Free tool · no email required

Does ChatGPT recommend your product?

Not “what does it know about my brand” — that question always flatters you. This builds the category buying questions a shopper actually asks, where the model picks a shortlist and you are either on it or you are not.

How ChatGPT shopping questions work

When someone asks ChatGPT for a product recommendation, the model answers with a shortlist assembled from what it can read: product data, reviews, and what third parties have written. OpenAI has not published a ranking system. What a store controls is legibility — whether its price, specs, shipping and returns exist in a form a model can quote. This tool tests whether that is currently working.

What this does today. It generates the question set and a scoring sheet. It does not query the model for you — that needs a service behind the page, and it does not exist yet. You run the questions; the page tells you exactly how and what to record. Also worth stating plainly: OpenAI withdrew Instant Checkout on 4 March 2026, so buying inside ChatGPT is not a shipped capability. The position we build for is discover in AI, buy on site. Checked 2026-09-09.

Build your question set.

Prefilled with a worked example — an illustrative brand and category, not a client. Describe your product in a buyer’s words and the set rewrites itself.

What they would type, not what you call it internally. “Greens powder”, not “daily nutritional foundation”.

The need, not the demographic. “Someone who hates the taste of greens” beats “health-conscious millennials”.

24 questions 

## Category shortlist — win: The brand appears anywhere in the answer.

- What are the best greens powder right now?
- Best greens powder for someone who hates the taste of greens?
- I need a greens powder recommendation. What should I look at?

## Budget-bound — win: The brand appears and the price quoted is correct.

- Best greens powder under $50?
- What is the best value greens powder for the money?
- Is there a good greens powder that is not expensive?

## Attribute-constrained — win: The brand appears and the attribute is described accurately.

- Which greens powder are unflavoured?
- I need a greens powder that is unflavoured — what are my options?
- Best unflavoured greens powder available now?

## Suitability — win: The model answers with detail that is correct, not hedged or generic.

- Is Daily Greens Powder any good for someone who hates the taste of greens?
- Would you recommend Example Brand for someone who hates the taste of greens?
- What do people say about Daily Greens Powder?

## Head-to-head — win: The comparison is factually correct on price, format and claims.

- Example Brand vs AG1 — which should I buy?
- How does Daily Greens Powder compare to AG1?
- Is Example Brand worth it compared to cheaper options?

## Substitution — win: The brand appears in the alternatives list.

- What are the best alternatives to AG1?
- I stopped buying AG1. What else is good?
- Cheaper alternatives to AG1 that are still good quality?

## Logistics and risk — win: Shipping and returns are stated correctly, not “check the website”.

- Which greens powder ship to the UK?
- Does Example Brand offer free returns?
- How long does delivery take from Example Brand?

## Repeat purchase — win: The subscription is mentioned with the correct cadence or discount.

- Best greens powder subscription?
- Can I get greens powder on a repeat delivery?
- Which greens powder give a discount for subscribing?

The scoring sheet downloads as CSV — one row per question, with columns for whether you were named, in what position, which competitors appeared and what the model cited.

How to run the test properly

The method matters more than the tool. Run it badly and you will get a result that makes you feel fine and tells you nothing.

  1. 1

    Use a logged-out session

    Your own account carries memory, history and everything you have ever asked about your own brand. That is the single fastest way to produce a flattering and worthless result.

  2. 2

    Run the same set against more than one model

    ChatGPT, Gemini and Perplexity read different sources and answer differently. A gap in one and not the others is a source problem; a gap in all three is a data problem.

  3. 3

    Record four things per answer

    Named or not, position, which competitors appeared, and what was cited. The citations are the most useful column — they tell you which third-party sources the model trusts in your category, which is a working target list.

  4. 4

    Read families, not answers

    One absence is noise. Absent from every budget question but present in the category shortlist means your price is not readable. Absent from attribute questions means your specs are not declared. The family that fails names the fix.

The taxonomy, published

Eight families. A prompt set you cannot inspect is a prompt set you cannot trust, so here is every shape, what it tests, and what counts as a win.

The eight buying-question families, and the win condition for each
Family What it tests Why it decides the sale A win looks like
Category shortlist Whether you appear at all when nobody names a brand. This is the question with the most buying intent behind it and the one no store can influence with copy alone. If you are absent here, everything else is decoration. The brand appears anywhere in the answer.
Budget-bound Whether the model knows your price well enough to place you in a bracket. Roughly half of category questions carry a budget. A model with no machine-readable price cannot put you in the right one, so you fall out of both the cheap and the premium answer. The brand appears and the price quoted is correct.
Attribute-constrained Whether your specs are legible enough to match a requirement. This is where itemised specs pay. A model can only match “unflavoured” or “fragrance-free” if the attribute is declared somewhere it can read, not implied in marketing copy. The brand appears and the attribute is described accurately.
Suitability Whether the model will vouch for you against a specific need. The question a considered purchase actually ends on. It needs specs, reviews and returns together — a description alone will not carry it. The model answers with detail that is correct, not hedged or generic.
Head-to-head Whether you are described accurately when set against a named rival. The model has to have enough data on you to draw a contrast. Where it does not, it either omits you or invents a difference — and the invented one is usually unflattering. The comparison is factually correct on price, format and claims.
Substitution Whether you are in the consideration set of people leaving a competitor. The highest-intent question in the set: the buyer has decided to switch and is asking who to switch to. Being absent here is the most expensive absence. The brand appears in the alternatives list.
Logistics and risk Whether shipping and returns are readable as data. The qualifiers that close a purchase and the ones buried in a policy page no crawler associates with your product. Almost no Shopify store publishes either in machine-readable form. Shipping and returns are stated correctly, not “check the website”.
Repeat purchase Whether your subscription offer is visible at all. Repeat-purchase brands sell a habit, not a unit. If the model never mentions that a subscription exists, the highest-value version of your offer is invisible. The subscription is mentioned with the correct cadence or discount.

