Frequently asked questions

Everything you need to know about ppl.studio — from getting started to pricing and features.

Getting Started

ppl.studio is an AI-powered UGC studio. You create AI experts with consistent faces, upload your real products, and generate authentic-looking user-generated content photos in about 60 seconds. It replaces expensive creator shoots and slow production timelines.

You pick or create an AI expert (a virtual person with a stable face identity), upload your product photos to the props library, describe the scene you want, and ppl.studio generates a realistic photo of that expert using your product. The AI handles lighting, poses, and composition automatically.

Yes. You get 10 free photos when you sign up — no credit card required. Create an account, add an AI expert and your products, and generate your first UGC photos right away.

No. ppl.studio is built for marketers, brand owners, and e-commerce sellers — not designers. You describe what you want in plain language and the AI handles the rest. No Photoshop, no camera, no studio setup.

Physical consumer products work best — skincare, supplements, food and beverage, apparel, electronics, home goods, and beauty products. Anything you can photograph and upload to the props library can be placed into generated scenes.

AI Experts & Personas

AI experts are virtual people with stable, consistent face identities. Once you create an expert, you can use them across hundreds of different scenes and their face stays the same. Think of them as your on-demand UGC creators.

Yes. You can generate new experts with specific characteristics — age range, gender, ethnicity, style, and more. Each expert maintains full face consistency across every photo you generate with them.

Yes. Face consistency is the core feature of ppl.studio. Use the same expert across product shots, lifestyle scenes, and testimonial formats and they look like the same person every time.

Yes. Every image you generate is yours to use however you want — ads, social media, product listings, packaging, websites. No licensing fees and no usage restrictions.

Very realistic. ppl.studio uses advanced image generation models that produce natural lighting, authentic poses, and lifelike expressions. Most viewers cannot tell the difference between AI-generated UGC and real creator content.

Packs, Video & Automation

A UGC pack is a campaign-ready set of assets produced in one run rather than image-by-image. Instead of prompting for one photo at a time, you pick a pack, and ppl.studio generates the full set — typically 5–10 ad variants, a mix of lifestyle photos and short video, and every aspect ratio a channel needs (9:16, 1:1, 4:5) from the same AI UGC creator and the same product. ppl.studio ships 21 ready-made packs built for Shopify, Amazon, TikTok Shop, and Meta Ads, so the output is a launch-ready campaign, not a folder of loose images you still have to assemble.

Yes — photo and video come from one identity. You generate lifestyle photos with an AI UGC creator, then turn those photos into short video clips with Animate, which is image-to-video conditioned on the same persona. Because the video is built from the creator's existing photo rather than a fresh generation, the face stays identical across both formats — the cross-format consistency standalone video tools cannot match. Video supports lip-synced speech: paste a script and the creator delivers it with natural mouth movement, ready as a 9:16 clip for Reels, TikTok, and Shorts.

You upload your product to the Props Library once, and it is composited into every generated scene rather than re-imagined from a text description. That preserves the details buyers and marketplaces check — correct colors, exact label placement, and right proportions — across hundreds of photos and videos. Product accuracy is the difference between AI UGC you can put on an Amazon listing or a paid ad and a generic AI image that only approximates what you sell, and it is why the same product reads as the same product in every asset of a pack.

Yes — automation is the core of the workflow. Each of the 21 packs runs a full sequence (research, casting, product prep, generation, and export) so a whole campaign is produced from one brief rather than assembled manually. For programmatic use, ppl.studio exposes a REST API so packs can be triggered from your own systems — useful when a catalogue change should regenerate the affected assets automatically, or when seasonal launches need every SKU refreshed on the same deadline. The unit of production is a campaign, not a single image.

Structure tests at the angle-and-hook level and let a pack produce the variants. A typical first test is 3 angles × 3 hooks × 2 formats (photo + video), which a single pack generates in one run using the same AI UGC creator and product. Because AI UGC costs 50–200× less per asset than commissioning human creators, testing at the volume Meta's algorithm actually needs to exit the learning phase becomes economically viable — the constraint stops being production capacity and becomes which angles to try next.

Yes. The pack library splits into photo packs (lifestyle sets, catalog-at-scale, Amazon A+ and marketplace imagery) and video packs (short-form UGC video, talking-head clips, TikTok and Reels sequences), because that is how buyers actually search and plan campaigns. Many packs deliver both — for example, a Meta Ads pack ships a photo-plus-video mix in Meta specs — so you can run a photo-only workflow, a video-only workflow, or a combined campaign from one AI UGC creator.

An AI UGC creator is a synthetic person with a stable, reusable face identity that holds your real product across every asset — the recognizability of a dedicated creator without the per-shoot cost, scheduling, or licensing. Stock models change from photo to photo and never hold your specific product; influencers give you authenticity but at high cost, limited volume, and usage windows that expire. An AI UGC creator gives you one consistent face and one accurate product across hundreds of on-brand photos and videos, generated on demand.

Product Photos

Go to the Props section in your dashboard. Upload product images and the system automatically removes backgrounds and prepares them for use in scenes. You can organize products into collections for faster access.

ppl.studio accepts PNG, JPG, JPEG, and WebP files. For best results, use high-resolution images with clean backgrounds. The system handles background removal automatically.

Yes. You describe the scene in the prompt — location, lighting, mood, pose, camera angle. You can also use scene presets for common setups like kitchen counter, outdoor cafe, gym, or studio backdrop.

There is no hard limit on photo generation. With a paid plan, you can generate as many images as you need. Generation uses your own API key so costs scale with your usage.

Generated photos are high-resolution and suitable for ads, social media, and e-commerce listings. Output resolution depends on the model you select in workbench settings.

Features

Storyboards let you plan multi-frame content sequences — think ad scripts, carousel posts, or before-and-after flows. Set up the frames, assign experts and props, and generate the full sequence at once.

Animate turns your static AI photos into short video clips. The latest version supports lip-synced speech — paste a script and your AI expert speaks it with natural mouth movements.

The props library is where you store and manage your product images. Upload once and your products are available in every scene you generate. The system handles background removal, tagging, and multi-angle support.

Yes. You can export photos formatted for Amazon, Etsy, Shopify, TikTok Shop, Instagram, and more. Each platform gets the correct aspect ratio and file specifications.

Yes. Every plan includes the ppl.studio API: 144 REST endpoints under /api/v1 that cover everything the app does — creators, product photos, UGC videos, scripts and social posting — with one API key from Account → API keys. There is also an MCP server for AI assistants — add https://ppl.studio/mcp as a connector in claude.ai, ChatGPT, Claude Code or Cursor and sign in, or run it locally to save results to your disk — and a Claude skill that teaches the workflows. See ppl.studio/features/api and ppl.studio/features/mcp.

Pricing & Account

ppl.studio is $1.99 per week. Image generation runs on your own API key so generation costs stay under your control. There are no hidden fees or per-image charges from ppl.studio itself.

Yes. Cancel anytime from your account settings. There is no minimum term and no lock-in. Your access continues until the end of the current billing period.

Yes. Every image you generate is yours. Use them in ads, on your website, in product listings, on social media, or anywhere else. No licensing fees and no usage restrictions.

You get 10 free photos when you sign up — no credit card required. After that, the platform is $1.99 per week. There is no permanent free tier, but the trial lets you test everything before committing.

ppl.studio accepts all major credit and debit cards through Stripe. Payments are processed securely and you can manage your subscription from your account settings.

Models & Technology

Images are generated with Google's Nano Banana 2.1, on your own Gemini API key. Video is generated with Gemini Omni 1.2 Flash, Google's video model on the Gemini API (it replaced Veo 3.1 there in October 2026). We deliberately keep the model stack small rather than offering dozens of mediocre options.

Yes. The Creator plan supports BYO key, so generation runs on your own quota and billing relationship with Google. ppl.studio charges only the $1.99/week platform fee for everything we build on top — Experts, Props, presets, storyboards, gallery, and Animate.

Animate is image-to-video by design, conditioned on a photo of an existing AI Expert. That means your persona stays identical across photo and video assets, which is the consistency standalone video models can't match when used in isolation. For cinematic brand films, a dedicated text-to-video tool may produce more polish; for UGC short-form, Animate ships an ad-ready 9:16 clip with synced speech in under 2 minutes.

No. Every image and video you generate on a paid plan is delivered watermark-free and you own all rights. Free-tier output is also watermark-free.

Consistent personas build brand recall the same way a recognizable creator does, but without the per-campaign creator cost. Brand-lift studies consistently show 40–60% higher recall on creative sets featuring one persona vs. scattered faces, and the algorithm rewards that recognition with cheaper impressions.

Brand Safety, Disclosure & Legal

Disclosure rules depend on platform and region. Meta, TikTok, YouTube, LinkedIn, and Pinterest all support AI-content labels and increasingly enforce them for synthetic media of real-looking people. The FTC's 2024 endorsement guides require disclosure when a synthetic persona is used as an endorser. The EU AI Act adds Article 50 disclosure obligations for deepfakes and synthetic media starting in 2026. We recommend disclosing by default — flip the platform's AI toggle on upload — which preserves regulatory headroom and consumer trust without measurably hurting paid-social performance in our tests.

Yes. Every image and video you generate is yours to use in ads, on your website, in product listings, in email, packaging, or any other commercial context. No licensing fees, no royalties, and no per-impression usage restrictions. The output is watermark-free on every paid plan and on the free tier.

No. AI Experts are entirely synthetic — generated by image models rather than derived from photographs of identifiable individuals. This is the legal foundation that allows you to use them commercially without consent forms, model releases, or talent contracts. Faces produced by ppl.studio do not resemble any specific real person and we run filtering to catch accidental likeness collisions.

