---
title: "The AI Visibility Report Teardown: Which Numbers Are Worth Keeping?"
description: "A page-by-page teardown of the typical AI visibility report: what its scores, share of voice, mentions, sentiment and content gaps reveal—and what they cannot prove."
url: "https://groas.com/post/teardown-the-typical-ai-visibility-repor"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ab6cb7f7-20f7-4a9d-8358-85c30f932096.png"
published: "2026-10-16T05:38:46.268Z"
modified: "2026-10-16T05:38:46.349Z"
---

October 16, 2026 · 10 min read

# The AI Visibility Report Teardown: Which Numbers Are Worth Keeping?

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![A shiny brass gauge reads in the green on a copper pipe that is leaking into a bucket: the metric looks great while the real problem drips on.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ab6cb7f7-20f7-4a9d-8358-85c30f932096.png)

In this article

1. [Page one: A visibility score with no repair instructions](#page-one-a-visibility-score-with-no-repair-instructions)
2. [Page two: A competitor pie chart that counts the wrong thing](#page-two-a-competitor-pie-chart-that-counts-the-wrong-thing)
3. [Pages three and four: Mentions that may lead nowhere](#pages-three-and-four-mentions-that-may-lead-nowhere)
4. [Page five: A green sentiment dial over a real objection](#page-five-a-green-sentiment-dial-over-a-real-objection)
5. [Page six: Content gaps that arrive as unassigned homework](#page-six-content-gaps-that-arrive-as-unassigned-homework)
6. [The missing page: What changed, and did the work cause it?](#the-missing-page-what-changed-and-did-the-work-cause-it)
7. [The five-item check before buying the dashboard](#the-five-item-check-before-buying-the-dashboard)
8. [Overall verdict: Keep the evidence, lose the theatre](#overall-verdict-keep-the-evidence-lose-the-theatre)

A visibility score jumps from 61 to 68, and Monday’s AI search report calls it progress. Progress toward what? I chose the standard six-page AI visibility report for this teardown because it repeats the vanity-metric mistake PPC agencies made in 2015: showing a client a number that moved without showing what the work achieved.

You know the artefact. Whether it comes from a standalone prompt tracker such as [Otterly](https://help.otterly.ai/kpis) or [Peec](https://docs.peec.ai/metrics/brand-metrics/share-of-voice), or from an enterprise suite, the familiar layout puts a headline score on page one, a competitor Share of Voice chart on page two, engine-by-engine mention counts on pages three and four, sentiment on page five, and a list of content gaps on page six. The figures in this sample report—a 68/100 score, 28% Share of Voice and +74% sentiment—are illustrations of that layout, not a finding about any particular company.

I remember the PPC version: a thirty-slide deck celebrating Quality Score, CTR and upper-funnel impressions while cost per acquisition stayed stubbornly high. The charts were real. The implied business result was not. As [Seer Interactive’s analysis of AI search tracking](https://www.seerinteractive.com/insights/ai-visibility-is-a-vanity-metric-prepare-your-execs) argues, tidy scorecards can create the same problem when they turn variable AI responses into apparent certainty. If you want to [measure brand visibility across ChatGPT, Gemini and Google AI Overviews](https://groas.com/post/run-your-own-ai-mention-audit-a-repeatab), you need to know what each number can tell you to change. Let’s turn the pages.

## Page one: A visibility score with no repair instructions

The first thing an executive sees is a dial reading 68 out of 100. Enterprise suites such as [Profound](https://docs.tryprofound.com/agents/nodes/data/visibility-score) use multi-engine prompt sampling; other trackers combine coverage and position into an index. The appeal is obvious. One number fits neatly into a Monday meeting.

The problem is what gets compressed to make that number. Was your brand named at all? Where did it appear in a list? How many prompts in the tracker’s library produced an answer? **A composite score hides the prompt-level result you need to act on.** The same 68 could describe strong coverage on five commercial prompts and no coverage on twenty others, or incidental mentions spread across all twenty-five. Those situations call for different work. The dial cannot distinguish them.

