Three things decide whether an AI assistant recommends your shop: whether it can reach your pages, whether it can work out what you sell, and whether you have given it enough to be confident recommending you. Most shops fail at the first or the second without ever knowing, because nothing about it shows up in your analytics.

If you only read one section, read the first gate. It breaks most often and takes the least time to fix.

What actually happens when someone asks

A shopper types something like "best waterproof hiking boots for wide feet under €150" into ChatGPT instead of into Google. They get an answer: two or three shops, each with a sentence explaining why.

There is no page two, and no list of ten links to work through. If your shop is not in that answer, you are not in the running, and you will never see it happen, because the visit never takes place. Nobody arrives, so nothing appears in your analytics. The loss is invisible, which is what makes this different from ordinary search.

So the question is not where you rank. It is what the assistant found when it went looking, and whether that was enough.

Gate one: can AI reach your pages?

Before anything else can matter, the assistant has to be allowed to fetch your pages. Many shops block it by accident.

Your shop has a file at yourshop.com/robots.txt that tells automated visitors which ones are welcome. It was designed for search engines. It now also decides whether AI assistants can see you, and the rules in it were usually written before any of these assistants existed, or by a security plugin that added a blanket block to keep bots off your site.

The distinction nobody tells you about

The bots that read your shop do two different jobs, and the difference matters.

Retrieval crawlers fetch your page to answer a shopper's question right now. Block one of these and you are invisible the moment somebody asks.
Training crawlers gather text to teach a future version of a model. Block one of these and the cost is real but slow and indirect. It affects a model release rather than today's answer.

Most advice online treats these as one group and tells you to either allow AI bots or block them. Both instructions are wrong, because the consequences are completely different.

Here are the names worth knowing, and what each one actually is:

Name in robots.txt	What it is for	Kind
`OAI-SearchBot`	ChatGPT search	Retrieval
`Claude-SearchBot`	Claude's search results	Retrieval
`PerplexityBot`	Perplexity's search results	Retrieval
`Bingbot`	Microsoft Copilot, and Bing itself	Retrieval
`Googlebot`	Google Search, including AI Overviews	Retrieval
`GPTBot`	OpenAI model training	Training
`ClaudeBot`	Anthropic model training	Training
`CCBot`	Common Crawl, which feeds many	Training
`Google-Extended`	Permission to train Gemini	Not a crawler
`Applebot-Extended`	Permission to train Apple Intelligence	Not a crawler

Two rows carry a warning. Google-Extended and Applebot-Extended read like crawler names and are not. They are permission settings describing how data that another bot already fetched may be used, so neither one decides whether you can be reached.

A third group is missing from the table on purpose: the bots that fetch a page because a shopper has just asked a question, such as ChatGPT-User. OpenAI and Perplexity both say robots.txt may not apply to those, on the reasoning that a person requested the page, so blocking them may not be a control you have at all. Anthropic makes no such exception for its own. The full reference is here, with a source for every line.

GPTBot is the name everyone recognises, and it is a training crawler. Plenty of shops have carefully blocked GPTBot believing they were making a decision about whether ChatGPT recommends them, while leaving the retrieval crawlers untouched. More often they have blocked everything and lost both.

Bingbot does two jobs at once. It feeds Microsoft Copilot, and it is also how your shop appears in Bing at all. Blocking it is rarely what anyone means to do.

What to do about it

Open yourshop.com/robots.txt in a browser. It is a plain text file and you can read it yourself. Look for lines that say Disallow: / under a User-agent naming any crawler in the table above, or under User-agent: *, which means everyone.

If you find one, decide deliberately rather than by default. Allowing retrieval crawlers is how you become recommendable. Whether to allow training crawlers is a genuine choice with reasonable arguments on both sides, and a different question from the first one.

The rules in that file go wrong in a small number of predictable ways, including one where a block naming a single crawler silently cancels every rule you thought applied to everyone. Those are worth reading through before you change anything.

Gate two: can AI work out what you sell?

An assistant that reaches your product page still has to understand it. The reliable way it does that is structured data: the hidden labels in your page's code that say "this is a product, this is its price, this many are in stock", instead of leaving a machine to guess from the layout. The longer explanation is here, including how to read your own with two free tools.

You probably already have some. Shopify, WooCommerce and most other platforms add it automatically. The question is whether it is complete, and it very often is not, especially once a theme has been customised or a plugin has taken over part of the page.

