What AI assistants say about your brand, and why

There is a whole industry selling special files and bespoke markup for AI visibility. Google’s documentation on AI features in Search is unusually blunt about it: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features. There’s also no special schema.org structured data that you need to add.”

The eligibility bar it does state is the ordinary one — “a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements” — followed by “there are no additional technical requirements.” And it declines to promise anything: “Just because a page meets all requirements, best practices, and complies with the policies, doesn’t mean that Google will crawl, index, or serve its content. Indexing and serving isn’t guaranteed.”

Read together, that is a specific and useful message. The work is being crawlable, being indexed, and having content clear enough to quote. The work is not a new file in your root directory. If a proposal in front of you is mostly a file and a score, it is selling you the easy half of a problem that does not have an easy half.

None of which makes structured data pointless — we put Product, FAQPage and BreadcrumbList markup on everything, because it earns rich results in classic search and it forces a discipline of stating facts explicitly. Just do not buy it as an AI cheat code, because the people who run the largest AI feature in the world have said in writing that it is not one.

Assistants use named crawlers, and each one does a different job. OpenAI documents GPTBot as the crawler for training generative models and OAI-SearchBot as the one that indexes for search results inside ChatGPT. It also documents ChatGPT-User as user-triggered fetching — a page a person asked the assistant to look at — and notes it is “not used for crawling the web in an automatic fashion,” so robots.txt rules may not govern it. Google, separately, offers Google-Extended as the control for limiting AI training and grounding in some of its other systems, distinct from Googlebot.

The distinction matters more than it sounds. Disallowing GPTBot keeps your content out of training. Disallowing OAI-SearchBot keeps you out of the search index the assistant reaches for when a customer asks for a recommendation. Plenty of brands have blanket-blocked both under one instruction to “stop AI scraping us” and taken themselves out of the shop window to protect a copyright position they had not articulated.

Have the conversation properly, with whoever owns brand and whoever owns legal in the room, and decide the two separately. We would generally allow search crawling and treat training as a policy call the business makes on its own terms. Either way, decide it — because the default is a decision too.

Shopify serves a default robots.txt and it is fine for most stores; Shopify notes that “the default rules are updated regularly to ensure that SEO best practices are always applied.” When you do need your own rules, the mechanism is a robots.txt.liquid template in the theme, which outputs groups of rules using the robots Liquid object. Shopify is specific that it “can’t be a JSON template” and that while you can replace the whole thing with plain text rules, it is “strongly recommended to use the provided Liquid objects whenever possible.”

Follow that recommendation. Iterating over robots.default_groups and appending your own rules means you inherit Shopify’s ongoing updates; pasting a static file means you inherit a snapshot from the day someone pasted it, and nobody revisits a robots.txt. We have seen a store lose a section of its catalogue from search for months because a hand-written file outlived the reason it was written.

This is the part clients least want to hear. If an assistant is asked to compare three brands in your category, the sentences it has to work with mostly come from people who are not you — reviews with substance, comparison articles, a forum thread where somebody described the fit, a retailer page with full specifications. A brand with a beautiful site and no third-party presence gives a retrieval system nothing to say beyond its own marketing, and marketing copy is exactly the kind of source these systems weight down.

The uncomfortable conclusion is that the highest-leverage AEO budget for many brands is not a website project at all. It is getting reviewed by people who write in detail, getting your specifications onto the places that list your category properly, and making sure your own factual claims are consistent everywhere they appear — because a contradiction between your site and a retailer’s listing is resolved by a machine picking one, and it may not pick yours.

What this does mean for the site: your facts need to be stated somewhere quotable, in text, in one obvious place. Materials, dimensions, care, compatibility, delivery times, returns terms, sizing. Every one of those we routinely find either absent, trapped in an image, or loaded by an app after the page renders — which for retrieval purposes is the same as absent.

All of the above is a content programme, and content programmes die on the same thing: publishing a page takes a developer two weeks. If answering a new customer question requires a ticket, the questions go unanswered and the category gets described by whoever did answer them.

This is why a composable section library is the piece of infrastructure that actually moves AEO for most stores. Merchandisers assembling a proper answer page — hero, comparison table, numbered steps, FAQ — on a Tuesday afternoon is worth more over a year than any markup you could add. The tactic list is cheap; the ability to execute it repeatedly is not.

Be honest that this is poorly measurable and refuse to pretend otherwise. There is no rank tracker for an assistant’s answer, and the visibility scores being sold are models of a black box rather than observations of one.

What you can do: sample. Write down the fifteen questions a real buyer in your category asks, run them across the assistants your customers use, and record what comes back and which sources it cited — the same fifteen, monthly, by the same method. That is a crude instrument, but it is the only one with actual signal in it, and it catches the thing that matters most: a factual error about your brand repeating across systems.

Then watch the things you can measure and accept that attribution is partial. Referral traffic from assistants where it identifies itself, branded search volume, and direct traffic that arrives already knowing what it wants. If someone offers you a single AI visibility number, ask how it is derived; if the answer is a proprietary model, you are buying a reassuring chart.

• Shopify section library
• Collection pages are your best SEO asset
• Faceted navigation and crawl budget