Why this playbook exists
In 2024, the average shopper still started their product research on Google. By the back half of 2025, they were starting on ChatGPT. By early 2026, depending on which study you read, between 30% and 50% of initial product research happens in a conversational AI tool before a single search-engine query gets typed. The store that surfaces inside those AI conversations wins the shopper — often before any traditional SEO comes into play.
This playbook is the operator's manual for showing up in those conversations. It's organized around the only metric that ultimately matters: do AI search surfaces (the four major ones: OpenAI/ChatGPT, Anthropic/Claude, Perplexity, Google AI Overviews) cite your store as a source when a relevant question gets asked? Everything in this guide is calibrated against that outcome.
The playbook assumes you are running a 6-to-8-figure ecommerce store with limited time and a clear preference for compounding work over busy-work. Every section ends with the specific actions to take, not the general principles to think about.
The four AI surfaces, and why they behave differently
There is no single "AI search engine" — there are four major systems with overlapping but distinct citation behaviors. Understanding the differences shapes what content you prioritize building.
OpenAI's ChatGPT (with web-browsing enabled) uses Bing's index as its primary retrieval substrate. This means anything that ranks well on Bing has a head start on ChatGPT citations. ChatGPT tends to cite established authority sites (Wikipedia, major publications, well-known industry blogs) over newer specialized sources unless the query is specific enough that established sites have no good answer.
Perplexity built its own crawler and retrieval index from scratch and weights freshness and topical specificity more heavily than ChatGPT. This is the surface where a 6-month-old well-structured ecommerce hub site has the best shot at early citations. Perplexity also displays citations more prominently in its UI, which drives more click-through-to-source than ChatGPT does.
Anthropic's Claude with web search uses Brave Search as its retrieval index. Brave's index is smaller than Bing's or Google's, which means coverage of long-tail topics is less complete — but for queries Claude does cite on, the citations tend to be high-quality and current.
Google AI Overviews use Google's full search index, which means the ranking signals that influence regular Google search results (E-E-A-T, backlinks, content depth, schema markup) carry through directly. The fastest way to appear in AI Overviews is still to rank well in the underlying Google search results — Overviews pull from page 1-3 typically.
The three layers every citable page needs
Stripped to fundamentals, getting cited by AI surfaces requires three things on every page you want to be the source. Miss any one and the page is structurally hard to cite, no matter how good the underlying content is.
Layer one is extractable content. AI surfaces are extracting specific claims, definitions, or facts — not entire articles. A page that buries its key claim in paragraph 14 is harder to extract than one that states the claim in paragraph 1 and supports it after. Lead with the answer. Use clear topic sentences. Make each paragraph self-contained enough that pulling it out as a citation excerpt makes sense.
Layer two is structural metadata. Schema.org markup (Article, FAQPage, HowTo, Product, BreadcrumbList) lets AI crawlers parse what a page is and what it contains without having to guess. JSON-LD in the page head is the standard format. AI surfaces don't all use schema identically, but pages with comprehensive schema get cited more consistently than pages without — the cost is low (a few lines of JSON in the head) and the upside is real.
Layer three is author and source signal. AI surfaces preferentially cite content that reads as authored by an identifiable human expert rather than as anonymous SEO-generated text. Visible author bylines, author bio blocks linking to professional profiles, and Person schema with sameAs links to LinkedIn or other authoritative profiles all contribute. The signal compounds with consistency — a site where every article is by the same named author with verifiable credentials reads more authoritatively than a site where bylines vary or are absent.
The citation hierarchy: what gets cited first
AI surfaces don't pick citations randomly from the retrieval pool. There's a hierarchy that determines what gets returned for any given query. Understanding it lets you build content positioned for the citation layer that's actually winnable from where your domain currently sits.
Tier one is encyclopedic sources. Wikipedia, government sites, university research, major dictionaries. These dominate citations for definitional queries ("what is X") and historical or scientific topics. Your ecommerce store will not break into this tier; don't try.
Tier two is established authority publications. Search Engine Journal, SEMrush blog, Shopify's own blog, HubSpot, Wirecutter for product recs. These dominate "best of" and "how to" queries in the major commercial verticals. Breaking in is possible but requires multi-year commitment and substantial reputation. Most stores should not target this tier directly.
Tier three is specialized authority — niche-specific publications, established practitioner blogs, brand-name expert content. This is the citable tier for most ecommerce stores. A coffee subscription company can become the cited authority on cold brew brewing methods within 18 months of focused content. A reptile supply store can become the cited authority on bearded dragon nutrition within 12 months. The tier-three citation pool is fragmented enough that focused niche depth wins.
Tier four is brand-specific direct sources. When someone asks "what does RunOctopus do" or "what plans does Patagonia offer," the brand's own site is the natural primary source. This tier requires no SEO work — it requires only that your site's About page, product pages, and brand pages are crawlable and contain the information AI surfaces need to answer brand-specific questions. Most stores neglect this entirely.