These are the labels that carry weight for a product:

Label	What it tells the assistant
`name`	What the product is called
`description`	What it is
`price` and `priceCurrency`	What it costs, and in what currency
`availability`	Whether it can be bought right now
`image`	What it looks like
`brand`	Who makes it
`aggregateRating` or `review`	What buyers thought
`sku` or `gtin`	Which exact product this is

Price and availability go missing most often, and they are the two an assistant is least willing to work without. Recommending a product at the wrong price, or one that turns out to be sold out, is the mistake these systems are most careful to avoid. Faced with a page where it cannot confirm either, the safe move is to recommend a shop where it can.

Gate three: is there enough to judge you on?

Passing the first two gates makes you readable. The assistant still has to decide whether to name you, and it does that on what your page actually says.

Think about the question the shopper asked: waterproof, wide feet, under €150. A product page reading "Classic hiking boot. Durable and comfortable. Available in several sizes." satisfies none of those three requirements. The writing is fine. It just contains nothing that answers the question. Which details a question is usually made of is a subject of its own, and the most checkable one on this page.

The shops that get named are the ones whose pages contain the detail the question is made of. Which fitting the model suits. What the waterproof rating is. What the upper is made of. What it weighs. The fix is more answers rather than more words.

Reviews work the same way. They are the closest thing to independent evidence on a page that is otherwise entirely your own marketing, and an assistant weighs them accordingly.

Why competitors get named instead of you

Usually one of three reasons, in order of how common they are:

They can be read and you cannot. A crawler was blocked, or the structured data is incomplete. This is not a marketing problem, and it is often a ten-minute fix.
They answer the question and you answer a different one. Their page addresses the specific thing being asked; yours describes the product in general.
They give more to be confident about: reviews, clear availability, a proper specification, consistent information across the page.

Budget is almost never the reason. You cannot buy your way into these answers, which is good news for a small shop: one with well-described products does get named alongside much larger ones.

Where to start

In this order, because each step is worth less until the one before it is done.

Read your robots.txt and check whether retrieval crawlers are blocked. This is the cheapest possible fix and the most common problem.
Check the structured data on one product page, the product that matters most to your business. Confirm price and availability are both there.
Reread that product's description from the shopper's side. Does it answer the question someone would actually ask, or describe the product in the abstract?
Then ask. Put the question your buyer would ask to an AI assistant and see which shops come back. There is a way to do this that does not fool you, and most people get it wrong on the first try.

Questions people ask

Does this replace SEO?

No. SEO works on where you appear in a list of results. This works on whether you are mentioned in an answer where there is no list. The two overlap, since a page nobody can crawl fails at both, but the fixes and the measurements are different. Where exactly they overlap and where they part is worth knowing before you decide how much of this is new work. If you have run into "GEO" or "AEO" while reading around this, those words are covered here.

Which assistant should I care about most?

Today, ChatGPT, because it is where most shopping questions are being asked. The first gate is not assistant-specific though: a page that can be fetched and understood is a page any of them can use.

Will fixing this show up in my analytics?

Not directly, and this is worth understanding. Some assistants send a visitor with a normal referrer, some send one with nothing attached, and a shopper who reads a recommendation and searches for you afterwards arrives looking like ordinary traffic. Treat the assistant's own answer as the measurement rather than your analytics.

How long does it take to work?

Longer than a page edit and shorter than SEO. The crawler has to come back and fetch the changed page before anything can be different, which is days to weeks depending on the assistant and on your shop.

Can I just pay to be recommended?

Not into the recommendation itself. OpenAI says its ads run on separate systems from the chat model and that advertisers cannot shape, rank or alter ChatGPT's responses, so the shops named in the answer are not there because somebody paid.

You can buy a slot beside it. ChatGPT has carried ads since 2026, labelled as sponsored and visually separated from the answer, shown to people on the free and Go tiers. That is the same split you already know from search, where the results and the ads above them are sold separately. Everything on this page is about the unpaid half.

Where we come into it

Everything above can be done by hand, and for one product on one shop, doing it by hand is entirely reasonable. It stops being reasonable at forty products, because the checks are repetitive, the structured data is tedious to read, and none of it tells you which competitor got recommended in your place or why.

That is what Shop on Radar does. It runs these checks across the products you choose, puts your buyers' real questions to ChatGPT, records who was named instead of you and what they did differently, and gives you the fixes in the order that costs you the most money. We explain how to fix things. We do not touch your shop.

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

Getting your shop recommended by AI assistants