Why ChatGPT Doesn't Recommend Your AI Tool
You asked for the best tools in your own category and your product was not in the answer. That is a diagnosable problem with seven usual causes — and they need completely different fixes, so the first job is working out which one is yours.
First, Make Sure You Measured It Properly
Most makers who believe they have AI visibility measured it with a leading prompt. If your product name is anywhere in the question, the model will discuss your product — that is the model being cooperative, not the model recommending you. The only reading that means anything is the blind one:
✗ Tells you nothing:
"Is [your product] a good [category] tool?"
✓ The real test:
"What are the best [category] tools right now?"
Run the second one in a logged-out or temporary session, so you are measuring what a stranger sees rather than what your own chat history has taught the assistant about you.
The 7 Reasons, With a Test for Each
These are not alternatives — several can be true at once. Work down the list in order, because the cheap causes invalidate the expensive fixes while they are still live.
Nothing about you exists on a page an assistant would retrieve
Most commonWhat it looks like: The assistant names five competitors and you are not among them, in any phrasing of the question.
Why it happens: Assistants answer recommendation questions largely from retrievable third-party pages — category roundups, directories, comparison pages, forum threads. Your own marketing site is the least likely source to be pulled, because it is the one page with an obvious incentive to say you are the best.
You are listed, but your category label is not the one people ask in
CommonWhat it looks like: The assistant finds you when you name the product, but never when you describe the job.
Why it happens: Retrieval matches on the language of the question. If your listing says 'agentic workflow orchestration' and buyers ask for 'a tool that automates my invoices', there is no lexical or semantic bridge between the two, and you never enter the candidate set.
Your facts disagree across sources
Under-diagnosedWhat it looks like: The assistant mentions you but gets your pricing, your free tier, or your feature set wrong.
Why it happens: When the same product has three different prices across three sources, the model has no way to pick, and the safest behaviour is to hedge or to prefer a competitor whose facts are consistent. Stale directory entries are the usual culprit — many are never updated after submission.
You have no comparison surface
CommonWhat it looks like: You appear in broad 'best tools for X' answers, but never in 'A vs B' or 'alternatives to A' answers.
Why it happens: Comparative questions are answered from comparative pages. If no page anywhere puts your product next to a named competitor, there is nothing for the model to compare, so it compares the products that do have such pages.
Your site blocks the crawlers that feed the assistants
Under-diagnosedWhat it looks like: Nothing about you surfaces anywhere, including facts you have published clearly on your own site.
Why it happens: AI assistants use several distinct user agents for search-time retrieval and for training, and blanket robots.txt rules or aggressive bot protection often block the retrieval agents along with the scrapers. This is usually accidental, added by a hosting default or a security plugin.
Your page is a JavaScript shell
CommonWhat it looks like: Your site is indexed, but assistants describe it vaguely or only from your meta description.
Why it happens: If your product facts — pricing, features, limits — only exist after client-side rendering, a retrieval fetch may get an empty shell. The page technically exists and technically says nothing.
You are genuinely new and there is no corroboration yet
CommonWhat it looks like: Everything above checks out, and you still are not recommended.
Why it happens: Assistants are conservative about recommending products with a thin evidence trail, and reasonably so. A product with one listing and no independent mentions looks to a model exactly like a product that does not really exist yet.
Cause 1 is the one you can close this week.
A listing here gives your tool a permanent, structured page in the category that ChatGPT, Perplexity and Google read when someone asks which tool to use. Free listings are reviewed by hand and usually live within a day.
Running the Diagnostic Properly
Five rules. Break any one of them and the reading tells you something other than what you think it does.
1. Run the blind prompt, not the leading one
Ask 'what are the best tools for [your category]?' — never 'is [your product] good?'. Naming your product in the prompt guarantees it appears in the answer and tells you nothing. Almost every self-assessment of AI visibility is inflated by exactly this mistake.
2. Run it across all three assistants, not just one
ChatGPT, Claude and Perplexity retrieve differently and cite differently. Being absent from one and present in another is a useful signal about which sources each is leaning on; being absent from all three is a different, larger problem.
3. Run it in a fresh session with no memory or personalisation
Your own account has months of context about your product. Use a logged-out or temporary session, or the result is a measurement of your own chat history rather than of what a stranger sees.
4. Record the competitors that do appear, and find out where they are cited from
The answer usually names the pages you are missing from. If four competitors appear and all four are on the same three roundups, those three roundups are your work list.
5. Re-run the identical prompts on a fixed schedule
A single reading is noise — assistant answers vary between runs. The signal is the trend across the same prompts over weeks, which is why this is worth automating rather than doing by hand when you remember to.
Or Have It Run For You
Our free scan runs blind prompts against your category across the major assistants and reports whether your tool is named, which competitors are, and which of the causes above your site trips. No product name in the prompt, so the reading is the honest one.
Frequently Asked Questions
How do I check whether ChatGPT recommends my tool?
Ask it for the best tools in your category without naming your product, in a fresh session with no personalisation, and see whether you appear. The critical detail is the blind prompt: if you ask 'what do you think of [my product]?' the model will discuss your product no matter what, and you will conclude you have visibility you do not have. Repeat across ChatGPT, Claude and Perplexity, and repeat over time, because a single run is noise.
Does getting listed in a directory actually make an assistant recommend me?
It is necessary far more often than it is sufficient. A listing gives an assistant something retrievable to cite, and directories with clean category structure are a shape these systems handle well. But if your facts disagree across sources, or your own site blocks retrieval agents, or no page anywhere compares you to a named competitor, a single listing will not carry you into the answer on its own. Treat it as the floor.
Is AI search optimization just SEO with a new name?
It overlaps heavily and diverges in two specific ways. The overlap: both reward being indexed, being on credible third-party pages, and being clear. The divergence: search returns ten links and lets the user choose, while an assistant returns one synthesised answer naming two to five products — so the gap between being on the list and being off it is far more binary. And assistants synthesise across sources, which makes consistency of your facts matter in a way ranking never did.
How long does it take to start showing up?
The config-level causes — blocked crawlers, JavaScript-only pricing — can change what an assistant sees within days of the next crawl. The corroboration-level causes take considerably longer, because you are waiting on third-party pages to be published, indexed and then retrieved. Expect fast movement on the technical fixes and a slow curve on the rest, and measure on a schedule so you can tell the two apart.
What should I fix first?
Work in the order of cheapness. Check for blocked retrieval agents first, because it is a fifteen-minute config change that invalidates every other effort while it is live. Then check whether your product facts are present in server-rendered HTML. Then fix factual inconsistency across existing listings. Only after those three is it worth investing in new coverage — otherwise you are adding sources to a product that still cannot be read correctly.
Related Guides
Affiliate disclosure: Some links on this page are affiliate links. If you sign up through them, AISO Tools may earn a commission at no extra cost to you. This never affects our rankings or reviews.
📬 Get the best new AI tools delivered weekly
One concise email with fresh launches, trending picks, and featured standouts.
Join thousands of professionals who discover the best AI tools every week. No spam — unsubscribe anytime.