![A cutaway visibility-score dial with loose dice and gears behind it.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/175ce30c-a3f3-49ad-864c-121a40d28b64.png)

Then there is response volatility. In conventional search, a ranking change gives you a relatively concrete investigation: inspect the page, competitors and technical changes. AI responses can vary even when the prompt does not. [Research published by OptimizeGeo](https://www.optimizegeo.ai/blog/ai-response-volatility-monitoring) reports 10.56% within-model output divergence under identical prompt conditions, with wider swings for more complex queries. A move from 61 to 68 in one weekly PDF might reflect a real change, sampling noise or both. The score alone cannot tell you which.

**Operator verdict: Discard the headline score as a performance verdict.** Keep the underlying prompt results if the tool makes them available. When the dial falls four points, it cannot tell you which page to inspect, which citation you lost or whether your team’s work had any effect.

## Page two: A competitor pie chart that counts the wrong thing

Next comes the Share of Voice donut: your brand at 28%, Competitor A at 42%, two smaller players splitting the rest. It borrows the credibility of Impression Share in Google Ads. But [Peec’s Share of Voice definition](https://docs.peec.ai/metrics/brand-metrics/share-of-voice) divides your brand mentions by mentions of all tracked brands. That calculation measures mentions in the tracked responses, not the proportion of buyers who saw you.

Picture ten buyer queries. Your software appears in four answers, so it has appeared in 40% of those responses. A competitor with three product tiers could be named twelve times across the same answers. If yours is named four times, its share of those sixteen mentions is 25%. Nothing about your four appearances changed. The denominator did.

The prompt mix can distort the chart again. A question about legacy CRM options and a question about the best CRM for a 50-person B2B sales team may each count once, although one is much closer to a purchase decision. And as [MaxAEO’s discussion of AI Share of Voice](https://maxaeo.ai/blog/what-are-the-best-tools-for-tracking-share-of-voice-in-ai/) notes, a mention does not tell you whether the model recommended your product or used it as a cautionary example. A pie chart gives both the same slice.

**Operator verdict: Rebuild or ignore.** Use a fixed set of unbranded commercial questions that reflect your buyers. Count whether the brand appears in each answer, then inspect its role in the answer. If the report cannot show its prompt set and counting method, the competitor chart is decoration.

## Pages three and four: Mentions that may lead nowhere

The next spread breaks mentions out across ChatGPT, Google AI Overviews, Gemini and Perplexity. This is a useful place to look: an engine-by-engine view is more specific than a blended score. But **a mention is not automatically a visit**, and one response is not a durable position.

Suppose a tracker runs twenty prompts across four engines once a week. That gives you eighty responses for the run. It can show what happened in those responses; it cannot establish that every buyer saw the same answers throughout the week. The [prompt-volatility benchmarks](https://www.optimizegeo.ai/blog/ai-response-volatility-monitoring) make repeat runs important, particularly before you treat a small change as a trend. A screenshot of one completion is evidence of one completion. I would not call it a ranking.

The bars also need to distinguish what the engine returned. A clickable citation to your product page gives someone a route to your site. An unlinked mention in a list of twelve products does not offer that direct route. Both might matter for awareness, but they do not have the same traffic mechanism. Put them in one bar and the result becomes harder to use.

**Operator verdict: Keep the engine breakdown; split the outcomes.** Show repeated responses, the prompts behind them, clickable citations and unlinked name-drops separately. Then check referral visits before calling a mention a pipeline win.

![An AI report placing a positive sentiment badge over critical customer reviews.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/a836531e-ddfc-4d03-bea1-e62d529ce605.png)

## Page five: A green sentiment dial over a real objection

Page five offers a reassuring +74% Net Positive Sentiment gauge. It is usually presented as an early-warning system for reputation. But a polite AI response can contain a serious buying objection: “Brand X is a well-established solution, though some users note pricing friction.” A sentiment parser may register the pleasant opening while a sales team would want to know exactly what “pricing friction” means.

That objection may also originate outside your site. As [ReachLLM’s sentiment-triage discussion](https://www.reachllm.com/blog/ai-sentiment-triage-source-vs-product-problems) explains, responses can draw on third-party reviews and user discussions, as well as genuine disputes about pricing or contracts. If an answer says buyers object to an inflexible contract, rewriting a page title will not settle the underlying issue. The source and substance of the objection matter far more than the gauge’s colour.

