What ChatGPT searches for before it answers

Published 23 October 2026 · 4 min read

Before it recommends a shop, ChatGPT runs its own short web searches. A shop that never comes back in them is rarely recommended. How to check yours.

When a buyer asks ChatGPT which hiking boots to buy, it often doesn't answer from memory alone. It writes a few short web searches of its own, reads what comes back, and builds the answer from those pages. Most buyers never look at those searches. They read the answer.

That step decides a lot. A study by Ahrefs of 1.4 million prompts found that 88% of the pages ChatGPT cited came through its searches. A shop that never comes back in those results is rarely recommended, however good its boots.

What those searches look like

The buyer asks a long question. ChatGPT turns it into shorter searches, closer to what a person would type into Google. A question like “which waterproof hiking boots are good for wide feet, under €150?” might become:

• “best waterproof hiking boots wide fit”
• “wide fit hiking boots review”
• “waterproof trekking boots wide feet under 150”

That list is an illustration. Nobody outside OpenAI can see the searches the ChatGPT app runs. Monitoring records them through each assistant's API, which gives a model the same kind of web search: close to the app, not the same. One thing follows from the buyer's side anyway. Someone who doesn't know your shop won't name it, so the searches name the product and leave your shop out.

What comes back

The pages ChatGPT cites are, Ahrefs found, mostly lists, reviews and high-authority sites: rankings, comparison sites, big marketplaces.

Ahrefs also found that how closely a page's title matches ChatGPT's own search wording is an important factor in being cited. Cited pages' titles were closer to those searches than to the buyer's original question. And pages whose addresses are made of readable words were cited more often (90% against 81%).

Check it yourself

1. 1 Take one real buyer question and write the 2–4 short searches you'd expect from it: the product, the one condition that matters most, and words like “best” or “review”. Leave your shop's name out.
2. 2 Run each search in Google and in Bing, in a private window so your own history doesn't colour the results. It won't be the list ChatGPT got, but it shows which kinds of pages hold these searches.
3. 3 Write down the first ten results for each and sort them: rankings and guides, comparison sites, marketplaces like Allegro or Ceneo, other shops' category pages. Is your shop anywhere? Are you on the rankings and comparisons that keep coming back?
4. 4 Open your own category and product pages and read their titles: the text on the browser tab, which is also the blue headline in Google. Do they use the words from your searches? “Waterproof hiking boots for wide feet | Granite Trail” can match a search. “Granite Trail” alone can't.
5. 5 Look at the addresses. granitetrail.eu/hiking-boots/wide-fit reads as words, granitetrail.eu/product.php?id=4817 doesn't. If yours are the second kind, ask your developer or your platform's support whether readable addresses can be switched on, with redirects (an automatic forward from the old address to the new one) so existing links keep working.

What to change first

Start with titles. They are quick to change, and they are the part of a page that Ahrefs tied to being cited. Give each category a title that says what is in it, in the words from your searches, with the shop's name at the end. Do the same on product pages: the product, the detail buyers ask about, then the shop.

Titles only help a page that search engines already know. Much of what decides lies off your own site: being indexed by search engines, links from other sites, and a place on the rankings and comparison sites the searches return. If one ranking of wide-fit boots keeps coming back and you sell three of the boots on it, that is worth a polite email to its author.

None of this guarantees a recommendation. It puts your shop where ChatGPT looks before it answers.