Four Numbers That Tell You Where Your Brand Stands in AI

Author : sopof smeth | Published On : 02 Nov 2026

A practical breakdown of the four signals worth recording in every AI visibility check, and why a blended score across platforms hides the findings that matter most.

Ask a marketing team how their brand is doing in AI, and the honest ones will admit they do not know. The checks happen, but they are informal: someone types a question into ChatGPT, screenshots the answer, and the screenshot goes into a deck. What is missing is not effort. It is a definition of what to count.

A structured AI visibility audit fixes this by recording four things for every answer, not one. Each of the four points to a different problem, and each has a different fix.

The first is the mention: did the assistant name the brand at all? This is the number everyone checks, and on its own it is close to meaningless. Being named is the entry ticket, not the result.

The second is the citation: did the answer link back to the brand’s own site, or to a source the brand controls? A mention without a citation sends the buyer somewhere else to verify, and that somewhere else may be a competitor’s comparison page.

The third is the description, and it is the one teams underestimate. Being named with the wrong location, the wrong price band or the wrong core service can cost a company more than not being named at all. Buyers act on descriptions, so a wrong one actively works against the brand.

The fourth is the source map: which outside pages did the assistant rely on to form its answer? This is the actionable number. AI answers are assembled from sources, and those sources can be identified, read and improved. If the assistant keeps citing a three-year-old directory listing with outdated details, that listing is the fix.

Recording four numbers only works if the questions stay the same. A panel of around twenty-five buyer-style questions, frozen at the start, is what turns scattered checks into a trend line. The moment questions are swapped for easier ones, every chart goes up and nothing has been learned.

Just as important is what not to do with the results: never average the platforms into one score. Published case data shows a brand can appear in hundreds of Google AI responses while scoring zero on Gemini. A blended average erases the zero, and the zero is usually the finding that matters. Buyers on the platform where the brand is invisible never encounter it at all.

Independent research points in the same direction. A Writesonic study of more than 160,000 prompts found that fewer than four percent of cited sources were shared across all four major engines for the same prompt. Visibility is platform-specific, so measurement has to be platform-specific too.

None of this requires new software. It requires a frozen question list, a spreadsheet, and the discipline to record four things on every run instead of one.