**Operator verdict: Cut the sentiment dial.** Preserve the actual passages where a commercial objection appears, trace their sources when possible and send the issue to whoever can address it. A positive average is no comfort if the answer a serious buyer sees contains the reason they will not buy.

## Page six: Content gaps that arrive as unassigned homework

The final page is the supposed deliverable: twenty to fifty “uncaptured AI content gaps,” usually prompts where a competitor appeared and your domain did not. If you are looking for [a platform that shows content gaps preventing your business from being cited by AI](https://groas.com/post/ai-prompt-tracking-tools-ranked-by-one-q), that table looks promising. Then someone has to decide what to make, change or fix. A prompt list alone does not make that decision.

![A stack of unread dashboard binders marked as homework on an office desk.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/50478060-9668-4157-9896-deb8de6c5da9.png)

A row saying “Competitor B is cited for enterprise data orchestration” leaves out the useful questions. Which competitor URL appeared? What claim or data point did the answer use? Do you have a relevant page, and what does it say? [Backlinko’s prompt-tracking breakdown](https://backlinko.com/llm-prompt-tracking) discusses the value of tracing the material that appears in AI answers. Without that trail, the marketing team can mistake “publish something about this prompt” for a plan. That is how a spreadsheet becomes a queue of generic posts rather than a diagnosis.

**Operator verdict: Keep the gaps only when they point to evidence and an asset.** A useful row names the prompt, the cited source, the relevant claim and the page your team should inspect. If the report supplies only the prompt, it has identified a question, not the fix.

## The missing page: What changed, and did the work cause it?

By now the report has shown scores, competitors, mentions, sentiment and gaps. It still has not answered the operator’s question: **What did we change before the result moved?** There is usually no timeline placing page edits, content updates or technical work beside changes in citation patterns. Without one, a rise in mentions might follow your work, a competitor’s publication or ordinary variation in responses. The chart does not establish causation.

The other omission is citation lineage. A recommendation is much easier to act on when you know whether the answer cited your documentation, a third-party discussion or an old review. When the source is not visible, the report should say so rather than pretend it has identified the cause. You cannot protect a useful citation, or repair an unhelpful one, by staring at a blended score.

This is the PPC reporting mistake I keep seeing in new clothes. When I managed Google Ads accounts by hand, a slide about clicks and impression share was easy to produce; explaining what changed in the account and what happened to CPA took more work. The first slide might be useful context. It was not the result. The AI visibility report makes the same trade: observation is packaged as progress, while the change log and the commercial outcome sit outside the deck. Paying for a sharper picture of a leak does not fix the pipe.

## The five-item check before buying the dashboard

Before signing an annual contract, I would read this [open letter to the owner about to buy an AI visibility dashboard](https://groas.com/post/an-open-letter-to-the-owner-about-to-buy) and ask for a sample export. Then I would check it against five requirements. The first four make the report diagnosable; the last tests whether anyone can connect movement to the work done.

1. **Fixed, unbranded commercial buyer prompts:** Show the questions and buyer personas; run each prompt at least three times daily rather than treating one answer as stable.
2. **Per-prompt citation lineage:** Show the exact brand and competitor URLs when available, and distinguish a primary recommendation from a footnote.
3. Engine-by-engine outcomes: Separate clickable referral links from unlinked text mentions; do not combine them in one victory bar.
4. **Pipeline and revenue connection:** Link attributable visits and qualified leads to CRM outcomes instead of declaring every mention a commercial win.
5. **A logged timeline of work:** Put code deployments, content updates and schema edits beside dated citation movement; do not imply that timing alone proves causation.

The **logged timeline of work** is the item most often missing. Without it, a team can spend months paying for reported movement while remaining unable to tell whether its own changes helped or the model simply answered differently.

## Overall verdict: Keep the evidence, lose the theatre

I would keep the fixed buyer prompts, the engine-by-engine view, source URLs where they are available and content gaps tied to a page worth inspecting. Those are pieces an operator can use. I would drop the executive dial, the undifferentiated mention totals and the cheerful sentiment gauge. None tells a business what changed because of work it paid for.