Yes — exports can carry C2PA-compatible provenance metadata so generated content includes a tamper-evident claim about what created it and when. Meta, TikTok, and YouTube now read C2PA claims to auto-populate their 'Made with AI' labels, which is the cleanest mechanism for disclosure compliance at scale.

We run automated face-similarity filtering on Expert generation to minimize accidental likeness to public figures or known individuals. If you spot a likeness issue in a generated image, email hello@ppl.studio and we will remove the relevant seed from the pool. Treat AI-generated faces the same way you'd treat a stock photo of a model — useful for general lifestyle context, not a substitute for a named endorsement.

E-commerce & Marketplace Listings

Yes — for Amazon main image and gallery images, Amazon's policy requires that the product itself be accurately represented, which ppl.studio's Props library composition handles. AI-generated lifestyle and A+ Content imagery is permitted as long as the product is accurate and any people depicted are clearly stylistic rather than implying a real endorsement. Many ppl.studio customers feed AI UGC into Amazon A+ Content, Brand Story, and Premium A+ to lift conversion vs studio-only imagery.

Yes. Shopify, Etsy, and TikTok Shop all permit AI-generated and AI-assisted product imagery as long as it accurately represents the product and complies with each platform's disclosure norms. Etsy specifically requires disclosure of AI use in the listing description. TikTok Shop encourages the AIGC toggle. Shopify has no specific rule beyond accurate representation — most stores use AI UGC freely without disclosure on the storefront.

Yes. Generate Amazon Main Image-spec (1:1, white background) for the listing slot, then generate lifestyle 4:5 or 9:16 for the secondary gallery, A+ Content modules, and Brand Story carousels. Our free Product Image Spec Checker (/tools/image-spec-checker) validates the file against each marketplace's exact rules before upload.

Across DTC and marketplace tests, AI UGC lifestyle photos lift PDP conversion 8–22% when placed in the second or third gallery slot — they provide the 'in-context' use case that pure-studio imagery misses. Traditional studio remains the safer Main Image; AI UGC supplements with the diversity (skin tones, scenes, demographics) that single shoots can't match cost-effectively.

Quick reference: Amazon Main Image — 1:1 square at 2000×2000+; Amazon A+ — module-specific (often 1464×600 or 600×450); Shopify product gallery — 1:1 or 4:5; TikTok Shop — 9:16 (3:4 secondary); Instagram Feed — 4:5 portrait; Reels and Stories — 9:16; Meta Feed — 4:5 or 1:1; Pinterest — 2:3. ppl.studio exports natively in 9:16, 1:1, and 4:5 from the same generation — no re-encoding, no quality loss.

AI Search & Visibility

Yes, if your content is structured for it. AI search engines (Perplexity, ChatGPT Search, Google AI Overviews, Amazon Rufus) prefer pages with explicit numeric statistics, named-entity comparisons (“X vs Y”), clear FAQ blocks with full-sentence answers, and last-updated dates. Pages built this way are cited 5–10× more often than long-form prose. ppl.studio’s glossary, comparison, and FAQ surfaces are all built to this pattern — the same approach you can apply to your own owned content.

GEO — sometimes also called AEO (Answer Engine Optimization) or AI Visibility — is the discipline of optimizing content to be cited inside AI-generated answers rather than ranked as a blue link. The high-impact patterns are: explicit numbered statistics, named-entity comparison pages, FAQ formatting with prose answers, structured product data, and last-updated dates. Domain authority still matters but less than in traditional SEO, which is why mid-size sites with structurally optimized content regularly outperform large brand sites in answer-engine citation share.

Not directly — AI search engines cite text and the structural patterns around it, not images. But AI UGC supports the AEO/GEO playbook indirectly: lifestyle photography on PDPs feeds Amazon Rufus and Google AI Mode the use-case context they need to answer “is this for me?” queries. The text on your page (specs, reviews, FAQ blocks) drives citation; the imagery drives the conversion that follows.

AI shopping assistants synthesize across listings, reviews, and third-party comparisons into a single conversational answer — meaning your PDP is no longer the only decision surface. Optimization requires clean product titles, prose-format bullets (not keyword strings), comprehensive use-case Q&A, accurate inventory and price feeds, and lifestyle imagery that answers “will this work for my situation?” AI UGC fills the lifestyle imagery layer at the volume Rufus needs across hundreds of SKUs.

They’re reshaping it, not eliminating it. Zero-click queries (where the user gets a synthesized answer without clicking) are growing fast, especially for comparison, definition, and recommendation queries. The defensible response is to optimize for citation share inside the AI answer (so your brand is mentioned even when no click happens) and to invest in mid- and bottom-funnel content where users still click through to evaluate or purchase. Pure top-funnel “what is X” content is the most exposed.

llms.txt is a proposed standard (introduced in 2024) for publishing a markdown-formatted, LLM-friendly map of your site at /llms.txt — similar to robots.txt or sitemap.xml but built for AI assistants to ingest. Perplexity, ChatGPT, and Claude honor it when present. If your site has 50+ canonical pages organized into topical clusters (glossary, guides, comparisons), publishing an llms.txt is one of the lowest-effort, highest-leverage GEO optimizations available. Early case studies show 20–40% citation lift inside Perplexity and ChatGPT for sites that implement it.

AI citation rate is the share of relevant LLM-generated answers in which your brand or page is cited as a source — the AI-search equivalent of organic CTR. Tools like Otterly.AI, Profound, Athena HQ, and Goodie sample queries representative of a target keyword set, run them through ChatGPT, Perplexity, Claude, Google AI Overviews, and Bing Copilot, and report citation counts and share over time. Brands in the top decile for citation rate in their category see attributable brand-search lift of 25–60% over 6 months.

Topical authority is the degree to which a site is recognized — by both search engines and LLM citation systems — as a deep, trustworthy source on a specific subject. You build it by publishing comprehensive coverage of a topic cluster: pillar pages, supporting articles, a glossary, comparison pages, FAQ blocks, and case studies, all internally linked. Topical authority compounds: each new page in the cluster lifts every other page's rankings and citations. The strategic answer to “one big article or ten focused ones” is almost always ten focused ones, internally linked.

Yes — strongly. E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) is the framework Google's quality raters use to evaluate content, which feeds back into ranking models. The same signals that earn human-rater trust earn LLM citation: named authors with credentials, cited primary sources, last-updated dates, original photography and data (not licensed stock), structured About and contact info. Pages with strong E-E-A-T signals are cited up to 60% more often than anonymous equivalents in answer-engine research.

Query fan-out is the technique generative search engines (notably Google AI Mode and ChatGPT search) use to expand a single user query into multiple parallel sub-queries, retrieve sources for each, and synthesize one answer. A query like “best AI UGC for Shopify under $100” fans out into “AI UGC tools 2026,” “AI UGC pricing comparison,” and “AI UGC for Shopify” — each retrieving its own candidate set. The implication: optimize only for the head query and you miss the citation opportunity. Brands that publish dedicated long-tail sibling pages per fan-out branch capture 2–4× more total AI citations than brands that publish only the pillar page.

Five schema types move the needle in 2026: FAQPage (every Q&A surface; 3–5× citation lift over unmarked content), Article (every blog and guide; supplies headline, author, publish/modified dates), HowTo (every step-by-step; structured step list maps to how engines compose how-to answers), Product + Offer + AggregateRating (every PDP; required for AI shopping citation), and Organization with sameAs links to verified social profiles (site-wide; supplies the entity signal AI engines use for brand disambiguation). Audited 2026 data shows under 30% of mid-market sites have FAQPage schema on pages with visible FAQs — closing that gap alone moves citation rate measurably within weeks.

Share of voice (AI) is the percentage of brand-relevant queries inside a defined topic set where your brand appears in the AI-generated answer — cited, mentioned by name, or recommended. SaaS tools like Otterly, Profound, Athena HQ, and Goodie run a curated query set (typically 20–100 queries per topic) across ChatGPT, Perplexity, Google AI Overviews, Microsoft Copilot, and Amazon Rufus on a weekly cadence, parse the cited sources and mentioned brands, and compute share of voice. The metric is asymmetric: top-3 cited sources in a topic capture 60–80% of all citations for that topic, so modest gains in share of voice translate into outsized traffic and brand-recall lifts.

Model context protocol (MCP) is the open standard introduced by Anthropic in late 2024 for connecting AI assistants to tools, data sources, and APIs in a uniform way — the USB-C equivalent for agentic AI integrations. By mid-2026, all major model providers and many enterprise tools support MCP. For marketers, the practical effect is that agentic creative-ops workflows (an AI agent that takes a brief, pulls inventory from Shopify, fetches brand assets, drafts ad copy, generates AI UGC visuals, uploads to Meta Ads Manager) are now economically feasible because each integration is build-once-use-anywhere instead of bespoke per tool. Brands not yet running MCP-connected workflows are on the trailing edge of a curve that’s steepening through 2026.

Treat citation count without a denominator as meaningless — count rises with engine activity even when share is falling. The mid-2026 thresholds worth planning around: 1–3% citation share is baseline (your brand surfaces occasionally but is not on the engine’s default shortlist); 5–8% is consistent presence; 10–15% is shortlist position where citation traffic shows up as a non-trivial channel in analytics and DTC brands report citation traffic converting at 2.3–3.6× the rate of paid social cold traffic; 20%+ is category-leader territory where the marginal lift on additional optimization investment compresses sharply and the better next investment is breadth (more query clusters) rather than depth (more pages inside a cluster).

