AI search visibility

AI search visibility for ecommerce.

AI search visibility is whether an AI answer engine — ChatGPT, Perplexity, Google AI Overviews, Gemini — names your store when a shopper asks it what to buy. It is not a ranking. There is no position four inside an answer: your brand is either in the shortlist the assistant writes, or it is absent from the decision entirely.

Deadpan Labs is an ecommerce growth company in Miami. We measure that visibility for Shopify and DTC brands with a monthly panel of the buyer questions their customers actually ask, and we run the work that changes the answer.

Last updated 2026-08-22 · Deadpan Labs · Miami, FL

The distinction

Being named is not the same as ranking.

Every instinct a store has about search was trained on a list of ten blue links. An answer engine does not produce a list of ten anything. It reads, decides, and writes a recommendation in a paragraph, and the shopper acts on that paragraph. Four things change as a result.

The unit

A named brand, not a ranked page

Classic search returns an ordered list of documents and lets the shopper choose. An answer engine writes a recommendation and names two to five brands inside it. The output is a sentence, not a results page, so the thing being optimized is whether your name appears in that sentence.

The sources

It answers from what it already trusts

An answer engine composes from sources it retrieves — independent roundups, review platforms, community threads, retailer listings, and product data it can parse. Your own site is one input among many, and rarely the deciding one for a category question.

The tail

There is no position four

Ranking degrades gracefully: position eight still gets some clicks. Being unnamed does not degrade. If the assistant writes a shortlist of four and you are not on it, you received nothing from that query, and the shopper never saw a page they could have scrolled past you on.

The variance

The same question gets different answers

Two runs of one query minutes apart can name different brands. That is normal behavior for these systems, and it is why a single screenshot proves nothing in either direction — good or bad. Only a repeated panel over time reads as a signal.

Why ecommerce first

Shopping questions are the ones assistants answer best.

An assistant is cautious about medical advice and legal advice. It is not remotely cautious about telling someone which moisturizer to buy. Product recommendation is the safest, most repeated, most obviously useful thing these systems do, so it is where the behavior moved first — and where it moved hardest.

Adobe Analytics, measuring more than a trillion visits to U.S. retail sites, reported that traffic referred by generative AI sources rose 693% year over year across the November–December 2025 holiday period, that revenue per visit from those referrals was up 254%, and that AI-referred shoppers converted 31% better than other traffic. The volume is still small next to classic search. The direction and the quality are not in dispute.

Three things make a store more exposed than a software company. A store has many products, so the question is asked per category rather than once about the brand — you can be visible for one product line and invisible for the next. A store competes against retailers and marketplaces that already own the roundups. And a store’s buyers are consumers, who adopted assistants faster than procurement departments did. The result is that a store is usually being evaluated in these answers long before anyone at the store has checked.

The diagnosis

Six reasons a store goes unrecommended.

Invisibility in AI answers is rarely one broken switch, which is why “we added a blog post” never moves it. It is a specific, diagnosable combination. In practice, almost every store we look at is carrying more than one of these at once.

1 · Nothing independent has been written about you. AI shopping answers lean hardest on third-party sources: maintained “best X” roundups, review sites, niche blogs, community threads. If no independent source has ever covered your product, the engine has nothing to cite, and no amount of on-site work creates a citation that does not exist.
2 · Your product pages are not machine-readable. Thin or missing Product structured data, specs that live only inside images, prices and availability rendered by client-side script. The symptom is a page that looks complete to a shopper and is close to empty to a parser.
3 · Your rating signals sit where the engines do not look. Testimonials on your own site carry far less weight than ratings on retailer listings and established review platforms. A store can have hundreds of happy customers and no rating signal the engine can read.
4 · Your language does not match the question. Shoppers ask assistants in problem language — “cream for a damaged skin barrier,” “a harness that a dog cannot slip out of.” If your pages and mentions only ever use category or brand language, the retrieval step never reaches you.
5 · Your entity is ambiguous. The engine is not certain what you are, who you are for, or that the brand named on three different sites is one company. Ambiguity is resolved by skipping you, not by guessing.
6 · A competitor already owns the default answer. Citations compound: the brand that keeps getting named keeps getting cited, and each new roundup copies the previous one. This one is not a defect in your store at all, which is why it is the one most often misdiagnosed.

Which of the six you are carrying decides everything about what the work costs and how long it takes. Two stores that are equally invisible can need completely different work. That is the whole reason the first honest step is a measurement rather than a proposal.

One thing to skip

llms.txt is not the answer, and the evidence is public.

For about two years, most advice on this subject has opened by telling stores to publish an llms.txt file. It is worth saying plainly that this is not where the win is.

Google’s John Mueller has said the file is not used for Search. Ahrefs analyzed 137,210 domains and found that 97% of the llms.txt files it examined received zero requests in May 2026 — nothing fetched them at all. It costs nothing to publish and it breaks nothing, so keep one if you like. Just do not mistake it for the work.

What the crawlers do read is unglamorous: server-rendered HTML they can parse without executing your JavaScript, complete structured data on product and organization pages, and consistent signals about who you are across the places you appear. We changed our own guide when this evidence came in, rather than leaving the older claim standing.