The families are our own, drawn from the shapes ecommerce buying questions take. They are not a published standard and no model vendor endorses them. What share of real shopping prompts each family represents is metric to confirm — we have not measured it and will not estimate it.

What each failure tells you to fix

  • Absent from the shortlist Nothing to work with. The model has no usable data about the product at all — start with the readiness checker before anything else.
  • Absent from budget questions only Price is not machine-readable, so you cannot be placed in a bracket. That is one property in the product markup.
  • Absent from attribute questions Specs are not declared. The attribute exists in your copy, where it cannot be matched against a requirement.
  • Absent from logistics questions Shipping and returns live in a policy page nothing associates with the product.
  • Named, but described wrongly The model is filling gaps from somewhere other than your product page. This is the most urgent result on the sheet.

What this test cannot tell you

It is a sample, not a rank. Model answers vary by session, account history, region and model version. Running the set once tells you what happened once. Running it monthly, logged out, in the same conditions, tells you whether something is moving.

It does not measure traffic. Being named in an answer is not a visit, and we have no way from the outside to know how often your category’s questions are actually asked. Anyone quoting you an AI-search volume figure is quoting something nobody publishes.

It cannot explain why. No vendor publishes how products are selected for an answer. What is observable is what the model could read, and that is what the fixes above address. Any account of the ranking mechanism — including ours — would be a guess.

It is not agentic checkout readiness. Instant Checkout was withdrawn on 4 March 2026. There is no purchase flow inside ChatGPT to be ready for. Discovery is the half that exists, and it is the half this measures.

Definitions

Category prompt
A buying question that names a need rather than a brand — the shape almost all real shopping questions take.
Brand-recall prompt
A question that names your brand. Reassuring, and close to worthless as a measure of visibility.
Answer set
The shortlist a model returns for a category question. Being in it is the thing being tested here.
Citation
A source the model links or refers to in its answer. The most useful column on the scoring sheet, because it names who it trusts in your category.
Discover in AI, buy on site
The position this work serves: shoppers find products through models and complete the purchase on the store. The half that exists today.

Questions about testing AI visibility

Does this tool ask ChatGPT for me?

Not yet, and the page says so rather than hiding it behind a spinner. Everything here runs in your browser, and a browser cannot query a model without an API key — putting ours in the page would mean anyone could spend our credits. So this generates the question set and the scoring sheet, and you run them. When we put a service behind it, the page will run the prompts and report the answers, and we will say clearly that it has changed.

Why not just ask ChatGPT what it knows about my brand?

Because it will tell you, and the answer means nothing. Ask a model about a named brand and it will almost always produce something — that is the tool working, not your visibility. Nobody types your name into a shopping question. They type “best greens powder for someone who hates the taste”, and the only thing that matters is whether you are in the shortlist that comes back. Every prompt on this page is that shape.

How does ChatGPT decide which products to name?

OpenAI has not published a ranking system, and anyone who tells you they know it is guessing. What is observable is what the model can read: your product data, your reviews, and what third parties say about you. That is also the only part you control, which is why the fixes at the bottom of this page are all data fixes rather than prompt tricks.

Can I buy placement in ChatGPT’s product results?

Not through us, and we would not advise trying. Treat anyone selling guaranteed placement in an AI answer as selling something that does not exist. What does exist is making your product legible enough to be picked accurately — specs, price, availability, shipping, returns, reviews — which is ordinary work with an unglamorous name.

What about Instant Checkout — can people buy inside ChatGPT?

OpenAI withdrew Instant Checkout on 4 March 2026. Any agency selling you “agentic checkout readiness” as a shipped capability is selling something that is not currently available. The position worth building for is the one that has not changed: people discover in AI and buy on your site. Everything on this page serves the discovery half.

I ran the same question twice and got different answers.

That is expected. Model answers vary by session, by account history, by region and by model version. This is a sample, not a rank. Run each family more than once, in a logged-out session, and read the pattern across a set rather than trusting any single answer.

We were named but the details were wrong.

That is the most actionable result on the sheet, and it is a data problem rather than a visibility one. A wrong price, a wrong format or an invented claim means the model is filling gaps from somewhere other than your product page. The readiness checker will tell you which fields are missing.

Related

Find out what you’re losing.

Before you commit to anything, we tell you exactly what you’re losing and what it costs to stop it. Two weeks. Fixed fee. Credited in full against any build you go ahead with.

Book your Revenue Recovery Audit →Or run the numbers yourself →
Fee
$1,500–$3,000, fixed
Duration
Two weeks
Credited
In full, against any build
You supply
Read access + one 45-minute call