Perplexity is usually the right first investment, even though it is smaller than ChatGPT Search or Google AI Mode in raw query volume. Its citation surface is fully numbered and inspectable (clean measurement), its query mix skews high-intent, and DTC brands report 3–8× higher CVR from Perplexity referrals than from Google organic. Optimize Perplexity first, then layer in Google AI Mode for top-of-funnel volume, Amazon Rufus if you sell on Amazon, ChatGPT Search as the broadest general-purpose surface, Microsoft Copilot for B2B pipelines, and Claude for technical or developer audiences. Optimizing all six from day one is a common mistake — measurement gets thin everywhere and nothing compounds.

AI Shopping Optimization (AISO) is the practice of structuring product content, reviews, schema, and brand entities so a product is preferentially surfaced inside AI shopping assistants — Amazon Rufus, Perplexity Shopping, ChatGPT Shopping, Google AI Mode shopping panels, and Microsoft Copilot. AISO sits between classic e-commerce SEO and broader generative engine optimization: the target artifact is the assistant’s 1–5 product recommendation (with click-through or assistant-driven add-to-cart), not a SERP position. High-impact AISO moves are clean Product/Offer/AggregateRating schema, review-corpus depth with explicit use-case language, comparison pages targeting category-defining queries, FAQ blocks answering literal buyer questions, and a persona-locked AI UGC photo library that fills the assistant’s multimodal-answer surface.

Perplexity already prefers content under 12 months old for most commercial queries; Google AI Mode is moving the same way. A practical refresh cadence: refresh every priority page (pillar, comparison, FAQ-heavy) at least every 9–12 months, and refresh any page that drops 25%+ in citation share week-over-week within two weeks of the drop. The refresh is not cosmetic — engines respond to updated statistics, refreshed publish/modified dates emitted in Article and FAQPage schema, new internal links, and (if applicable) a refreshed visual layer that fills the inline-image carousel. Set-and-forget content programs lose citation share materially faster than refresh-cadence programs through 2026.

A rationale snippet is the short justification fragment an AI engine surfaces alongside a citation — the ‘best for sensitive skin’, ‘reviewers report it lasts through a workout’, or ‘works well with the matching toner’ that the engine lifts from your page. Capturing rationale snippets alongside citations is the single highest-signal artifact for content rewrites, because it tells the writer precisely which passage the engine is leaning on. Competitor rationale snippets are the cleanest input to the weekly content backlog — they reveal the use cases, claims, and phrasings the engine has decided are good answers in your category. AI-visibility tools that surface rationale (Profound, Otterly, AthenaHQ) shorten the loop between observation and content action by 30–50% vs. tools that only count citations.

Pure last-click attribution misses 60–80% of AI-search-influenced revenue in 2026 because Amazon Rufus never leaves Amazon, Claude proxies the majority of citations, ChatGPT Search sends a partial referrer on roughly 55% of clicks, and Google AI Mode looks identical to Google organic at the referrer layer. The working 2026 model layers three signals: citation telemetry from an AI-visibility tool (Profound, Otterly, Peec.ai, AthenaHQ), referrer + UTM reconciliation engine-by-engine, and a post-conversion ‘where did you first hear about us?’ survey on the order-confirmation page with AI search broken out by engine. Reconciled together, the model lands inside ±15% of channel-revenue truth on engines where ground-truth checks exist. Report a defensible range, not a point estimate — point estimates lose credibility on the first cross-check.

Citation drift is the week-over-week delta in citation share on a single query, single engine — the early-warning indicator that a competitor has published into your space, that an engine has re-weighted its retrieval substrate, or that one of your pages has decayed below the freshness window the engine prefers. The standard trigger is 25%+ negative drift over two consecutive weeks on a priority query — flag it immediately for content refresh; brands that act on drift alerts inside 14 days recover original citation share at ~3× the rate of brands that respond inside 60 days. A 30%+ positive drift on a non-target query is the surge signal: scale the page’s depth before competitors notice.

Roughly 22% of mid-2026 brand-query citation misses are root-caused to entity disambiguation failure — the engine retrieved the right content but routed the citation to a competing entity with the same surface name. Brands with a complete Organization + sameAs schema block linking to Wikipedia, Wikidata, Crunchbase, and core social profiles see 18–30% higher citation share than equivalent brands missing the sameAs links. Wikipedia presence remains the single highest-correlated entity-graph signal with citation share inside ChatGPT Search and Perplexity (r = 0.62 across a 200-brand sample, mid-2026). Run a brand entity graph audit before assuming page-level rewrites are the bottleneck — for many brands, the entity layer is the actual constraint and the rewrite is wasted effort.

The multimodal-answer surface is the inline image carousel AI engines now surface alongside the text answer on commercial queries — Perplexity on ~35% of commercial queries, Google AI Mode on ~55%, ChatGPT Search on ~25%. The engines route 20–35% of total citation weight through this surface on commercial queries, which means brands that ignore the inline-image carousel cap their citation-share ceiling well below the optimized-competitor benchmark. Pages with a fresh persona-locked AI UGC photo set earn cited image slots in roughly 3.2× the multimodal answers vs. equivalent text-only pages. The visual layer has moved from supplementary to a first-class citation channel with its own retrieval substrate and freshness window (4–12 weeks for fast-moving categories — materially shorter than the text freshness window).

Assistant handoff rate is the percentage of AI-assistant recommendations on a defined query set that result in a click-through to the brand’s own surface — the PDP, the homepage, the comparison page — vs. completing inside the assistant. It is the agentic-shopping analog of organic CTR. Average mid-2026 handoff rates: ~62% on Perplexity, ~48% on ChatGPT Search, ~35% on Google AI Mode, ~12% on Amazon Rufus. Handoff rate is projected to fall 20–35% across the major assistants through 2027 as in-flow checkout flows ship from OpenAI, Perplexity, and Amazon. Brands tracking handoff rate per engine now identify AISO investments that compound assistant-side revenue 6–9 months ahead of brands that only track aggregate citation share — early enough to compete for shortlist position before crystallization.

A rationale snippet audit is the recurring (typically weekly) process of capturing the rationale snippets AI engines publish alongside their citations, organizing them into a tabular artifact, scoring them on a five-axis rubric (use-case specificity, claim density, visual reference, competitive framing, freshness reference), and converting the output into a Friday content backlog the writers can ship from. The audit is the bridge between raw AI-visibility data and actionable content strategy: citation count tells you whether the engine picked your page, but rationale tells you why — and that ‘why’ is what a content brief actually needs. Run the audit weekly on a stable priority query set, group by source URL for briefing, and cap the backlog at ten briefs per week (most teams cannot ship more than ten page-level rewrites per week). Brands that run a rationale audit from week one ship content rewrites 40% faster than brands that only count citations.

Yes if your site has 50+ canonical pages organized into topical clusters (glossary, guides, comparisons, FAQ); otherwise fix on-page schema and brand entity disambiguation first. llms.txt is a proposed standard (introduced by Jeremy Howard in September 2024) that exposes a curated, markdown-formatted map of a site at /llms.txt — modeled on robots.txt and sitemap.xml. Perplexity, ChatGPT, and Claude all honor it; sites that ship a maintained llms.txt see 20–40% citation lift inside Perplexity and ChatGPT for their topical areas. The single most important implementation rule is to generate the file at build time from the same data the site is built from — hand-maintained files decay inside two quarters. Add llms-full.txt (full markdown content of the top 30–80 pages) only after the link-map version has stabilized and only if site depth justifies the maintenance overhead.

In-flow checkout — the AI-assistant feature that lets a user complete a purchase inside the assistant’s own surface, with no click-through to the brand’s site — is shipping or has shipped across OpenAI’s ChatGPT Shopping, Perplexity Shopping, and Amazon Rufus Buy through Q4 2026. The structural effect is that a meaningful slice of revenue lands without ever touching the brand’s analytics stack, which means assistant handoff rate falls structurally and the standard three-signal attribution model (citation telemetry + referrer reconciliation + post-conversion survey) needs a fourth signal: the engine-side merchant report the assistants publish for checkout merchants. Most brands will not light up the fourth signal until Q1 2027; brands that wire it now buy a 6–9 month planning advantage. The operational fix is to model assistant-side revenue as a separate P&L line from brand-site revenue.

Shortlist position is the slot a brand occupies on the 1–5 recommendation list an AI shopping assistant surfaces in response to a category-defining query. It matters more than aggregate citation share for commercial outcomes because the assistant absorbs the user’s alternatives evaluation — the buyer often never sees the brands the assistant did not surface. Position 1 captures 38–52% of assistant-driven traffic on average across the major engines in mid-2026; positions 4 and 5 capture under 6% combined. Brands that hold position 1 or 2 on the top 5 category-defining queries earn the bulk of agentic-shopping revenue even when their long-tail citation share trails optimized competitors. The shortlist crystallizes 9–18 months after an assistant launches in a category; brands that hold position by month 12 are difficult to displace without a category-shifting product or content event. Optimize for position on the top queries first, breadth across long-tail second.

Yes. A signed AI policy is the brand-controlled declaration — published at /robots.txt, /ai.txt, or via the in-development IETF AI Preferences Working Group standard — that names which AI crawlers and assistants are permitted to ingest, index, and cite the site’s content. By mid-2026, only ~35% of mid-market sites have a clear, single-source AI-bot allow/deny posture; the rest have either no policy or a contradictory policy across robots and meta tags, and citation share suffers because engines route around ambiguous targets. Sites with a clean policy see 12–18% higher citation share inside ChatGPT Search and Claude than sites with contradictory policy, controlling for content quality. The right minimum: explicitly allow GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot, anthropic-ai, and Applebot-Extended (or explicitly deny, but pick one); pair the policy with a maintained llms.txt so engines know both what they can ingest and how to retrieve it cleanly.