Check it yourself

How to tell if this is happening to you.

This takes an afternoon and costs nothing. Do it before you talk to anyone, including us.

Open ChatGPT, Perplexity and Google’s AI results, ask them the questions your customers ask before buying — in their words, not your brand’s — and read the answer rather than the links. If you are not named, the useful question is not “why” but “who is, and what did the engine read to decide.” Answering that reliably, month after month, is the part that takes a method; ours is a fixed monthly panel, and the first read is free.

Most owners find one of two things. Either they are not named anywhere, which is uncomfortable and useful. Or they are named in one category and absent in three others, which is more common and more actionable, because it means the machinery works and the coverage is uneven.

Measurement

What a monthly prompt panel measures.

A prompt panel is a fixed set of real buyer questions, asked across the answer engines on a fixed schedule, with the results logged the same way every time. Most of the difficulty is in that word “fixed”, and it is the reason a screenshot and a measurement are different things.

Three things come out of it:

1 · Whether you are named, and how consistently — a single answer proves nothing in either direction, good or bad.
2 · Who is named instead of you.
3 · Which sources the engines read to decide, because that is where the work has to happen.

What a panel deliberately does not produce is a single number that goes up. Anyone selling you an AI visibility score with two decimal places is selling you precision these systems do not have. The honest output is a direction and a list of the specific sources deciding the answer.

We run these panels monthly, including on our own brands. That is not a case study and we will not dress it up as one — we have no signed client work to point at yet. It is simply the reason we know what the run-to-run variance looks like.

Side by side

Classic search and AI answers, compared.

You need both. They are not substitutes, and the second does not replace the first on any timeline anyone can honestly name.

A comparison of classic search optimization and AI search visibility across seven dimensions.
	Classic search (SEO)	AI answers (GEO)
What you win	A position in a list of links	Your name inside the written answer
Where the shopper decides	On your page, after a click	In the assistant, often before any click
Primary levers	Crawlability, speed, page structure, depth, links	Parseable product data, extractable answers, third-party citations, entity clarity
How it is measured	Rank, impressions, clicks — daily, from Search Console	Named / not named across a fixed panel of buyer questions — monthly, by asking the engines
How fast it moves	Weeks to months, fairly legible cause and effect	Weeks to months, and noisier — the same query answers differently between runs
Failure mode	You are on page two	You are not in the conversation
Who it favors	Domains with authority and technical hygiene	Brands that independent sources have actually written about

If you are choosing which to fund first, we wrote that decision up honestly in SEO vs GEO: what your Shopify store needs first. The short version is that the answer depends on whether you currently have traffic, and for a lot of stores the correct answer is the boring one.

Questions buyers ask

Frequently asked.

Is AI search visibility the same as SEO?

No, though they overlap. SEO decides whether your page is ranked in a list of links. AI search visibility decides whether your brand is named inside an answer an assistant writes. The technical hygiene that helps one helps the other, but the deciding factor in an AI answer is usually off-site: whether independent sources the engine trusts have written about you. You can rank first on Google and go entirely unnamed in ChatGPT.

Does my store need this if my traffic is fine?

Traffic being fine today is a measurement of the buyers who still search the old way. Adobe Analytics reported that traffic referred to U.S. retail sites by generative AI sources rose 693% year over year across the November–December 2025 holiday period, and that revenue per visit from those referrals was up 254%. That is a channel changing shape while the existing one still works, which is the ordinary way a channel changes shape.

How long does it take to change?

Weeks to months, and it does not move in a straight line. On-site work — making product data parseable, writing the answer a category question actually needs — can register within weeks. Third-party citations take longer, because you are waiting on other people to publish and on the engines to re-crawl what they published. Anyone quoting you a date is quoting you a guess.

Should I publish an llms.txt file?

It will not hurt, and current evidence says it will not help. Google’s John Mueller has stated the file is not used for Search. Ahrefs analyzed 137,210 domains and found that 97% of the llms.txt files it looked at received zero requests in May 2026 — nothing fetched them at all. Treat it as optional housekeeping, not as the thing standing between you and an AI answer.

Can I do this myself?

Yes, and you should do the first part yourself regardless — asking the assistants your own category questions costs nothing and takes an afternoon. What is hard to sustain in-house is the repetition: the same panel, the same wording, every month, logged consistently enough that the trend means something. Most teams run it twice and stop. We wrote a longer honest comparison of the three ways to do this in the guide on agency versus in-house versus software.

What does Deadpan Labs actually do here?

We measure it and we run the work that changes it, for Shopify and DTC stores. We do not sell a scorecard and leave. We run our own products on the same platform we sell, and we run the same monthly panels on ourselves — one of our own brands went from zero to more than 900 downloads organically, with no ad spend, which is the receipt we can show you honestly. We have no client claims to make, because we are not going to invent any.

Keep reading

The rest of it.

The free check

Want to see where you stand?

Tell us your category. We’ll run your brand against ChatGPT, Perplexity and Google AI and show you whether you’re named — and who is named instead.

See who gets named insteadFree check · run by us, delivered to your inboxTalk to us →