AI assistants don't read your website first — here's where they actually research your brand
This week we ran one of our deeper tests for a just-launched supplement brand: thirty realistic shopper questions — "what should I take for X", "good gift for someone who Y" — asked live across Claude, ChatGPT, and Gemini, with web search on and the brand never mentioned. The kind of questions this store's product is a genuinely good answer to.
The brand was named in zero of the thirty answers.
That part wasn't the surprise — it's a young brand, and new stores start invisible. The surprise was where the assistants went to decide who to recommend instead. We log every page they cite while answering, and the list for this one category looked like this: a supplement review site, read in five separate answers. A health-content portal, read in three. WebMD. A recovery-clinic blog. A newspaper's product roundup. Amazon listings. A competitor's own blog, which several assistants treated as a neutral source on the whole category.
Not one answer was built primarily from any brand's own website.
The assistant is a researcher, not a visitor
When a shopper asks a web-connected assistant what to buy, it behaves less like a customer browsing your store and more like a hurried journalist: it runs a few searches, skims the pages that rank, and synthesizes an answer from what those pages say. Your homepage copy — the positioning you spent weeks on — usually isn't in the room when the recommendation gets made.
The pages that are in the room are mostly third-party: "best X for Y" roundups, category review sites, health and hobby portals, Reddit threads, news listicles. Whoever is featured on those pages gets named. Whoever isn't, doesn't exist — no matter how good the product is or how polished the website looks.
We see the same pattern in category after category we test: the assistants lean on a fairly small, stable set of sources per niche. Five to ten pages, cited over and over, quietly deciding who gets recommended to every shopper who asks.
It gets stranger: the assistants can confuse you with someone else
The same test surfaced a second problem we now check for routinely. This brand shares its name with two completely unrelated companies in adjacent categories. Asked about the store directly, one assistant confidently described it as selling a product line that actually belongs to one of the name-twins — it had stitched an identity together from several same-named brands' pages and presented the mix as fact. A shopper would never know.
If your brand name is a common phrase, or collides with another company's, this is worth testing today: ask each assistant about your store by name, with your URL, and read closely. What they get wrong is rarely malicious and usually traceable — it came from a page about somebody else.
What you can actually do
The uncomfortable version of this finding: AI visibility is mostly earned on pages you don't own. The useful version: those pages are findable, finite, and pitchable. This is work a founder can start this week, free.
1. Find your category's source list. Ask ChatGPT, Claude, and Gemini a handful of unbranded shopper questions in your category ("best [your category] for [your customer's problem] — which should I buy?") and note which sites they cite. You'll see the same names repeat quickly. That repetition is the point — it's a short list.
2. Check who's featured on them. Open each one. Is your brand there? Are your competitors? Every page that features them and not you is converting shoppers you never see — the referral happens inside the chat.
3. Pitch the ones that don't know you. Review roundups and category sites update; most accept products for consideration, and niche ones answer their email. This is classic PR, aimed by data instead of guesswork: you know exactly which placements the assistants actually read.
*4. Make what they do read of you machine-clear.* When an assistant does land on your site, it should find unambiguous, machine-readable facts: real product schema, honest descriptions, policies, and — if you share a name with someone — enough distinct identity (consistent name, domain, product identifiers) that you can't be confused with them. We covered the fundamentals in why AI agents recommend some products and skip yours.
The metric that makes it manageable
None of this is one-and-done — it's a number you move. That's why our detailed report now measures it directly: how many of ~30 real shopper answers name you, who gets named instead, and the exact pages the assistants read before answering, flagged by whether you're featured on them. Run the outreach, re-run the report, watch the number change.
The free version tells you where you stand today — it takes two minutes at assessment.sqrly.ai/check.
Sqrly helps online retailers show up — knowledgeably — for every shopper and every AI assistant. Join the waitlist.