A citation footprint is the full inventory, for a single brand, of every URL the major AI engines cite when that brand is the cited source — paired with the rationale snippet the engine attached, the query that triggered it, the engine, and the week it surfaced on. Mapped against two or three of your highest-shortlist-position competitors, the footprint tells you what to publish next in the language of the brands the engines have already shortlisted in your category. Mid-2026 cohort programs that map three competitor footprints weekly ship 30–50% more high-leverage briefs per quarter than programs that only score their own citations because the gap analysis is a direct read from the data rather than an inferred strategy artifact. Map weekly, pivot by URL template tag (comparison, use-case, FAQ, glossary, PDP), and cap at three competitors — past five, the writer cannot read the map fast enough on a Monday to act on it the same week.

Re-use the priority query set from your visibility dashboard, capture cited URLs across the two or three highest-shortlist competitors weekly using a six-column table (query, competitor + week, cited URL, URL template tag, rationale snippet, engine + position), pivot by URL template to surface missing template coverage, pivot by query on a 0–3 axis to surface share-of-voice gaps, cluster rationale snippets by claim type to surface the language the engine has decided is citable in your category, then draft briefs one-per-(template, query-cluster) capped at ten per week with the verbatim competitor rationale quoted into each brief. Pair every text brief with a visual brief — pages cited in multimodal answers won't close from prose alone. Expect first measurable citation lift on a gap-targeted query at week 9–11; three to five weeks after publish, matching the engines' retrieval-refresh cycles.

URL template coverage is the count of distinct cited URLs a brand holds per content template — comparison pages, use-case pages, category pillars, glossary, FAQ, PDPs, case studies, reviews, integration pages, location/service pages. It's the most actionable axis a citation footprint carries because content teams ship in templates, not in one-offs. The mid-2026 cohort benchmark for a category-leading brand is 10–25 cited URLs per primary template across the four highest-volume AI engines (Perplexity, ChatGPT Search, Google AI Mode, Microsoft Copilot). Comparison-page coverage is the single template most correlated with category-defining-query shortlist position (r = 0.58 across a 120-brand sample in mid-2026); brands with zero coverage on the comparison template lose roughly 38% of available category-defining-query citation share to competitors that ship the template.

Modern AI engines retrieve and cite at the passage layer — the citation chip resolves to a #:~:text= URL fragment that scrolls the reader to the cited paragraph rather than the page header. The implication is that pages are no longer the unit of content optimization, paragraphs are. On pages with five or more weekly citations, a single paragraph typically carries 60–85% of total citation share in roughly 70% of cases. Rewriting that paragraph lifts the page's aggregate citation share 15–35% within two refresh cycles in 60% of cases. Track citation share at the URL-fragment layer, not the URL layer, for any page above the five-citation-per-week threshold — the fragment data tells the writer which passage to defend and which to rewrite.

Google AI Mode and ChatGPT Search fan out roughly 78% of commercial queries into 2–6 sub-queries before synthesizing the answer — a query like 'best AI UGC for Shopify under $100' becomes 'AI UGC tools 2026', 'AI UGC pricing comparison', and 'AI UGC for Shopify' run in parallel, with sources retrieved per sub-query. The practical implication is that optimizing only for the head query misses the citation opportunity entirely. Brands publishing dedicated long-tail sibling pages per fan-out branch capture 2–4× more total AI citations than brands publishing only the pillar page. Multimodal-answer surfaces appear on roughly 40% of fan-out branches vs 30% of head queries, so sibling pages with a persona-locked visual layer compound the citation opportunity rather than dilute it.

Multimodal answer share is the percentage of priority queries on which your brand holds at least one cited slot inside the AI engine's inline image carousel, weighted by slot position (position 1 = 1.0, position 2 = 0.6, position 3 = 0.3, positions 4–5 = 0.1). It is the visual analog of share of voice and the headline metric for the multimodal-retrieval pipeline. The two surfaces respond to different signals — ImageObject schema density, image freshness, and persona stability drive the carousel, while prose rationale density and entity disambiguation drive text citations — so a brand can be winning one and losing the other. Brands tracking multimodal answer share quarter over quarter identify carousel decay 6–10 weeks ahead of brands tracking only aggregate citation share — early enough to refresh the visual library before slots drop.

The visual freshness window runs 4–8 weeks on fast-moving categories (apparel, beauty, food, supplements, accessories), 8–12 weeks on mid-velocity categories (home, fitness, pet, baby), and 12–24 weeks on slow-moving categories (B2B SaaS, financial services). Across the board, the visual freshness window is materially shorter than the 6–18 month text freshness window. Engines drop carousel slots for stale images an average of 5–7 weeks before they drop text citations from the same page. The implication is operational: quarterly photo-shoot cadence is an order of magnitude slower than the engines retrieve at, which is why every category-leading brand on the multimodal surface in mid-2026 ships visual refresh on a weekly or bi-weekly loop.

ImageObject inside Product is the single highest-leverage piece of structured data the multimodal-retrieval pipeline reads — pages that emit ImageObject with contentUrl, caption, name, description, and creator earn cited image slots in roughly 2.8× the multimodal answers vs. equivalent pages with raw img tags. Article schema with image populated as ImageObject lifts blog and guide carousel slots by 1.9×. Author/creator on the ImageObject feeds visual entity disambiguation, the bridge to the textual brand-entity graph. caption that mirrors the cited surrounding paragraph (not a generic alt-text duplicate) lifts carousel inclusion by 1.6×. Roughly 82% of mid-market e-commerce sites do not emit a full ImageObject on PDPs by mid-2026 — closing that gap alone moves multimodal answer share measurably within one or two refresh cycles.

Yes — by 3–5× on the inline image carousel. The multimodal-retrieval pipeline reads persona stability across a page set as a visual entity-disambiguation signal, the same axis the textual pipeline reads off sameAs schema and Wikipedia presence. Pages with a single recognizable persona across the priority page set out-cite pages with stock or unrelated photography because the engine can confirm a stable visual identity for the brand. The operational implication is that ad-hoc model selection across a category page set destroys the disambiguation signal; single-persona discipline compounds it. Brands shipping a single persona on the priority set first and expanding to two or three personas in month four outperform brands shipping three personas from day one on cumulative multimodal answer share by 35–55% over six months.

The two pipelines (text and multimodal) read partially different signals — a page can win the text citation and miss the carousel for one of five repeating reasons. Off-product hero imagery (the engine cannot confirm a product match). Stale image last-modified header (the engine reads the file as 18+ months old). Missing ImageObject schema (the multimodal pipeline reads the URL but not the image-side metadata signals). Persona inconsistency across the page set (no stable visual entity for the brand). Off-topic OG image (the fallback retrieval surface presents a logo rather than a product shot). The fix for each is mechanical: re-shoot the hero, re-export and re-upload the image, upgrade to full ImageObject schema, lock a single persona across the priority set, and replace the og:image with a product-in-context shot from the same persona-locked visual set.

Passage level, by mid-2026. Every major engine through 2026 segments each crawled page into 6–18 chunks of ~600–900 characters, embeds each chunk independently, and retrieves the single best-matching chunk per fan-out sub-query rather than the page as a whole. Roughly 84% of mid-2026 citations resolve to one specific chunk inside a longer page — the page is the host, the chunk is the unit. A page can win the URL-level click and still lose the passage-level retrieval if the chunks inside it do not segment cleanly into ~600–900 character heading-bounded passages. The implication is that page-level word counts and topical authority are no longer the right optimization targets — chunk-level retrievability is.

600–900 characters is the sweet spot across Perplexity (~700-character chunks), Google AI Mode (600–900), ChatGPT Search (~900), and Microsoft Copilot (matches Google AI Mode). Below 400 characters the substrate treats the chunk as a low-context fragment and down-weights it; above 1,000 characters the substrate forces a mid-paragraph split and both halves lose their context anchor. The cleanest mechanical fix is heading cadence — h2 or h3 every ~600–900 characters lands every chunk on a heading-bounded passage, which is the optimal segmentation shape. Pages with one h2 followed by a 3,000-character wall of prose run 3–4 oversized sections the substrate splits arbitrarily; pages with consistent h2/h3 discipline run 8–14 retrievable chunks on the same word count.

Read the text-fragment anchor on the citation URL. Google AI Mode, Microsoft Copilot, and roughly 80% of Perplexity citations append a #:~:text= URL fragment that highlights the exact passage that was retrieved. Visit the citation URL with the fragment intact and the browser scrolls to and highlights the cited chunk. The chunk you see is the chunk the substrate retrieved; everything else on the page was indexed but did not win the retrieval for that query. This is the cleanest signal for a chunk audit table's cited-which-chunk column, and it also surfaces the chunk-pattern clusters competitors are winning citations on. ChatGPT Search rarely uses the text-fragment shape and requires manual inspection of the cited page to identify the matching passage.

A self-anchoring opening sentence restates the page-level context the chunk operates inside — without it, the retrieved chunk arrives at the engine as a context-less fragment that loses the entity match. 'The mid-2026 inline image carousel surfaces on roughly 35% of commercial queries on Perplexity' is self-anchoring; 'It surfaces on roughly 35% of those' is not. Self-anchoring chunks retrieve at 1.5–2.0× the rate of context-stripped chunks at the same chunk size. The structural reason: the substrate's chunk overlap window is only ~80–100 characters, small enough that a sentence starting with a load-bearing pronoun loses the antecedent during retrieval. Rewriting chunk openings to name the brand or product entity explicitly is the single highest-leverage editorial change in a chunk audit — most programs find roughly 28% of priority-page chunks need the rewrite on first audit.

