AI search is becoming another way shoppers research products: they ask ChatGPT, Perplexity or Google’s AI Overviews which option suits them, and those tools answer with a summary and a few cited sources. For AI search, ecommerce stores get cited mostly on the strength of things that already matter for SEO: crawlable pages, clear structured data, a consistent brand identity across the web, and content that answers questions directly. There is no special file or trick that guarantees a citation, and some popular tactics do nothing at all.
How AI search tools find and cite stores
The tools work differently, but they share a basic pattern: they retrieve pages relevant to the question (from their own crawler, a search index or both), then write an answer and cite the pages they relied on. That means two things have to be true for your store to be cited:
- Your pages must be retrievable. The relevant crawler must be allowed, and the content must be visible in the HTML rather than only after scripts run.
- Your page must be the best source for a specific point in the answer: a clear fact, a comparison, a definition, a recommendation with reasons.
Google states plainly that there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible to show a snippet, and that its usual SEO guidance applies. That is a useful anchor: most of what helps here is good technical SEO done properly.
Let the right crawlers in
Check your robots.txt against the crawlers that matter. OpenAI documents separate agents for search (OAI-SearchBot) and for model training (GPTBot); Perplexity uses PerplexityBot; Google’s AI features use Google’s normal crawling. You can allow search-related crawlers while blocking training crawlers if that is your policy; the important thing is to decide deliberately rather than inheriting a default from a security plugin or a CDN setting. We regularly find stores that block AI crawlers without anyone having decided to.
Also check that key content does not depend on client-side rendering. Product details, prices, specifications and FAQs should be in the HTML the server sends. Some crawlers do not run JavaScript at all.
Structured data: say what things are
Structured data will not make an AI tool cite you on its own, but it removes ambiguity about what a page is and what it contains, which helps every system that reads it.
- Product pages: Product markup with name, brand, price, availability, and reviews if you have genuine ones.
- Organization on the homepage or about page: legal name, logo, contact details and
sameAslinks to your official profiles. - Breadcrumbs on product and category pages.
- Articles for guides, with a real author.
Validate it, and keep it in sync with what the page shows. Markup that claims a price or rating the page does not show is a quality problem, not an optimisation.
Entity signals: be a clearly identifiable brand
AI tools answer questions about brands by combining what many sources say. You help by being consistent and easy to identify:
- The same business name, address and contact details on your site, marketplace listings and business profiles.
- An about page that says what you sell, since when, where you ship from and who runs the business.
- Official profiles linked from your site and linking back.
- Mentions in places that are relevant to your category: supplier listings, industry associations, genuine reviews, press.
None of this is new. It is the same work that builds trust with search engines and shoppers, now also read by language models.
Write answer-first content
The pages that get cited most in our experience answer one question clearly near the top, then support the answer with detail. For stores that means:
- Buying guides that start with the recommendation (“For wet trails, choose a shoe with…”) and then explain why.
- Comparison pages with a table, not only paragraphs.
- Specific, checkable facts on product pages: dimensions, materials, compatibility, care. Vague marketing copy gives a model nothing to quote.
- FAQs written from real customer questions, answered in two or three sentences.
- Category buying help, which we cover in ecommerce category page SEO.
A short summary at the top of a long guide (like the one on this page) is useful for readers, and it also gives any system summarising the page an accurate version to work from.
A worked example: one product page
Take a typical product page for a hiking backpack. The version we often see opens with a tagline (“Adventure awaits”), has three sentences of mood copy, a specifications tab that only loads when clicked, and reviews rendered by a script after the page loads. A person can use it. A system trying to answer “which 30-litre backpack fits carry-on limits?” finds very little it can quote.
The rewritten version keeps the design but changes what is on the page:
- An opening sentence that states what the product is and who it is for: capacity, weight, what it fits, where it is not suitable.
- Specifications in the HTML, visible without clicking: dimensions in centimetres and inches, materials, volume, weight.
- A short “Will it fit?” section answering the carry-on question directly, with the dimensions that decide it.
- Three or four FAQs taken from real customer questions.
- Product and Breadcrumb structured data that match the page, and reviews in the HTML if you have them.
Nothing about this is written for machines. Customers asking the same question in your own site search get a better answer too, which is usually the best test that a change is worth making.
llms.txt: useful, but not a ranking switch
llms.txt is a proposed plain-text file at the root of a site that lists its most important pages with short descriptions, written for language models. It is cheap to publish, and it can help tools that choose to read it find your key pages quickly. It is not a standard that the major AI search products have committed to use, and Google does not require any such file. We publish one on our own site because it costs little and documents our key pages well; we would not expect it to change citations by itself.
What doesn’t work
| Tactic | Why it doesn’t help |
|---|---|
| Mass-produced AI articles targeting every question | Thin, repetitive pages rarely become the best source for anything, and they can drag down how search engines assess the whole site |
| Hidden text or text written only for bots | Violates search spam policies; models favour content users can see |
| Keyword-stuffed FAQs | Answers that say nothing specific give nothing to cite |
| Markup that doesn’t match the page | Inconsistent data reduces trust rather than adding it |
| Blocking crawlers “to protect content” while wanting citations | Tools cannot cite pages they cannot read |
| Relying on llms.txt alone | Optional and not universally read; it complements crawlable pages, it does not replace them |
How to measure it
Measurement is still rough. What we use:
- Referral traffic from AI tools in analytics (chatgpt.com, perplexity.ai and similar referrers), tracked as its own channel.
- A fixed list of [X] questions your customers ask, checked monthly in each tool, noting whether and how you are cited.
- Search Console for Google, where AI Overviews traffic is reported within normal web search performance.
Trends over months matter more than any single check; answers vary between sessions.
Where to start
- Check robots.txt and server-rendered content on your key templates.
- Fix and validate Product, Organization and Breadcrumb structured data.
- Rewrite your top category and product pages to answer questions first, with specific facts.
- Make your brand details consistent everywhere they appear.
- Set up measurement before you change anything else.
This work sits between our ecommerce SEO service, which covers crawling, structured data and content, and our AI automation work. If you are also thinking about using AI inside the business rather than being found by it, read what to automate first, and what to keep human.