You can find out whether an AI assistant can reach your shop, read it, and name it, in about an afternoon, with free tools and no technical knowledge. This page is the method. It ends with an honest account of what the method cannot tell you, because that part matters as much as the checks.

The four checks follow the three gates in getting your shop recommended by AI assistants, plus one that asks whether it is already happening.

First, how to ask without fooling yourself

Most people run this test wrong on the first try, in the same four ways. The answer they get is reassuring and means nothing.

Ask the buyer's question, not your shop's name. "Do you know Kowalski Outdoor?" tests whether an assistant can recall a brand. That is a different question from whether it would recommend you, and it will often say yes to the first while never naming you in the second. Ask what your customer would ask: "best waterproof hiking boots for wide feet under €150".

Use an unpersonalized temporary chat, or log out. ChatGPT's memory draws on your saved memories and your past conversations to tailor answers. Your own account is the worst possible place to test this, because it has spent months learning that you run this shop.

A temporary chat alone is not enough to fix that, which is the detail worth getting right. When you start one, ChatGPT asks whether you want it personalized or unpersonalized, and the personalized option can still use your existing memories and custom instructions. Only the unpersonalized one ignores them. Choose it before your first message, because OpenAI's documentation says you cannot change the setting once the conversation has started. Neither option writes anything back to memory, so your test will not teach ChatGPT anything new about you either way.

Ask more than once. These systems do not return the same answer twice. Run each question three or four times, in separate chats, and record all of it. One run tells you almost nothing. Four runs start to show a pattern, which is the thing you are actually looking for.

Remember where you are. Results vary by country and language. If you sell in Poland, ask in Polish, because that is the question your buyer types.

Check one: ask the question your buyer would ask

Write down three to five questions a real customer might ask. Not categories. Questions, with the constraints people actually carry: a budget, a use, a problem to solve, a size or a fit.

For each one, in a fresh temporary chat, ask it and write down two things: was your shop named, and which shops were named instead. Repeat until you have run each question three or four times.

What you are looking for is not a yes or no. It is which of these you are in:

Named consistently. You appear in most runs. This gate is working.
Named occasionally. You appear sometimes. You are a borderline candidate, which usually means the page is readable but thin on whatever the question asked about.
Never named, competitors are. The interesting case, and the one worth the rest of this page.
Nobody is named, including competitors. The assistant answered generally rather than by recommending shops. Your question was probably too broad, or it did not go and look. Make the question more specific and try again.

If the assistant shows its sources, read them. When it searches the web it usually lists what it used, and that list tells you which shops were reachable and readable for that question.

Check two: can they reach you at all

If you were never named, this is the first thing to rule out, because it is the most common cause and the cheapest to fix.

Open yourshop.com/robots.txt in a browser and read it. You are looking for Disallow: / under User-agent: * or under any named crawler. The mistakes in that file are predictable, including one where a block naming a single crawler silently cancels the rules you thought applied to everyone.

For a second opinion that is not your own reading, Google Search Console is free once you verify that you own the site. Its URL Inspection tool takes one of your product URLs and tells you whether Google was allowed to crawl it, saying "Blocked by robots.txt" in as many words if it was not. That only speaks for Googlebot, but Googlebot is also what governs whether you appear in Google's AI Overviews, so it is worth knowing on its own.

Check three: can they understand your product page

Take the product that matters most to your business. Run its URL through the Schema Markup Validator, which lists every label it finds, and through Google's Rich Results Test, which says which result types the page qualifies for and what is missing.

You are checking four things in particular: that a Product block exists at all, that it has a price, that the price has a currency beside it, and that the stock status is there. Those four are what an assistant needs to be confident enough to name you. The longer explanation is here.

Then read the page as your buyer would. If the question was about wide feet and waterproofing under €150, does the page say anything about fitting, about the waterproof rating, about the price? A page can be perfectly readable and still answer none of what was asked.

Check four: find out whether it is already happening

Two places show you real evidence rather than a test you constructed.