Rewrite the existing priority pages first. Adding new URLs while existing priority pages have under-cited chunks doubles the substrate's crawl load without lifting citation share — the new URLs accumulate the same chunk-shape failures the existing pages have, and the program runs in place. Most programs that publish new URLs first report 4–8 weeks of crawl-load growth with no citation lift, then return to rewriting the existing pages anyway. The right sequencing is: chunk-audit the existing priority set, rewrite the worst 20–40 chunks against the five retrievability properties, capture the resulting citation lift, then expand URL-level coverage once the chunk discipline is baked into the editorial pattern. The rewrite-first cohort lifts citation share 2–3 weeks ahead of the new-URLs-first cohort on the same content budget, and the editorial discipline carries forward into every new URL the program ships.

Per-engine planning anchors for mid-2026: Perplexity runs 9–12 months on commercial text and 12–18 on informational, with images at 4–8 weeks. Google AI Mode runs 6–9 months on commercial text and 9–14 on informational, with images at 5–10 weeks. ChatGPT Search runs 10–14 months across query types with images at 6–12 weeks. Microsoft Copilot runs 8–11 months with images at 5–9 weeks. Amazon Rufus runs an asymmetric window where review-corpus freshness (12–24 weeks) and PDP image freshness (4–8 weeks) carry far more weight than copy freshness. Claude runs 11–15 months with the smallest multimodal surface. Treat these as planning anchors rather than precision numbers — engines ship substrate updates every 8–14 weeks that compress the windows modestly, and category velocity (apparel and beauty cycle faster than B2B SaaS) shifts them further. The right operational posture is to plan refresh cadence 20–25% inside the engine's nominal window to absorb the natural variance.

Five signals reconcile into the substrate's freshness timestamp on every priority page. (1) HTTP Last-Modified header — emitted by the CMS automatically on every edit; static-site generators that cap it at the deploy date compress the per-page signal materially. (2) Schema dateModified on Article, FAQPage, Product, and HowTo — emitted on every refresh from the same source-of-truth field the visible date renders from. (3) Visible last-updated date rendered in server-side HTML at the top of the page where the substrate can read it without JavaScript execution. (4) Content-diff hash the substrate computes between successive crawls — cosmetic date updates without real prose changes are discounted within two crawls. (5) Image last-modified and content hash for pages on the multimodal surface, which runs on its own track from the text signals. Refreshes that touch one or two signals and leave the rest stale are discounted; the discipline is to drive all five from one source-of-truth date field in a single edit window with real content changes that survive the diff.

Citation share on an un-refreshed priority page does not fall off a cliff at the window edge — it rides a measurable three-phase decay curve. Latency phase (months 0–6): share holds steady or trends mildly up as inbound links accumulate and the rationale snippet stabilizes. Compression phase (months 6–12): share compresses 15–35% relative to peak as the substrate weights age into the retrieval ranking and the page loses long-tail share first while still holding its strongest rationale clusters. Drift phase (months 12+): share drifts toward zero on commercial queries at 8–12% per month, and recovery requires a full content refresh rather than a cosmetic date update. The curve is observable per page when citation share is captured weekly — programs that score weekly detect the compression phase 6–10 weeks ahead of the drift phase, which is enough time to ship a refresh sprint before the page exits the active citation pool.

Four tiers, with quarterly rotation. Tier 1 (5–15 pages, refresh every 4–6 weeks): highest-traffic, highest-citation, category-defining pages — typically pillar posts, head-keyword comparisons, and top-cited PDPs. The cadence sits well inside the shortest engine freshness window so these pages never leave the active citation pool. Tier 2 (30–60 pages, refresh every 8–12 weeks): strong mid-citation pages with stable rationale snippets and growing trajectory — this is where most well-engineered programs spend the bulk of refresh hours. Tier 3 (the long tail, refresh every 6–9 months): lower-cited pages with steady performance held at the outer edge of the engine freshness window. Tier 4 (deprecate or merge): pages with near-zero citation that have not responded to refresh — 301 into a better-cited sibling or merge into a pillar. Rotate the tier assignment quarterly so the calendar does not ossify around last quarter's priorities — Tier 2 performers promote into Tier 1; Tier 1 underperformers demote.

Because the multimodal substrate scores images on a freshness window of 4–12 weeks — materially shorter than the 6–18 month text window — and the carousel slot decays 5–7 weeks ahead of the text citation on freshness-driven drift. Pages on the multimodal surface need an image-refresh queue that runs on its own track, independent of the text refresh sprint, or the carousel slot is lost while the text refresh is still inside its window. The image refresh has to ship a fresh content hash to clear the substrate's discount — re-uploading the same image file produces no hash change and is discounted. The right operational shape is re-export the persona-locked image from a fresh generation, update the ImageObject schema's contentUrl, and re-deploy on the 4–12 week image-window cadence. Persona-locked AI UGC at ppl.studio's production cadence is the operational fit for the image side of the freshness calendar — same persona, fresh file content, every cycle.

The freshness window is the time horizon inside which the engine treats a source as actively citable (9–18 months on text). The engine refresh cycle is the cadence on which the substrate re-crawls and re-embeds an already-indexed source against the latest fan-out queries (11–22 days on priority URLs across the major engines). Two distinct concepts that compose: the freshness window sets when a page exits the active citation pool; the refresh cycle sets how fast a refresh edit registers as a freshness lift. A refresh that lands the day after a re-embedding pass waits a full cycle (11–22 days) for the freshness signal to register inside the substrate's ranking — a refresh aligned to the next re-embedding pass lands in the next crawl and earns a 1–2 week head start on competitor recovery. Refresh-cadence engineering reads both the window and the cycle to time edits for maximum substrate response.

Per-engine planning anchors for mid-2026 commercial queries: Google AI Mode fans 82% of commercial queries into 3–6 sub-queries (average 4.1). ChatGPT Search fans 74% into 2–5 (average 3.4). Perplexity fans 68% into 2–4 (average 2.8) but surfaces more cited URLs per branch. Microsoft Copilot inherits a Google-AI-Mode-shaped fan-out but weights freshness more in synthesis (fresher source on the same retrieval rank carries roughly 1.4× synthesis weight). Amazon Rufus runs an asymmetric tree with one product-discovery branch plus 1–3 use-case branches against the review corpus. Claude has the shallowest fan-out (61% of commercial queries into 2–4 sub-queries) but the highest per-citation synthesis weight. Treat these as planning anchors — fan-out depth shifts with query intent (commercial fans wider than informational), category velocity (apparel fans wider than B2B SaaS), and substrate updates (engines often compress fan-out depth briefly post-update before re-expanding).

Sibling page architecture ships one focused page per dominant mid-layer fan-out branch — the pillar still anchors the cluster, but every branch the engine fans into has its own dedicated brand-aligned candidate in the per-branch retrieval set. Pillar-only architectures retrieve against the head branch alone and forfeit citation weight on every other branch the engine fans into. Audited mid-2026 cohort: brands shipping siblings across at least 60% of dominant mid-layer branches capture 2.4–3.8× more total AI citations than pillar-only brands on the same head query set, and the gap widens on multi-branch fan-outs where the head branch contributes less than 30% of total citation weight. Sibling architecture also defends against competitor pillar launches — brands on the sibling pattern lose only the head-branch citation when a competitor ships a strong new pillar; pillar-only brands lose the entire head-query citation.

The tree is inferred from the citation surface, not published. Five-step inference: (1) Capture cited URLs and rationale snippets per head query per engine on a weekly cadence — the same capture pipeline the rationale audit reads off. (2) Cluster rationale snippets per head query by intent slice — specification (constraints like 'under $100'), use-case (scenarios like 'for Shopify'), comparison (named-entity comparisons like 'Tool A vs Tool B'). Each cluster surfaces a dominant mid-layer fan-out branch. (3) Score brand coverage per inferred branch — yes (dedicated sibling targets the branch), partial (page mentions but doesn't target the branch), no (no page on the branch at all). (4) Compute the coverage ratio — dedicated sibling pages divided by inferred mid-layer branches. (5) Convert gaps into the editorial backlog sorted by inferred citation weight. Mid-2026 cohort medians sit at 0.35 on mid-market programs and 0.62 on category-leading programs; ratios above 0.7 are category-leading.

Three rules compose into one prioritization score per branch. (1) Highest-rank branch first — the top branch contributes 24–38% of total answer citation weight, so the highest-impact gap closes first even if lower-rank gaps are easier writes. (2) Multimodal-active branches before text-only branches — pages on carousel-active branches with persona-locked visual sets capture both the text citation and the carousel slot, doubling the per-page citation contribution. (3) Branches with weak competitor coverage first — branches where the top citation is held by a low-authority source (a generic listicle, a thin comparison page) are structurally easier to displace than branches dominated by a category-defining competitor. Compose: rank weight × multimodal multiplier × inverse displacement difficulty, sort descending. Cap the weekly backlog at 4–6 new sibling pages — past the cap, chunk-level quality drops below the citation threshold and siblings dilute rather than compound the cluster.

