Why ChatGPT doesn't recommend your SaaS (and what to fix this week)

A model can know your SaaS exists and still leave it out when a buyer asks an open question about the category. That is the gap worth diagnosing: recognition is not the same as recommendation.

We saw this distinction in a live buyer-question map for a wishlist app. The models named the product in 24 of 36 answers overall, but in 0 of the 12 questions that did not contain its name. The full example is in our post on query fan-out and buyer questions. It is why the audit reports branded and open questions separately.

Start with the questions, not a generic diagnosis

List the questions a buyer asks before choosing a product. Cover the category, the main use cases, comparisons, objections and the context in which the product has to work. Our definition of buyer-intent prompts explains the five shapes and why open prompts are the ones that measure recommendation.

Ask each question without putting your brand name into the prompt. Keep the answers and mark three states:

• Your brand is named
• A competitor is named instead
• No product is named clearly

That gives you a map of specific gaps. It does not turn one answer into a general claim about what every ChatGPT user sees.

What to publish this week

Choose one open question where a competitor appears or nobody is named, then put the answer on the page that owns the topic.

Comparison question. If the question asks who should choose your product instead of a competitor, add an honest comparison page. State who each option fits and the facts a buyer needs to decide.

Pricing or fit question. Put the answer on the pricing or product page. Say what is included, what is not and who the offer is for.

Implementation or use-case question. Add the question as an H2 inside the relevant product page and answer it directly below the heading. Do not move it to a separate FAQ hub just because it is phrased as a question.

That last recommendation comes from our own citation sample. Across 505 live buyer answers, the models cited 438 URLs. Of those, 400 pointed to a home page, 38 to product, pricing or service pages, and none to a FAQ or help-center path. The method and caveats are in FAQ page or H2 sections: what AI models actually cite.

Measure the same questions again

After the page is published and available to crawlers, repeat the same questions. New questions create a different test. The useful comparison is whether the cells you worked on change while the rest of the map remains visible as context. The full method for building and scoring that test suite by hand is in how to measure your AI visibility.

---

Want to build the map for your own brand? Run the free audit: it asks the models live and shows the open buyer questions where your brand appears, where a competitor does and where nobody does.