AI shopping visibility: how to measure yours

Somewhere right now, a shopper is asking ChatGPT which of the things you sell they should buy. The answer comes back as a tidy carousel: product photos, prices, a one-line verdict. Either your store is in that carousel or a competitor's is — and no dashboard you own tells you which. That blind spot has a name, AI shopping visibility, and the uncomfortable part is that it can be measured. Most stores just never do it.

This guide covers what the term actually means, what the numbers tend to look like when a store measures for the first time, and a measurement routine you can run yourself this week with a spreadsheet and an hour. If what you need is to get your products into ChatGPT Shopping in the first place — feeds, catalogs, the three doors — that is a different guide. This one is about knowing where you stand.

What AI shopping visibility actually is

AI shopping visibility is the share of shopping-shaped AI answers in your category where your products appear. Not "do I show up if I ask about my own brand" — that question mostly measures whether the model has heard of you. The metric that pays your invoices is what happens when a buyer who has never heard of you asks the assistant for the best option in your category, and the assistant picks a handful of products to put in front of them.

Two properties of that definition do all the work. First, it is a share, not a rank. There is no ordered list of your category sitting inside ChatGPT waiting to be scraped. The assistant generates each answer fresh, and may name a different set of products for the same question tomorrow. Second, it is defined over many phrasings, not one keyword. Buyers ask "best trail running shoes for wide feet", "trail shoes that won't wreck my knees", "what should I buy for rocky terrain" — and the assistant internally rewrites each of those into its own queries. Your visibility is your presence across that whole cloud of questions, sampled, the way a poll samples voters.

OpenAI's own documentation is useful for calibrating what you are measuring. The Help Center states that product results are selected independently by ChatGPT — not ads, not influenced by partnerships — and that the ranking weighs availability, price, quality signals, and whether you are the maker or primary seller of the product, using structured metadata from first-party and third-party providers. In other words: the inputs are knowable, the selection is probabilistic, and nobody can sell you a guaranteed slot.

One question, three storefronts

Before measuring, know what surface you are looking at, because a store shows up in AI answers in three distinct places and they move independently. There is the category recommendation ("best X for Y" — the carousel), the product card (the assistant describing your specific product, with a price and details it assembled from data you may not control), and the informational answer that names products while explaining something. We broke down the three races — and why winning one gives you nothing in the others — in Ecommerce AI visibility: what actually gets measured.

For a shopping visibility measurement, the carousel is the surface that matters most, for a blunt reason: it appears exactly when the person is deciding what to buy. The other two feed it, but the carousel is where the sale changes hands.

What the numbers look like when a store finally measures

Aggregates from Inspeccia audits between June and August 2026, across businesses whose owners asked the same question you are asking now — where do I stand in AI answers:

• In roughly 1 in 4 audits (24%), ChatGPT actively recommends the audited business when asked about its category.
• In 71%, the model knows the business exists but does not put it in the recommendation — the "knows you, doesn't pick you" middle where most stores live.
• In 40% of audits, the business's own website earns zero citations in the AI answers about its category: everything the model says about that market comes from other people's pages.

The practical reading: absence is the default, not the exception. If you have never measured, the base rate says the carousel is probably showing someone else — and the interesting question becomes who, and what data the assistant used to pick them.

How to measure yours this week

You need a spreadsheet, an hour, and discipline about repetition. The protocol:

1. Write the prompts a buyer would write

Ten to fifteen prompts, in the buyer's words, not yours. Not your keyword list — nobody types "buy artisanal leather goods online" into ChatGPT. Think "gift for a father-in-law who has everything, under $100", "most durable laptop bag for daily commuting". Include two or three where you'd honestly expect a competitor to win; a prompt set you can only pass is a mirror, not a measurement.

2. Run them clean, and log four things

Fresh session, logged out where possible, so memory and past chats don't contaminate the sample. For each prompt record: did shopping results appear at all; are you in them; which stores or products appear instead; and whether the price, stock and details shown for you (or for the competitor) are actually right. That last column earns its place — a carousel that shows your product at an old price loses the sale as surely as absence, and it is the one failure you can usually fix fast. When you find wrong data, this guide covers where it comes from.

3. Repeat weekly. The trend is the number

One run is an anecdote: the same prompt can produce a different carousel an hour later. Run the same set weekly and compute the share of appearances. After four weeks you have something no one-off audit gives you — a baseline with variance, so when the number moves you can tell signal from noise. Add a couple of new prompts each month as you learn how buyers actually phrase things; keep the original set intact for comparability.

Why two tools will give you two different numbers

If you have already tried a couple of visibility tools and gotten numbers that disagree, nothing is broken. A visibility score depends on which prompts were run, how many times, how recently, and against which model — and every vendor chooses differently. A tool that runs a handful of prompts once a month and a tool that runs hundreds weekly are both "measuring AI visibility" the way a thermometer and a weather satellite are both measuring weather. The disagreement is the nature of the metric, and it is why the honest presentation is a share with variance, not a single proud number. We dissected this in Why AI visibility tools don't agree.

The corollary for your own measurement: pick one protocol and stay loyal to it. A consistent imperfect ruler beats a rotating cast of perfect ones.

What actually moves the number

Measurement without levers is just documented sadness, so, in the order we see them matter in audits:

• The data doors. Assistants assemble carousels from product data they can actually reach: crawlable product pages, complete Product schema, and — where available — a feed or catalog integration. Which doors exist and which one is automatic is the subject of the ChatGPT Shopping guide.
• Data hygiene. OpenAI lists availability and price among its merchant ranking factors. Price on your site that contradicts the price in third-party sources, stale stock, orphaned variants: each one is a reason for the model to pick a product it trusts more.
• Being the primary seller of your own brand. The ranking factors explicitly favor the maker or primary seller. If a marketplace outranks you for your own products in AI answers, that is a measurable, fixable leak.

Frequently asked questions