The fan-out coverage ratio is the headline metric of a sibling page architecture program: the count of dedicated sibling pages divided by the count of inferred mid-layer branches in the fan-out tree. Compute per head query and average across the priority set. A ratio above 0.7 is category-leading; 0.5–0.7 is competitive; below 0.5 is exposed. Mid-2026 cohort medians: 0.35 on mid-market programs, 0.62 on category-leading programs. The ratio is the single best predictor of total citation share lift per editorial hour spent — it outperforms URL count, word count, and backlink count as a lift predictor. Coverage ratio and total citation share move together with a 6–10 week lag — programs that lift the ratio in a sprint see citation share lift land at week 6–10 post-publish, matching the engines' re-embedding cycles. Track the ratio quarter over quarter alongside total citation share and reweight investment toward the head queries with the largest open gap.

Every major AI search engine through 2026 runs a three-stage retrieval-rerank-synthesis pipeline. Retrieval returns 40–120 candidate chunks per sub-query via embedding similarity; a cross-encoder reranker scores each (sub-query, chunk) pair jointly and prunes the candidate set to the top 3–8; synthesis composes the answer from only the reranked set. A page that retrieves into the candidate set but fails rerank never reaches the cited surface — which is why retrieval-only optimization caps citation share at the retrieval ceiling. Roughly 41% of mid-2026 citation share losses on previously cited pages trace to rerank failure, not retrieval failure. Programs scoring retrieval alone read 70% coverage and assume health while rerank survival sits at 12%; the gap between retrieval and synthesis is the rerank-survival opportunity and the highest-leverage chunk-level investment most programs still skip.

Per-engine planning anchors: Google AI Mode retrieves 80–120 candidate chunks per sub-query and reranks to 5–8 (6–10% survival). ChatGPT Search retrieves 60–90 and reranks to 4–7 (6–11% survival). Perplexity retrieves 40–80 and reranks to 5–8 (8–15% survival, the highest of the major engines, driven by a tighter candidate set up front). Microsoft Copilot inherits a Google-AI-Mode-shaped retrieval but runs a freshness-tilted rerank — the same chunk reranks 1.6× higher at three months old than at fifteen. Amazon Rufus runs asymmetric pipelines where review-corpus chunks survive at 12–20% on use-case branches vs PDP chunks at 4–7%. Claude has the smallest candidate set (30–60) and the highest selectivity (4–8% survival) but weights passage-level entity grounding the most heavily. Treat as planning anchors — rerank thresholds shift with substrate updates, query intent, and category velocity.

Five properties compose multiplicatively into the rerank survival lift. (1) Claim specificity — chunks with explicit numeric statistics survive rerank at 1.8–2.3× the rate of vaguely phrased equivalents because the cross-encoder reads numeric density as a citability proxy the embedding stage discounts. (2) Named-entity grounding — chunks that name the brand, product, or category entity explicitly survive at 1.6–2.0× the rate of context-anchored chunks. (3) Rationale-shaped opening sentence — chunks that open with a citable claim (rather than a topical introduction) survive at 1.7–2.1× the rate of equivalent chunks because cross-encoders read leading-sentence weight more heavily than tail-sentence weight. (4) Freshness stack alignment — the reranker discounts chunks whose freshness stack is stale even at equal embedding scores; the discount is steepest on freshness-tilted engines (Copilot, Perplexity). (5) Schema scaffolding — chunks inside fully-scaffolded pages (FAQPage, HowTo, Article, ImageObject) survive at 1.4–1.7× the rate of chunks inside schemaless pages. Composed multiplicatively, these can lift rerank survival 3–5× over baseline with 15–30 minute per-chunk edits.

Five-step audit: (1) Capture the synthesis surface (cited URLs and rationale-fragment anchors) per priority sub-query per engine weekly — same capture pipeline as the rationale audit. (2) Infer the candidate set per sub-query from competitor candidate-set membership (chunks the engine rotates through over a 4–8 week window even when not cited daily) and chunk-pattern competitor analysis (chunks with the same shape, entity density, and claim specificity as the cited set are very likely candidates). Inference precision lands within ±30% of the true set — enough to drive editorial priority. (3) Score rerank survival rate per brand chunk (cited slots / candidate-set slots) on a rolling 4-week window. (4) Score each chunk on the five rerank properties to surface the failing axes. (5) Convert the gap into the rerank-edit backlog sorted by retrieval frequency × failed property count, capped at 10–15 chunk-level edits per week. Mid-2026 cohort medians: 12% survival on mid-market programs, 22% on category-leading programs.

Phase 1 — Freshness drift: chunk retrieves but the freshness stack has aged into the discount window; rerank survival compresses 25–40% over 8–12 weeks. Fix is the five-signal freshness refresh (HTTP Last-Modified, schema dateModified, visible last-updated date, content-diff hash, image last-modified) driven from a single source-of-truth edit window. Recovers 60–80% of lost survival. Phase 2 — Claim erosion: chunk retrieves with current freshness, but competitor chunks have sharpened claim specificity on the same sub-query; survival compresses another 20–35% as the cross-encoder picks the more specific claim. Fix is a chunk-level leading-sentence rewrite with the verbatim rationale-snippet pattern from the competing synthesis surface. Recovers 70–90% of additional loss. Phase 3 — Entity dilution: chunk retrieves on shape but the entity slot has shifted in the engine's entity graph; survival drops to near zero. Fix is an entity-graph rebuild (sameAs schema, Wikipedia entry, structured author bio, explicit named-entity grounding) — 6–10 weeks to re-anchor. Scoping the wrong fix for the phase produces no lift and burns editorial bandwidth.

The text rerank pipeline runs in parallel with a multimodal rerank pipeline on the inline image carousel. The visual reranker reads its own candidate set (image chunks indexed by the multimodal substrate), runs its own cross-encoder pass scoring image-side properties — ImageObject schema density, persona stability across the page set, image freshness (4–12 week window, materially shorter than the text freshness window), caption alignment with the surrounding cited paragraph, and content-hash recency — and surfaces 3–6 top images per sub-query into the carousel slot. A page can survive the text rerank but lose the visual rerank, holding the text citation while losing the carousel and halving the per-page citation contribution. The fix is a parallel visual rerank-survival audit alongside the text audit on every multimodal-active sub-query, with both edit tracks briefed together (text-side leading-sentence rewrite plus image-side ImageObject upgrade and content-hash refresh).

AI Search Synthesis & Multi-Turn Citation

The synthesis stage is the third and final stage in the retrieval-rerank-synthesis pipeline every major AI search engine ran through 2026. Retrieval returns 40–120 candidate chunks per sub-query; rerank prunes to 3–8; synthesis composes the rendered answer from the union of surviving chunks. The synthesis stage runs the citation-vs-paraphrase decision per chunk — render verbatim with a quoted span, paraphrase into the engine's voice without a quoted span, or drop the chunk and relegate the source to the further-sources panel below the answer. Roughly 38% of chunks that survive rerank still fail synthesis in mid-2026 — read into the synthesis prompt but neither cited nor paraphrased. Programs that score rerank survival alone read 22% survival, assume the program is healthy, and miss that inline verbatim citation rate sits at 26% on rerank-surviving chunks.

Per-engine planning anchors for the share of rerank-surviving chunks that land as verbatim citations rather than paraphrase or further-sources: Perplexity 32–44% (highest of the general-purpose engines, surfaces quoted spans directly inside numbered source chips); ChatGPT Search 26–36%; Microsoft Copilot 22–32% (lifted on fresh chunks because the synthesis prompt is freshness-tilted); Google AI Mode 18–26% (lowest of the general-purpose engines, prefers paraphrase on short factual claims); Amazon Rufus 38–48% on the discovery branch from review snippets and 12–22% on the use-case branch from PDP body; Claude 16–24% but with the highest per-citation brand-recall lift because the answer prose is sparser. Mid-2026 cohort medians across rerank-surviving chunks: 26% verbatim rate on mid-market programs, 42% on category-leading programs.

The further-sources panel is the secondary source list AI search engines render below (or beside) the synthesized answer — chunks that survived retrieval, survived rerank, were read into the synthesis prompt, but were not rendered into the answer's prose with a quoted span or paraphrased sentence. Click-through on further-sources chips runs 8–15% of an inline verbatim citation's CTR at equivalent answer-position. 32–48% of rerank-surviving chunks land here in mid-2026 — the largest single-surface improvement opportunity for most programs. Roughly 60% of further-sources chunks recover into inline citation with a single leading-sentence rewrite into the citable-claim shape ([Entity] [verb] [quantified claim] [optional qualifier]). The recovery is the highest-leverage chunk-level edit a synthesis-stage audit identifies.

Five chunk-level properties compose multiplicatively into the verbatim citation lift. Claim specificity (chunks with explicit numeric statistics earn verbatim at 2.4× the rate of vaguely phrased equivalents); self-containment (chunks whose load-bearing claim does not require surrounding paragraph context survive synthesis at 1.6–2.0×); rationale-shaped opening sentence (chunks opening with a citable claim rather than a topical introduction survive verbatim at 1.7–2.1×); source authority signaling (Article schema with author name and credentials lifts verbatim by 1.3×); lexical distinctiveness (chunks with phrasing the engine cannot easily compress into its own voice survive verbatim at 1.5×). Composed multiplicatively, these lift verbatim citation rate 1.6–2.2× over rerank-survival-optimized baselines without adding any new pages — which is why the synthesis-stage audit is the highest-leverage chunk-level investment a program can ship after the rerank-survival audit lands.