Bing Webmaster Tools is free, and its AI Performance report is currently in public preview. It shows how often your pages are cited in Microsoft Copilot and in Bing's AI-generated answers, which URLs were referenced, and the queries the system used when it went looking. That last part is unusually useful: it is a list of the questions your content actually got pulled into, written by the machine rather than guessed by you.

Your server's access logs record every visitor, including the user agent each one announced. Searching them for OAI-SearchBot, PerplexityBot, Claude-SearchBot and Bingbot tells you which AI crawlers have actually fetched your pages and when. Which of those names matters and why is a subject of its own. Most hosting panels expose logs somewhere; if yours does not, whoever maintains your shop can get them.

Keep a record, not a score

Write it in whatever you already use. The point is only that next month's run has something to compare against.

Question asked	Named in	Named instead
waterproof boots, €150	1 of 4 runs	two shops, both times
boots for wide feet	0 of 4 runs	three shops

Resist turning this into a number. A score out of a hundred invented from four chats would feel like progress and would mean nothing, because the inputs move on their own. What the table is for is noticing that a question you never appeared in starts returning you after you fixed a currency field.

What this method cannot tell you

This is the part most articles leave out, and leaving it out is what makes the method look more conclusive than it is.

Three or four runs is a small sample of something that varies. You are looking at a handful of answers from a system that does not repeat itself. A pattern across four runs is a hint. It is not a measurement, and treating it as one will have you rewriting a product page because of noise.

It covers one product at a time. Everything above is per product and per question. A shop with forty products and three real buyer questions each is a few hundred chats, run four times over, which is not an afternoon any more.

It tells you that you lost, not why. When three competitors are named and you are not, you can see the outcome. Working out what those three pages have that yours does not means opening each one, reading its markup, comparing its description against the question, and doing that for every question where you lost.

It only covers what you thought to ask. You are testing your own guesses about what customers type. The questions you did not think of are the ones where you are probably losing quietly.

It tests one assistant at a time, and the answer from one is not the answer from another.

None of that makes the method worthless. It is genuinely the right first move, and for one shop with a handful of products it may be all you ever need. It stops being the right move at the point where the checking costs more attention than the fixing.

Questions people ask

How long does the whole thing take?

An afternoon for one product and a few questions, if your robots.txt is readable and you can get into Search Console. The asking is quick. Reading the results honestly takes longer than people expect.

Should I use ChatGPT, or something else?

Start with ChatGPT, because it is where most shopping questions are being asked. If you have the appetite, run the same questions through Perplexity and Copilot, which will often name different shops and tell you something by the difference.

The assistant said my shop is great. Am I fine?

Probably not, and this is the trap. If you asked about your own shop by name, you tested whether it could find a page about you, which it can. Go back and ask the buyer's question with no brand name in it.

It named a competitor that is objectively worse than me. Why?

Usually because they are easier to read rather than better. A complete product page with a price, stock status and a specification beats a better product described vaguely. That is frustrating and it is also good news, because it is fixable.

Do I need to pay for anything to do this?

No. Every tool named on this page is free: temporary chats, the schema validators, Google Search Console and Bing Webmaster Tools. The cost is your time.

Where we come into it

Run the checks above. Genuinely, run them, before you consider paying anyone including us. If you sell twelve products and your customers ask three kinds of question, this method may be the whole answer, and we would rather you found that out than bought something you did not need.

Where it stops working is where the arithmetic turns. Forty products, several questions each, four runs apiece to see past the variation, every month to know whether a change helped. That is the point at which doing it by hand stops being thoroughness and starts being a second job, and the point at which the questions you did not think to ask become the expensive ones.

Shop on Radar is the same method with the repetition taken off your hands: the checks across the products you pick, the questions asked enough times to see past the variation, the competitor comparison done for you, and the findings ordered by what they cost. We explain how to fix what we find. We do not touch your shop.

We are not open yet. Join the waitlist and we will tell you when we are.

How to check by hand whether AI can see your shop