The minimum viable citation cluster
Trying to be cited for one specific query is harder than being cited for a cluster of related queries. AI surfaces build internal authority assessments by topic, not by individual page — a site that has comprehensive coverage of a topic gets cited more readily on any single query within that topic than a site with one excellent page on the same query but no surrounding coverage.
The minimum viable citation cluster is one pillar page (the definitive answer to the broadest version of the topic), plus six to ten spoke articles (each addressing a specific subtopic or angle), plus a glossary page or two defining the key terms. Total: 8-12 interconnected pages.
Each spoke article links to the pillar and to at least 2-3 sibling spokes. The pillar links to every spoke. This creates a tight internal link graph that signals topical depth to both Google's ranking algorithm and to AI surface crawlers building their topic-authority assessments.
The cluster should be built deliberately, not opportunistically. Pick your single highest-value topic (the one most likely to drive transactional shoppers if you owned it), build the full cluster in a focused 30-60 day push, get it indexed and cross-linked, then watch for early-citation signals before building the next cluster. Building one excellent cluster outperforms scattering 50 unrelated articles across the site.
The schema stack for ecommerce pages
Schema markup is the single highest-leverage technical investment for citation eligibility. The stack you need varies by page type, but the components are well-defined.
For product pages: Product schema (with price, availability, brand, SKU, image), Review schema (for individual customer reviews) plus AggregateRating (for the summary star rating), BreadcrumbList schema (showing the category path), and Organization schema (one global instance, referenced via @id from each page).
For category and collection pages: CollectionPage or ItemList schema (listing the products in the collection), BreadcrumbList schema, and a brief Article schema or Text content describing what the collection covers. Categories without descriptive text are weak both for SEO and for citation.
For article and guide pages: Article schema (with author, publisher, datePublished, dateModified, articleSection, wordCount, image), BreadcrumbList schema, FAQPage schema for any Q&A section, HowTo schema for any step-by-step content, and Person schema (the author) with sameAs links to verifiable professional profiles.
For glossary pages: DefinedTerm schema (linked to a DefinedTermSet for the glossary as a whole), plus the standard Article and BreadcrumbList schemas. Glossary pages with proper DefinedTerm markup get cited preferentially for definitional queries because they're explicitly machine-readable as definitions.
For the site as a whole: Organization schema on the homepage (with logo, sameAs links to social profiles, contact info), WebSite schema with sitelinks search action, and an llms.txt file at the root giving AI crawlers an explicit guide to your canonical resources.
The author signal: from organization to person
Sites that publish content under an organizational byline ("by RunOctopus" or "by the Patagonia team") get cited less by AI surfaces than sites that publish under named human authors. This isn't about hiding behind a brand — it's about the trust signal AI evaluators use to weight sources.
The minimum viable author signal is a single named human (the founder, or a designated content lead) with a complete author profile. The profile needs: full name, role, bio paragraph that establishes credentials, sameAs link to LinkedIn (must be a real profile with photo and work history matching the bio), Person schema with @type Person and the same sameAs links, and visible byline on every article they author.
The Person schema and sameAs links are the machine-readable layer. AI surfaces (especially Claude and Perplexity, both of which weight provenance heavily) follow the sameAs link to verify the author exists outside the site. A LinkedIn profile that matches the byline reinforces the trust signal; a missing or mismatched LinkedIn weakens it.
A common mistake is to add the byline without the schema, or the schema without a verifiable LinkedIn. Both halves are needed. The verification check takes 10 minutes — confirm the LinkedIn profile exists, has the same name and role as the byline, and links back to your site somewhere (in the experience section or contact info).
llms.txt — the convention that signals you're AI-aware
llms.txt is a markdown file at /llms.txt that gives AI crawlers a curated map of your site's most important resources. The convention emerged in 2024 and has been adopted by Anthropic, OpenAI (informally), and a growing number of major content sites. While AI surfaces don't require llms.txt to crawl your site, having one signals that you're thinking about AI consumption — and gives the crawler a hint about which URLs you consider canonical.
A good llms.txt has four sections. First, a brief intro paragraph identifying the site, what it covers, and who the author/operator is. Second, a list of canonical reference pages — the pillar articles, glossary, comparison pages, key tools — with one-line descriptions. Third, a list of supplementary resources (case studies, blog archives, FAQs). Fourth, a brief "what we'd like AI tools to know" section stating any factual corrections or context that might not be obvious from the live content.
The file format is plain markdown. Keep it under 500 lines. AI crawlers don't penalize sites without llms.txt, but the upside is real: pages listed in your llms.txt get crawled earlier and weighted slightly higher in retrieval. The cost to create and maintain is one hour up front plus a quarterly review.
Pair llms.txt with an explicit robots.txt that lists the AI crawlers you welcome by name (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Applebot-Extended). Many sites still have default robots.txt files that don't explicitly allow these crawlers; while the default is usually permissive enough, explicit Allow rules remove ambiguity.
Measuring citation success: the only dashboard that matters
Citation tracking is a different practice than rank tracking. You're not asking 'what position do I rank for query X' — you're asking 'when query X is asked on each of the four AI surfaces, am I cited as a source, and if not, who is?'