Citation slot weight is the per-position multiplier the synthesis stage applies to surviving reranked chunks when composing the rendered answer. Mid-2026 anchors across the major engines: slot 1 carries roughly 1.0 weight (routes the answer's opening sentence on 60–80% of rendered answers); slot 2 carries 0.55–0.7; slot 3 carries 0.3–0.45; slots 4–8 carry 0.1–0.2 combined. The slot weight composes with the verbatim-vs-paraphrase disposition: a slot-1 verbatim citation drives 4–6× the click-through of a slot-3 verbatim citation at equivalent rerank-survival score because the user reads the answer top-down and the opening citation anchors attention. Slot-weighted citation share (sum across answers of slot-weight × disposition-weight) is 1.7–2.3× more predictive of brand-search lift than raw citation count on the same priority sub-query set.

The conversational follow-up turn is the user's second (and subsequent) message in an AI search conversation — a refinement, comparison ask, clarification, or pivot following the engine's first answer. Through 2026 every major engine carries conversation state across turns and uses it to bias both the retrieval substrate and the synthesis prompt of the follow-up answer. Mid-2026 cohort decay: brands cited verbatim on turn 1 retain verbatim citation on turn 2 at 48–61% of head-turn rate; an additional 31% drops from turn 2 to turn 3; another 26% drops from turn 3 to turn 4. Compounded, turn-4 verbatim citation runs at roughly 31% of turn-1 verbatim citation without explicit multi-turn engineering. Roughly 62% of mid-2026 commercial AI search sessions include at least one follow-up turn before the user converts or exits — head-turn-only optimization misses the majority of the conversion-driving turn surface.

Per-engine planning anchors: ChatGPT Search carries 8–12 turns of state on commercial sessions (longest of the general-purpose engines); Perplexity carries 5–8 turns with higher per-turn citation density; Google AI Mode carries 4–7 turns with aggressive state decay after turn 3; Microsoft Copilot carries 6–10 turns with freshness-tilted state-weight decay (the most recent turn carries 70% of the state weight); Amazon Rufus carries 3–5 turns with asymmetric weighting toward product-discovery state; Claude carries 10–15 turns with the deepest entity-graph carryover. The most recent 1–2 turns carry 60–80% of state weight across every engine. Conversion-driving commercial decisions cluster on turns 2–4, well inside every major engine's retention window — the multi-turn engineering opportunity is structurally available, not theoretical.

Signal 1 — Turn-level entity grounding: chunks that name the brand or product entity inside the leading sentence hold citation across the follow-up turn at 1.6–2.0× the rate of chunks that rely on surrounding paragraph or page-level entity context. Signal 2 — Thread-resilient claim shape: chunks anchored on the category and use-case rather than the specific head-query phrasing survive follow-up turn pivots at 1.5–1.9× the rate of head-query-anchored chunks. Signal 3 — Cross-sub-topic schema coverage: pages scaffolded for adjacent sub-topics the follow-up pivots into (FAQPage on the pivot question, HowTo on the pivot use case, Article with named author on the pivot category) survive follow-up pivots at 2.1–2.8× the rate of pages with head-turn-only schema coverage. Signal 4 — Visual entity stability: persona-locked visual sets across the priority page set compound carousel state across turns at 1.8–2.3× the rate of rotating imagery because the multimodal substrate reads persona continuity as a brand-entity stability signal.

Conversation-weighted citation share sums per-turn citation count weighted by turn-position commercial value across captured multi-turn conversations. The metric matters because turns 2–4 carry roughly 2× the conversion weight of turn 1 in mid-2026 cohort data — the user has invested in the conversation by turn 2 and is closer to a purchase or decision by turn 4. Programs scoring raw head-turn citation share underread the conversion-driving citation surface; programs scoring conversation-weighted citation share read where commercial value actually lands. Mid-2026 cohort: conversation-weighted citation share is 1.9–2.4× more predictive of brand-search lift and DTC conversion than head-turn-only citation share on the same priority head-query set. Track it as a separate quarterly metric alongside head-turn citation share — the two trajectories diverge sharply for programs that have shipped multi-turn engineering vs programs still optimizing only the head turn.

The synthesis prompt composes the text answer and the inline image carousel in the same pass on multimodal-active sub-queries — the two surfaces are jointly composed, not pipelined. The carousel's image-1 slot binds to the text's slot-1 chunk on 78% of multimodal-active answers in mid-2026, and a slot-1 verbatim text citation pulls the cited page's persona-locked visual asset into the carousel's top slot at 2.6× the rate of an equivalent slot-3 text citation. Pages that hold the slot-1 text citation but lose the carousel slot halve the per-page citation contribution — the parallel visual synthesis signals (ImageObject schema density, persona stability across the page set, caption alignment with the cited paragraph, image freshness on the 4–12 week window, content-hash recency) need their own audit alongside the text-side synthesis audit. ppl.studio's persona-locked visual production is the operational fit for the visual side of the joint composition.

AI Search Answer-Anchor Sentence Engineering

The answer-anchor sentence is the single sentence-length hyperlink text an AI search engine renders as the visible citation on the answer surface. Every major engine through 2026 renders a verbatim citation as three surfaces — a numbered source chip, a short quoted or bolded fragment, and a sentence-length hyperlink anchor the user actually clicks. The anchor is not the whole cited chunk; it is one sentence inside the chunk the synthesis stage picks as the click-target. Roughly 34% of mid-2026 verbatim citations lose click-through because the engine picks the wrong sentence — a topical framing sentence, an entity-generic sentence, or a truncated tail sentence — even when the chunk itself survives synthesis. Chunk-level metrics report the chunk as cited; sentence-level metrics report the anchor as lost. The two readings look identical to a program scoring only verbatim citation rate, which is why anchor-slot survival needs its own metric, audit cadence, and editorial backlog.

Google AI Mode anchors on the leading sentence 68% of the time when it carries an explicit numeric statistic or a named-entity subject; ChatGPT Search anchors on the highest-numeric-density sentence in the chunk 58% of the time (content-shaped rather than position-shaped selection); Perplexity rotates the anchor sentence across sessions and stabilizes on the highest-engagement sentence over a 4–8 week window; Microsoft Copilot anchors on the freshest-signal sentence 46% of the time (dated-claim chunks shift the picker even when the leading sentence is stronger); Amazon Rufus runs an asymmetric anchor policy (product-specific sentence on the discovery branch, scenario-descriptive on the use-case branch); Claude anchors on the most reasoning-dense sentence 54% of the time. Across the four highest-volume general-purpose engines, roughly 62% of verbatim citations anchor on the chunk's leading sentence, 22% on the second sentence, 10% on a middle sentence, and 6% on the tail sentence.

Five sentence-level properties compose multiplicatively into anchor-slot survival. (1) Anchor signal density — sentences with a numeric statistic and named-entity token in the first 40% of the sentence win the anchor slot at 2.1× the rate of equivalents with either signal alone. (2) Fragment integrity — sentences that render coherently in isolation (no dangling pronouns, no unresolved forward references, no list-continuation markers) win at 1.7–2.0×. (3) Anchor token load — sentences in the engine's preferred anchor length range (17–28 words desktop, 12–22 mobile) win at 1.4–1.8×. (4) Anchor typography compliance — sentences avoiding em-dashes, semicolons, parenthetical asides, and inline citations that force fragment truncation render as coherent anchors at 1.5×. (5) Head-query anchor-word presence — sentences with the head-query keyword in the first 60% of the sentence win at 1.6–1.9×. Composed multiplicatively, these lift anchor-slot survival 2.4–3.8× over synthesis-stage-optimized baselines.

Anchor drift is the shift in which sentence inside a rerank-surviving chunk wins the anchor slot across sessions. Roughly 28% of anchor sentences on rerank-surviving chunks shift within a rolling 8-week window even when the chunk itself has not been edited. Perplexity runs the highest drift rate (roughly 38% within 8 weeks) driven by active anchor sentence rotation; the other general-purpose engines run 18–24%. Four causes require different fixes: substrate re-embedding (the engine re-scores sentence relevance under a revised embedding model), query-intent drift (the head query's intent distribution shifts and re-weights sentence-level scores), competitor sharpening (a competitor's sharper anchor sentence pulls click-through weight on the rendered answer surface without shifting the within-chunk selection), and Perplexity-style anchor rotation (some engines actively rotate to test user engagement). The audit captures anchor sentence identity weekly per priority sub-query per engine, computes drift rate on a rolling 4-week window, and diagnoses the cause before scoping the fix — misdiagnosed drift fixes produce zero lift and let the drift stabilize against the program rather than for it.

Anchor-slot survival rate = anchor slots won / verbatim citations available on the same rerank-surviving chunk universe, scored per priority page on a rolling 4-week window. A rate above 68% is category-leading; 50–68% is competitive; below 50% is exposed. Mid-2026 cohort medians: 46% on mid-market programs, 66% on category-leading programs. Anchor-slot verbatim citations earn 1.5–2.1× the click-through of same-chunk non-anchor verbatim citations at equivalent answer-position — the delta is the operational proof that closing the anchor-slot gap converts existing verbatim citation share into rendered-answer click-through without adding a single new URL. Track anchor-slot survival rate quarter over quarter alongside the synthesis-stage verbatim citation rate; the two metrics move together with a 3–6 week lag once sentence-level edits start shipping.

Loss mode 1 — Anchor picker chose the wrong sentence: the leading sentence reads as topical framing rather than a citable claim, and the picker skips to a weaker candidate or falls through to the tail sentence. Fix is a leading-sentence rewrite into the citable-claim shape ([Entity] [verb] [quantified claim] [optional qualifier]) — roughly 65% of wrong-sentence losses recover the anchor slot on the first cycle. Loss mode 2 — Anchor picker chose the right sentence but the rendered fragment truncates: sentence exceeds the engine's preferred anchor length, or typographic breaks (em-dashes, semicolons, parentheticals) force early truncation. Fix is sentence-length reduction to the target range or typographic simplification (em-dash → comma, remove parentheticals, split into period-separated micro-sentences). Loss mode 3 — Anchor drift under a competitor-sharpening event: a competitor's sharper anchor sentence pulls click-through weight on the rendered answer. Fix is a sharpening rewrite that outperforms the competitor under the current anchor picker weights. Scoping the wrong fix produces no lift and burns editorial bandwidth.

