Why does ChatGPT recommend your store one day and not the next?

Why the same question gets a different answer

AI assistants do not look up a stored list of recommended stores. Each time someone asks a question, the assistant builds a new answer from scratch, and there is a bit of randomness in how it words and orders things. Two runs of the same question can name the same brands in a different order, or swap one brand for another.

For product questions, the assistant also searches the web as it goes. Which pages it finds, and which parts of those pages it keeps, can differ from one run to the next. A store sitting on the edge of being included will drop in and out.

We have no published figure on how often answers change for product questions, so be wary of anyone who quotes one. What is clear is that they do change, and that store owners should plan around it.

What changes behind the scenes between runs

AI assistants split one question into many smaller searches and then combine the results. Ahrefs reported that ChatGPT's deep research mode ran 420 searches for a single prompt about buying a red phone case. With that many searches, small differences in what comes back can change the final list of brands.

The web itself also moves. Pages get updated, new roundups appear and old ones go stale. In Ahrefs' study of 17 million citations across 7 AI platforms, content that AI assistants cited was 25.7% fresher than content in normal Google results. A competitor refreshing their buying guide this week can push your store out of an answer next week.

The wording of the question matters too. "Best beard oil" and "best unscented beard oil for sensitive skin" are different questions, and they can produce different recommendations.

• The searches the assistant runs
• The pages it finds and keeps
• How recently those pages were updated
• The exact words the shopper typed

Each AI assistant gives a different answer

ChatGPT, Gemini, Perplexity and Google AI Overviews do not share one list of recommendations. They lean on different sources. In Ahrefs' study, only 7 of the top 50 most cited websites appeared on the lists for all three of Google AI Overviews, ChatGPT and Perplexity.

Ahrefs found Google AI Overviews lean on YouTube, Reddit and Quora, ChatGPT on publishers and news sites, and Perplexity on niche and regional sites. So being named by one assistant says little about the others. Checking only ChatGPT gives you part of the picture.

Why one answer proves almost nothing

If you ask ChatGPT once and it names your store, you may have caught a lucky run. If it does not, you may have caught an unlucky one. Either way, one result is too small a sample to act on.

This is the reason GEO asks each buying question twice on each assistant and reports averages rather than single answers. Two runs is not perfect, but it smooths out some of the luck. Averaging across many questions and several assistants smooths out much more.

The same logic applies to a quick check you do yourself. Ask several specific buying questions, more than once, and count how often you appear. A share like "named in 6 of 20 answers" is far more useful than "it named me today".

What a trend tells you that a single answer can't

A trend is how often you are named, tracked over weeks. If your share of answers climbs from rarely to often across a month or two, something real has changed: more sites mention you, your pages are clearer, or a roundup has added you. If it falls steadily, a competitor may have taken your place on the pages AI reads.

Week to week, expect wobble. A dip one week followed by a rise the next is usually just variation. Look for direction across several checks before deciding a change worked or failed. New mentions on other sites can take weeks to months to show up in answers.

Pair the trend with money where you can. Clicks and orders from AI assistants show whether being named turns into sales, though those figures always undercount, because some AI apps pass no referrer and some shoppers search your name on Google afterwards.

• One answer: a snapshot, mostly luck
• Several runs of one question: a rough read
• Many questions, several assistants, several weeks: a trend you can act on

What tends to move the trend, and what doesn't

You cannot pay to be in the answer itself. ChatGPT now shows ads, but OpenAI's help centre says in 2026 that "Ads do not influence ChatGPT's answers" and that advertisers "have no ability to shape, rank, or alter ChatGPT's responses." Ads sit below the answer and are labelled. Google is different: it allows ads above, below or within AI Overviews in Australia and other countries, but those are labelled ads, not the recommendation.

What tends to help is the slow work: credible sites mentioning your products, product pages that state clear facts near the top, pages kept up to date, and making sure the bots that power AI search are not blocked in your robots.txt. These raise your odds across many runs rather than in any single one.

Skip the shortcuts. Fake reviews, posing as a customer on Reddit, buying links and stuffing keywords go against honest practice and are not worth the risk to your store's reputation.

How to check your own store without fooling yourself

Pick 10 to 20 specific questions a shopper would ask right before buying, such as "best gift for a new gardener under $50". Ask each one on ChatGPT, Gemini and Perplexity, in a fresh chat, more than once. Write down which brands are named and whether you are one of them.

Repeat the same questions on a schedule, for example monthly, using the same wording. Compare your share of answers each time and note which competitors and which websites keep appearing. Then change one thing at a time and watch whether the share moves over the following weeks.

• Use the same questions every time
• Use fresh chats, not one long conversation
• Count a share of answers, not a yes or no
• Judge changes over weeks, not days