ChatGPT can put your company name on screen without suggesting that anyone hire you. It can cite your website to support a fact about the trade, or recommend your business while linking to someone else's page. Those are three different results.

Record them separately: Was the business named? Was a page connected to the answer? Was the business presented as a suitable choice for the customer's need? A flattering screenshot can hide a weak result.

One answer can contain three different results

Start with the words and links on the screen. Do not infer a recommendation from the mere presence of a company name, and do not assume that a citation endorses the business behind the page.

Mentioned

The business name appears. The answer recognizes or discusses the company. It may be positive, neutral or negative. A name in a directory-style list still counts only as a mention unless the wording goes further.

Cited

A visible link supports the answer. The linked page may belong to the business, a directory, a publisher or another source. Open the link and check which nearby statement it supports.

Recommended

The business is presented as a fit. The answer offers the company as a suitable option for the service, place and constraint in the customer's question. The stated reason still needs checking.

A source link can help another company

Suppose ChatGPT compares roofers for a slate repair. It could cite your service page for the point that slate work needs specialist experience, then recommend another roofer whose public record more clearly covers slate repairs in the customer's town. Your page was cited. Your business was not recommended.

The reverse can happen too. ChatGPT may recommend a business while the visible link goes to an association profile, news article or review site rather than the company's own website. The business appeared in that answer, but the owned site did not earn the visible citation.

That distinction changes the work. A citation without a recommendation calls for a closer look at customer fit and how the business is described. A recommendation supported by an inaccurate third-party page calls for a correction at the source carrying the error.

What four official help pages call a source

On August 21, 2026, I reviewed current official help or documentation pages for ChatGPT Search, Claude web search, Perplexity and Google's generative search features. I recorded how each product described visible source evidence and looked for any statement equating a source link with a business recommendation.

The products do not use one shared label. Their documentation also leaves room for different interfaces and answer types, which is why a cross-platform report should preserve the product and mode instead of merging every link into one citation total.

Use a strict rule for recommendations

In an AnswerPrism audit, I count a recommendation only when the answer presents the business as suitable for the need in the question. A bare name in a list remains a mention. This is a reporting rule for keeping results honest, not a claim about OpenAI's ranking system.

The fit judgment should be visible in the answer. Phrases such as ‘a good option for,’ ‘worth considering for’ or a direct answer to ‘who should I hire?’ can qualify when they refer to the stated service and location. A sentence that merely says the company exists does not.

  • Mention: record the exact sentence containing the business name
  • Citation: save the linked URL and mark it as owned or independent
  • Recommendation: copy the language that presents the business as a fit
  • Accuracy: check the reasons, service area, hours and qualifications separately

Read the combinations, not only the total

The combinations tell you more than a single visibility score. They show whether the gap sits in recognition, visible evidence, customer fit or factual accuracy.

  • Mentioned, cited and recommended: preserve the wording and verify every material reason before treating it as a strong observation
  • Mentioned and cited, but not recommended: inspect the stated reason, exclusion or missing fit evidence
  • Cited without a mention: the page is visibly connected to the answer, but the brand did not enter the customer's choice
  • Recommended without an owned citation: inspect the independent sources and check whether the website supports the same facts
  • Mentioned only: recognition is present, but the screenshot does not yet show evidence or a fit judgment

Save the result before changing anything

Keep the exact question, date, location wording, product or search mode, full answer and every visible source. Run the same question in a separate fresh conversation and keep both outcomes, including a less flattering second result.

Then place mention, citation, recommendation and accuracy in separate columns. If the first run recommends the business and the second only names it, the result is mixed. Do not rewrite a correct website to chase the better screenshot.

Fix the first problem a customer could feel. Wrong hours deserve action before a missing citation. A recommendation for a service the company does not offer deserves action before a clean mention count.

Sources

  1. OpenAI: Searching the web with ChatGPT
  2. Anthropic: Enable and use web search
  3. Perplexity: How Perplexity works
  4. Google: Optimizing for generative AI features on Search