Guide

Why doesn’t ChatGPT recommend my product?

Because the pages it builds the answer from do not mention you. That is the whole reason, and it is a different problem from the one most people go and fix. Here is the evidence, including the part that is uncomfortable for us.

The thing you are probably fixing is the wrong thing

When a product does not get named, the usual response is to go and work on the website: add schema markup, publish an llms.txt, write an FAQ block, tidy the headings. All reasonable. None of it answers the question, because an assistant answering “what is the best X” is not reading your site. It runs a search and writes from the pages that come back.

If those pages do not mention you, no amount of work on your own domain changes the answer. You are optimising a page the model never retrieves.

What we measured

On 26 July 2026 we took 3 software categories, put six real buying questions from each to ChatGPT, Gemini and Perplexity, and then did the part almost nobody does: we read every source the 18 answers were built from, 131 in total.

Three categories, three completely different gatekeepers. In AI SEO tools, a blog called onelittleweb.com appeared in the sources of all six answers, ahead of Ahrefs and Semrush. In email marketing software, emailvendorselection.com and zapier.com each fed all six. In help desk software, Reddit fed all six.

And the finding that should change a budget: across all 18 answers, G2 appeared exactly once and Capterra never. Review directories were 0.8% of the sources deciding these categories. Reddit, by comparison, appeared 12 times.

That is one measurement of three categories on one day, not a universal law, and we would rather you checked your own than took ours. The point is that the answer is knowable, and most teams have never looked.

The four reasons, in the order they actually bite

1. 1

   You are absent from the pages that get retrieved

   The dominant reason. The answer is assembled from a handful of third-party pages, and you are in none of them. Fixing this means getting into those specific pages: the comparison article, the category roundup, the Reddit thread, the integration directory that happens to rank.

2. 2

   You are there, but described as something else

   Your product is mentioned in a page the model reads, but framed as a different category, so it never surfaces for the question your buyers ask. This is the quiet one, because you look present and convert nothing.

3. 3

   The category has one gatekeeper and it is not you

   Some categories are decided by a single site. In one of ours, a blog most people in the industry have never heard of fed every single answer. Winning that category means being in that blog, and nothing else moves the number nearly as much.

4. 4

   The assistants disagree and you only checked one

   The three do not agree with each other. Checking only ChatGPT gives you roughly a third of the picture, and a product named by Perplexity and invisible to ChatGPT looks like a success or a failure depending purely on which one you opened.

What to do instead

In order, and none of it starts with your own website.

First, find out whether you are named at all, and by which of the three. Second, read the sources behind those answers and write down the domains that keep reappearing: those are your category’s gatekeepers, and there are usually fewer than five. Third, work out which of them you can realistically get onto, which is a mix of pitching, being genuinely useful in the places your buyers already are, and publishing the thing the gatekeeper would want to cite. Fourth, re-run the same questions on a schedule, because a single measurement is a snapshot and the whole point is whether the number moves.

That loop is the entire product we sell, so treat the recommendation with the suspicion it deserves. The first step of it is free and needs no account, which is the part we would rather you judged us on.

Questions people ask next

Why doesn't ChatGPT recommend my product?

Almost always because the pages it builds its answer from do not mention you. An assistant answering a buying question runs a web search and writes from what comes back, so being absent from those specific pages is the same as not existing. It is rarely about your own site: in the categories we measured, the sources deciding the answer were third-party pages, not vendor sites.

Is it because my site is not optimised for AI?

Usually not, and this is the most expensive misdiagnosis in the category. Adding schema markup, an llms.txt file or an FAQ block to your own site does not put you in a comparison article that a model is reading. Those things help once someone is already looking at you. Getting named is about the pages other people have written.

Do I need to be on G2 or Capterra?

On the evidence, far less than the budgets suggest. Across 18 buying answers built from 131 cited sources, G2 appeared once and Capterra never, which is 0.8% of the sources. Reddit appeared 12 times. That is one measurement of three categories, not a law, but it is worth checking your own category before renewing a directory listing on the assumption that AI reads it.

How long does it take to start showing up?

Nobody honest can give you a number, because it depends on whether the pages that decide your category are ones you can get onto at all. What you can do is measure it: run the same questions on a schedule and watch whether the answer changes. Anyone promising you a ranking date is selling you a number they invented.

Does more traffic help?

Not directly. An assistant is not reading your analytics. It is reading a set of pages retrieved for that one question. A page with modest traffic that happens to be the one the model retrieves is worth more than a popular page it never sees.

Measurements taken 26 July 2026 against ChatGPT, Gemini and Perplexity on one shared live web search per question. Assistants change their answers over time, so every figure here is a record of a moment. The full question sets and source lists are published in the linked category write-ups.