Why ChatGPT doesn't recommend you (and how to find out in 30 seconds).

A buyer asks ChatGPT for the best software in your category. It names three competitors. The answer is fluent, specific and entirely untroubled by your absence.

That does not necessarily mean the model has judged your product and rejected it. More often, it means your brand was not available to the model in the right frame, for that question, on that platform, at that moment.

The distinction matters. “ChatGPT does not recommend us” sounds like a verdict. In practice, it is a measurement problem with several possible causes.

iSeer classifies these causes rather than collapsing them into one visibility score. Across the 1,610 valid platform checks currently in the iSeer ledger (the latest complete report per scanned domain, measured as of 11 July 2026), the brand was mentioned in 46% of answers and absent from 54%.

The failure modes below are the qualitative labels iSeer uses to separate why a brand is absent or mis-described. They describe the different ways a brand can disappear from an answer.

You may be absent from the category.

The bluntest case is category absence. The model answers the buyer's question, names plausible products, and never mentions you.

This can happen even when your company ranks well in conventional search. Search visibility and answer inclusion are related, but they are not the same thing. A search engine can return your page because it matches a query. A language model must also infer that your brand belongs in the comparison set, understand how it differs, and decide that it deserves one of a small number of mentions.

Across the ledger, 84 of 163 scanned domains — 52% — were absent from more than half of the buyer-intent answers measured.

That number should not be read as a universal industry benchmark. It is a description of the reports in the iSeer ledger at the time of publication. The sample, prompts and categories determine what it means.

You may be weakly positioned.

A model may know that you exist without knowing what to do with you.

Your site can contain a large amount of accurate copy and still fail to provide a clean answer to basic category questions:

• What is the product?
• Who is it for?
• Which problem does it solve?
• What makes it different from the obvious alternatives?
• In which situations should someone choose it?

When those answers are scattered across vague landing-page language, feature grids and campaign pages, the model has to reconstruct the positioning. Competitors with clearer category language are easier to place.

This is not an argument for writing for machines instead of people. It is an argument for saying plainly what the product is.

You may be outframed.

Sometimes the model mentions your brand, but under the wrong interpretation.

A product built for regulated teams may be framed as a lightweight tool for freelancers. A specialist platform may be treated as a generic alternative. A company known for an older product may remain attached to that product long after its positioning has changed.

This is an outframing problem. The brand is visible, but the surrounding explanation works against it. Outframing can be more damaging than simple absence because it creates the appearance of recognition while directing the buyer towards the wrong conclusion.

The useful question is not merely “Were we named?” It is “What did the answer say we were?”

iSeer records the answer and classifies the failure mode so the evidence can be inspected rather than inferred from a single score.

The gap may be platform-specific.

ChatGPT and Claude do not reliably produce the same shortlist.

They use different models, retrieval systems, product surfaces and source mixtures. A brand can appear consistently on one platform and disappear on the other. Treating one platform as a proxy for all AI discovery hides that divergence.

That is why iSeer measures both ChatGPT and Claude. It does not convert one model's answer into a claim about “AI” in general.

The platform question is especially important when a team reacts to an anecdote. One employee runs one prompt in ChatGPT, sees the brand, and concludes that visibility is fine. Another runs a similar prompt in Claude and reaches the opposite conclusion. Both observations may be true. Neither is a sufficient measurement.

The companion field note, ChatGPT and Claude disagree about you more than you think, examines that problem directly.

One answer is not a measurement.

Language-model output varies. Wording, ordering and inclusion can change between runs even when the question is unchanged.

A single answer is weak evidence. It may reveal a problem, but it cannot establish the size or persistence of that problem.

iSeer samples weekly, records the underlying runs and reports uncertainty using statistical confidence rather than pretending that one response is a stable ranking. The methodology is described in How iSeer measures AI visibility. Terms such as AIRS are defined in the glossary.

This is the less exciting version of AI visibility: prompts, runs, receipts and a willingness to say when the evidence is inconclusive. It is also the version that can be acted upon.

What to check first.

Start with the questions a buyer would ask before knowing your brand name.

“Best project management software” is broad. “Best project management software for a 20-person architecture practice” is closer to a buying decision. “Alternative to X for teams that need Y” is closer still.

For each question, inspect four things:

1. Whether your brand appears.
2. Where it appears.
3. How it is described.
4. Which alternatives appear instead.

Then repeat the measurement. Do not turn one favourable answer into a success story or one unfavourable answer into a crisis.

The purpose is not to persuade a model to repeat your marketing copy. It is to discover whether the market position you intended is present in the answers buyers receive.

The 30-second version.

Enter your domain. iSeer builds a free report from buyer-intent questions, measures ChatGPT and Claude, and shows whether the problem is absence, weak positioning, outframing or a platform-specific gap.

The result is not a promise that the model will recommend you next week. It is a record of what was measured now, what appears to be wrong, and what to test next.