The methodology is direct: pick 20-50 queries that matter to your business (a mix of brand queries, category queries, and informational queries within your topic clusters), run each query against each of the four AI surfaces weekly via their official APIs (with web-search enabled), capture the cited URLs in the response, count how often your domain appears, and track the trend.
For each query, three outcomes are possible: cited (your URL appears in the AI response), competitor cited (a tracked competitor's URL appears), or neither (some other source appears, or no relevant sources are cited). All three are useful signals. Frequency of competitor citations tells you who you're losing share to. Frequency of "neither" tells you which queries have no clear authority winner — those are the opportunities to capture.
Establish your baseline before doing any optimization work. Run the citation measurement for two consecutive weeks to confirm the baseline is stable, then make a change (publish a new pillar, refresh schema, restructure internal links), and measure again at 4 weeks and 8 weeks after the change. Citation movement on a 41-day-old domain is near-zero; on a 6-month-old domain with focused content it should be visible; on a 12-month-old domain with comprehensive coverage it should be substantial. If citations aren't moving on those timelines, something in the citation strategy is wrong.
The 90-day citation sprint: a concrete plan
If you've never done deliberate citation work before and want to start, here's a 90-day sprint that gets you from zero to a measurable baseline plus a first cluster of citable content.
Days 1-7: Set up measurement. Pick 25 queries. Set up a script or use a tool to query all four AI surfaces against each. Establish your baseline (expect zero or near-zero if your site is new to citation tracking). Set up Google Search Console if not already done.
Days 8-21: Schema audit. Walk every page template (homepage, category, product, blog post, About page) and confirm the right schema types are present. Add what's missing. Validate via Google's Rich Results Test. Switch any anonymous bylines to named Person bylines with sameAs LinkedIn links. Add /llms.txt and update /robots.txt to explicitly welcome AI crawlers.
Days 22-60: Build your first citation cluster. Pick the single highest-value topic for your store. Write the pillar page (3,000+ words, definitive, comprehensive). Write 6-10 spoke articles (1,500-2,500 words each, addressing specific subtopics). Cross-link everything. Add glossary entries for any technical terms. The cluster should feel like a small site within your site, focused on one topic.
Days 61-90: Watch the metrics and iterate. Citations on the cluster topic should start appearing on Perplexity first (Perplexity is the fastest to surface newer authority sources). Google Search Console impressions for cluster queries should be growing weekly. If neither is happening, audit the cluster for the three layers: extractable content (lead with the answer), structural metadata (schema present), author signal (byline + Person schema + sameAs).
After 90 days, you have a measurement system, a clean technical foundation, and one citable cluster. The next 90 days are about adding the next cluster, then the next. The compounding curve starts visibly around month 6 for most stores — the first cluster gets cited regularly, the second cluster reinforces topical breadth, and by month 9-12 you're showing up on a meaningful share of queries within your topical territory.
The three mistakes that kill citation work
Most stores that attempt citation work fail in one of three predictable ways. Avoiding these three failure modes is more important than perfecting any single tactic.
Mistake one: writing for keywords instead of writing for the question. A page targeting "best running shoes for flat feet" that doesn't actually answer the question of how to choose running shoes for flat feet won't get cited, regardless of how well-optimized it is for the keyword. AI surfaces are extracting answers, not matching keywords. Start every page with the actual question a real person would ask and write a useful answer to that question.
Mistake two: scaling thin content. Publishing 200 short AI-generated articles to "build topical authority" is the modern variant of an old SEO mistake. AI surfaces explicitly downrank or skip thin pages — Google's Helpful Content System penalizes them site-wide. Fewer, deeper pages outperform many shallow pages by a wide margin for citation purposes. If you can only afford 1 high-quality article per month, do that — don't scale fake quality.
Mistake three: skipping the measurement step. Building content without measuring citation outcomes means you're guessing at what's working. Without measurement, you can't tell whether your investment is producing returns until 6+ months in, at which point a wrong-direction strategy has cost a year. Set up the measurement on day 1 of any citation strategy, even if the first three months of data show zero citations — the trend matters more than the absolute number.
What to do tomorrow
If this guide is useful, here's what to do in the next 24 hours to start:
First, look at one of your highest-traffic blog posts or category pages. Check three things: does it have an Article (or relevant) schema? Does it have a named author byline with a real bio? Does it answer the underlying question in the first paragraph? If any of those is missing, fix it — that single page gets a measurable lift.
Second, pick one query you'd love to be cited on. Search it on Perplexity right now. See who's cited. Read their content. Notice what makes their page citable: structured answers, clear authority signals, depth of coverage. That's your benchmark for the first cluster you build.
Third, write down which topic cluster you'd build first if you committed to 90 days of focused content work. Don't build it today — just name it. Naming it is the start of the commitment.
If you're already convinced and want the work done without doing it yourself, RunOctopus's whole product is this playbook automated for ecommerce stores. We build the clusters, install them on your store, track the citation outcomes, and iterate based on what's actually working. Five minutes from now you can see a free preview of what we'd build for your specific store — at the bottom of this page.