Anchor-sentence engineering is the sixth layer in the AI-search optimization stack — it sits downstream of the synthesis-stage audit because paraphrase citations do not carry an anchor sentence. The synthesis-stage audit converts rerank-surviving chunks into verbatim citations; the anchor-sentence audit converts verbatim citations into visible anchor slots and clicked hyperlinks. The two layers are compositional: a chunk that fails synthesis never enters the anchor-sentence universe, and a chunk that survives synthesis but fails the anchor-sentence pass caps at the verbatim citation rate without capturing the anchor-CTR compounding. Programs that ship both layers lift rendered-answer click-through 2.5–4.6× over rerank-survival-optimized baselines — 1.6–2.2× from synthesis-stage optimization and another 1.5–2.1× from anchor-sentence optimization on top.

AI Search Citation Card Rendering

The citation card is the visible source panel an AI search engine composes alongside the anchor sentence — favicon, publisher-badge string, truncated page-title fragment, timestamp chip, thumbnail slot, and card position in the rendered source strip. It is a first-class rendered element the engine composes from a small set of signals pulled off the cited page and off the publisher record, not the URL preview a browser paints under a hovered link. Roughly 27% of mid-2026 rendered-answer click-throughs come from the citation card surface rather than from the anchor sentence — the user reads the card first, decides whether the source is credible, and only then clicks the anchor. A card that renders in a degraded shape (broken favicon, publisher badge collapsed to bare domain, page title truncated mid-brand-name, stale timestamp, missing thumbnail, or card demoted to position four) suppresses the click even when the anchor sentence is well-formed. The loss is silent from the anchor-CTR metric because the anchor was still visible, which is why card-rendering compliance needs its own metric, audit cadence, and editorial backlog.

Google AI Mode renders favicon + publisher-badge string + 48-character page-title fragment + timestamp chip; top-3 slots capture 84% of card-surface click-through. ChatGPT Search renders favicon + publisher-badge string + 60-character page-title fragment; no primary-surface timestamp chip; top-3 slots capture 78%. Perplexity renders favicon + publisher-badge string + 55-character page-title fragment + thumbnail slot; top-3 slots capture 74% (shallowest distribution because users scan deeper); Perplexity is the engine most sensitive to card-thumbnail quality. Microsoft Copilot renders favicon + publisher-badge string + 44-character page-title fragment + timestamp chip + publisher-authority checkmark badge on recognized entities; top-3 slots capture 86% (tightest position concentration). Amazon Rufus runs an asymmetric card policy (product image + price + rating on the discovery branch; favicon + publisher badge + 40-character title on the information branch). Claude renders a sparse card — favicon + publisher-badge string + 70-character page-title fragment, no timestamp chip, no thumbnail slot; top-2 slots capture 88%.

Six page-level and publisher-level properties compose the visible click surface. (1) Favicon trust signal — sharp, distinct favicons at 32×32 or larger source resolution lift card-surface CTR 1.2–1.4× over blurry, generic, or missing favicons. (2) Publisher badge recognition — publishers that resolve to a recognizable brand string lift card-surface CTR 1.4–1.6× over publishers rendering as bare domains. (3) Card title truncation compliance — page titles within the engine's card-title budget (44–70 characters depending on engine) capture card-surface CTR 1.3–1.5× over titles that truncate mid-word or mid-brand-name. (4) Card timestamp signal — pages with an aligned dateModified + visible last-updated copy + HTTP Last-Modified header render a timestamp chip that lifts card-surface CTR 1.2–1.4× on engines that render one. (5) Card thumbnail slot compliance — pages exposing a card-preview-ratio image via og:image + twitter:image + ImageObject schema render a thumbnail that lifts card-surface CTR 1.5× on Perplexity and 1.2× on Google AI Mode. (6) Citation card position weight — top-3 source slots capture 74–88% of card-surface click-through depending on the engine. Composed multiplicatively, the six lift card-CTR 1.9–2.8× over card-rendering-defaulted baselines.

Citation card drift is the shift in citation-card composition — publisher badge, favicon rendering, page-title truncation, timestamp chip, thumbnail, or card position — on an anchor-slot-winning chunk across sessions, even when the underlying page has not been edited. Roughly 24% of citation cards on anchor-slot-winning chunks shift within a rolling 8-week window. Four causes require different fixes: publisher-registry re-scoring (the engine re-runs the publisher-badge resolver against an updated registry and previously resolved brand strings collapse to bare domains), UI redesign (the engine ships a card-format revision that changes title budgets or introduces/retires timestamp chips and thumbnail slots — Google AI Mode has shipped three card-format revisions since December 2025), card-position demotion under competitor sharpening (a competitor's publisher-authority strengthens and pulls card position), and thumbnail slot substitution (the engine picks a different image from the cited page as the thumbnail). The card-drift audit captures citation-card identity weekly per priority sub-query per engine, computes drift rate on a rolling 4-week window, and diagnoses the cause before scoping the fix — misdiagnosed drift fixes produce zero lift and burn editorial bandwidth.

Card-rendering compliance rate = citations where all six card-rendering properties passed / anchor-slot-winning citations available on the same rerank-surviving chunk universe, scored per priority page on a rolling 4-week window. A rate above 72% is category-leading; 54–72% is competitive; below 54% is exposed. Mid-2026 cohort medians: 51% on mid-market programs, 70% on category-leading programs. Card-rendering-compliant citations earn 1.4–1.9× the click-through of same-anchor non-compliant citations at equivalent card position — the delta is the operational proof that closing the card-rendering gap converts existing anchor-slot survival share into rendered-answer click-through without adding a single new URL or rewriting a single anchor sentence. Track card-rendering compliance rate quarter over quarter alongside the anchor-slot survival rate; the two metrics move together with a 4–8 week lag once card-rendering edits start shipping.

Loss mode 1 — Favicon or publisher-badge failure (broken favicon, unrecognized publisher badge, bare domain rendering). Fix is a sitewide favicon + apple-touch-icon + Organization schema alignment — publisher-level fix that affects every card on every priority page. Loss mode 2 — Page-title truncation on a load-bearing token (title truncates mid-brand-name or mid-topic, dropping the recognizable string that would have driven the click). Fix is a title-tag rewrite to front-load the load-bearing token in the first 40 characters — page-level fix with roughly 60% first-cycle recovery. Loss mode 3 — Missing or stale timestamp chip (dateModified misaligned with visible last-updated copy, or HTTP Last-Modified header disagreeing with schema). Fix is timestamp alignment on next refresh — page-level fix that composes with the freshness-window audit. Loss mode 4 — Card thumbnail substitution or absence (og:image missing, wrong ratio, or replaced with a generic content image). Fix is an og:image + twitter:image + ImageObject schema update to a card-preview-ratio persona-locked visual asset — page-level fix that composes with the multimodal answer audit. Scoping the wrong fix produces no lift and burns editorial bandwidth.

Card-rendering engineering is the seventh layer in the AI-search optimization stack — it sits downstream of the anchor-sentence audit because a chunk that fails to win the anchor slot never renders as a citation card the user sees. The anchor-sentence audit converts verbatim citations into visible anchor slots; the card-rendering audit converts anchor-slot wins into clicked hyperlinks. The two layers are compositional: a chunk that fails anchor-slot survival never enters the card-rendering universe, and a chunk that survives the anchor slot but fails the card-rendering pass caps at the anchor-CTR ceiling without capturing the card-CTR compounding. Programs that ship both layers lift rendered-answer click-through 2.1–4.0× over anchor-sentence-only programs — 1.5–2.1× from anchor-sentence optimization and another 1.4–1.9× from card-rendering optimization on top. Composed across the full seven-layer stack, rendered-answer click-through on the priority set lifts 3.5–7.2× over rerank-survival-only programs.

Ad Performance & Workflow

Yes — UGC-style creative consistently outperforms polished brand creative on CTR and CPA across Meta, TikTok, and Pinterest, because it looks native to the feed rather than like an ad. AI UGC delivers the same visual qualities at 50–200x lower per-asset cost than human creators, which is what makes ad-creative testing at meaningful volume economically viable.

Meta's algorithm needs ~50 conversions per ad set to exit learning, so structure tests at the angle/hook level rather than per-image. A typical first test is 3 angles × 3 hooks × 2 formats (photo + video) = 18 variants. Use our free ROAS Calculator to set your break-even threshold before spend goes live.

Plan on refreshing every 7–14 days for active campaigns — our free Creative Fatigue Calculator sets the exact cadence based on your spend and audience size. Each refresh reuses the same AI Expert and Props but swaps hooks and scenes.

Anywhere UGC-style content performs: Meta paid social (Feed and Reels), TikTok ads and TikTok Shop, Instagram Reels and Stories, Pinterest, YouTube Shorts, Amazon listing photos and A+ content, Shopify product pages, email marketing, and influencer-style organic content. The same export covers most ratios.

Yes. Visit /workflows for end-to-end funnels including Creator UGC Video, TikTok + Reels short-form video, TikTok carousels and animated story ads — each one walks through research, scripting, casting, production, and launch with the exact tools to use at each step.

Still have questions? Reach out at hello@ppl.studio or visit the contact page.

Try ppl.studio free