What a GEO report shows, and what it doesn’t

GEO, or Generative Engine Optimization, reporting shows how AI tools answer the questions you chose, over time: where your brand appears and where competitors do. It doesn’t show every conversation your buyers have with AI, and it can’t prove a business outcome on its own. Read it with those limits in mind.
What does prompt tracking record?
A prompt is the question or instruction you type into an AI tool. According to Semrush, prompt tracking records the answers to a chosen set of questions over time. You see where your brand and your competitors show up, and how the answers shift.
The same question may not return the same answer on every run. Semrush notes that its tool checks tracked prompts daily, but that daily checks don’t widen coverage; they only show whether an answer is stable. A mention that keeps appearing counts as stronger evidence than a one-off result.
What can’t a report see?
Semrush lists three ways an AI visibility report can mislead. I suggest using them as a checklist when you receive a report.
- The tracked list may leave out important buyer questions.
- Answers to tracked questions can change from one run to the next.
- The AI platform itself may have changed.
List length tells you little on its own. Semrush says a dashboard tracking 500 or 1,000 prompts can look like broad coverage. The number means nothing, though, until you know how many platforms, markets and languages it includes. Checking 100 prompts on three platforms yields different evidence than checking 300 on one. Even thousands of prompts can keep repeating the same few buyer situations.
A list built in category language can also make a brand look stronger than it is. Buyers often describe their problem in plain words, not industry terms. In Semrush’s example, a brand turns up when the category name is searched. It doesn’t turn up when someone asks how to tell whether ChatGPT recommends it.
Why should branded and unbranded queries get separate rows?
A branded query names a brand; the person asking already knows you and is checking before they decide. An unbranded query describes a category, a problem or a use case. According to HubSpot, new buyers meet you in unbranded queries, or they never do. HubSpot also notes that Google added a branded-query filter to Search Console in November 2025, and argues that AI search needs the same split. I’m relying on the parts of HubSpot’s articles I could access, not the full text.
Give prompts that name only a competitor their own label. HubSpot’s point is that these are neither unbranded nor yours.
Without a competitor benchmark, you can’t tell whether a drop belongs to you or to the market. HubSpot names competitors, an industry baseline or the brand’s own history as reference points. It calls fixing the wrong problem the costliest mistake.

What happens when visibility is high but the pipeline is shrinking?
Bill Hunt describes a project in Search Engine Journal. The company appeared often in AI answers, yet its lead pipeline kept shrinking. Hunt wrote 38 detailed, criteria-based prompts and ran them across five AI surfaces. That produced 190 responses.
The company was mentioned in 98.4% of the responses and appeared as a top pick or shortlisted option in 93.2%. But when Hunt checked those same responses against the criteria in the prompts, 73.7% showed weak or missing evidence for important claims. In many cases, Hunt says, the capability existed; the information explaining it fell short.
This is one project at one company, so don’t generalize from it. Still, a report that gives only a mention rate leaves out the second half: whether the evidence is sufficient.
What should a sample GEO report include?
The template below is our suggestion, not a standard from the sources. In a report, you’d want to see these sections:
- Question list: grouped into families by buyer decision, written in both industry language and plain language.
- Separate rates for branded, unbranded and competitor-only questions.
- Competitor benchmark: is the drop yours, or the whole market’s?
- Repeated runs: one-off mentions flagged separately.
- AI referral traffic and branded search in GA4.
- Evidence row: is solid proof of your key claim, such as a test result or a customer review, publicly available somewhere?
- A “what this report can’t see” section: which platforms, languages and buyer questions are out of scope.
A simple way to start: ask ChatGPT, Gemini and Perplexity ten questions your customers actually ask, once a month, and log whether you’re mentioned. That number is a measurement baseline. It says nothing about how much of the buyer conversation you cover.
Visibility reporting is one of the deliverables in Voie’s SEO & GEO (AI-driven Search) work. If you’re unsure what your current report measures, send us your brief and we’ll read it together.