That is also the distinction [groas](https://groas.com/) is built around. It is a fully autonomous growth engine for paid and organic search, not another PDF layer: specialized models carry out campaign, content, bidding, targeting, budget and optimization work continuously, while a named human strategist sets direction and owns accountability. The model is a flat monthly fee, no setup fee and month-to-month cancellation, rather than paying for a stack of recommendations someone else must execute. I would still ask to see the actions and the commercial outcomes. A report earns its place when it helps answer that question; the rest is dashboard furniture.

## Frequently Asked Questions

### Should I use the overall visibility score in an AI visibility report as a measure of progress?

No. A composite score like 68/100 compresses different situations that call for different work, and AI responses vary even when the prompt does not—one study reports 10.56% within-model output divergence under identical conditions. Keep the underlying prompt-level results instead and discard the headline dial as a performance verdict.

### What does Share of Voice actually measure in AI visibility reports?

It divides your brand mentions by mentions of all tracked brands, so it measures mentions in tracked responses rather than the proportion of buyers who saw you. The prompt mix and whether a mention is a recommendation or a cautionary example are not reflected, so the pie chart can give very different situations the same slice.

### Are engine-by-engine mention counts in AI reports worth keeping?

Yes, keep the engine breakdown, but split the outcomes. A clickable citation to your product page sends traffic to your site, while an unlinked mention in a list of twelve products does not, so they should not share one bar. Check referral visits before calling a mention a pipeline win.

### Can a positive sentiment score in an AI visibility report hide a serious problem?

Yes. A polite AI response can contain a serious buying objection, such as users noting pricing friction, which a sentiment parser may register as positive. The source of the objection, such as third-party reviews, and its substance matter far more than the gauge's colour.

### How should I use the content gap list from an AI visibility report?

Keep the gaps only when they point to evidence and an asset. A useful row names the prompt, the cited competitor source, the relevant claim and the page your team should inspect. A prompt list alone is a question, not a fix, and can lead to a queue of generic posts.

### Why can't an AI visibility report tell me whether my work caused the changes in my metrics?

These reports usually lack a logged timeline placing page edits, content updates and technical work beside changes in citation patterns, and they often omit citation lineage. Without a timeline, a rise in mentions could follow your work, a competitor's publication, or ordinary variation in responses—the chart does not establish causation.

### What should I check before buying an AI visibility dashboard?

Ask for a sample export and check five things: fixed, unbranded commercial buyer prompts run at least three times daily; per-prompt citation lineage distinguishing primary recommendations from footnotes; engine-by-engine outcomes separating clickable links from unlinked mentions; connection of visits and qualified leads to CRM outcomes; and a logged timeline of work beside dated citation movement.

## Pay For Results, Not For Hours

Businesses buy the outcome, agencies resell it, and groas answers for it either way.

[See If You Qualify](https://groas.typeform.com/to/xC1bQNUT)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/teardown-the-typical-ai-visibility-repor#webpage","url":"https://groas.com/post/teardown-the-typical-ai-visibility-repor","name":"The AI Visibility Report Teardown: Which Numbers Are Worth Keeping?","description":"A page-by-page teardown of the typical AI visibility report: what its scores, share of voice, mentions, sentiment and content gaps reveal—and what they cannot prove.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ab6cb7f7-20f7-4a9d-8358-85c30f932096.png"},"breadcrumb":{"@id":"https://groas.com/post/teardown-the-typical-ai-visibility-repor#breadcrumb"},"dateModified":"2026-10-16T05:38:46.349Z","mainEntity":{"@id":"https://groas.com/post/teardown-the-typical-ai-visibility-repor#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/teardown-the-typical-ai-visibility-repor#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"The AI Visibility Report Teardown: Which Numbers Are Worth Keeping?","item":"https://groas.com/post/teardown-the-typical-ai-visibility-repor"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/teardown-the-typical-ai-visibility-repor#article","url":"https://groas.com/post/teardown-the-typical-ai-visibility-repor","mainEntityOfPage":{"@id":"https://groas.com/post/teardown-the-typical-ai-visibility-repor#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"The AI Visibility Report Teardown: Which Numbers Are Worth Keeping?","description":"A page-by-page teardown of the typical AI visibility report: what its scores, share of voice, mentions, sentiment and content gaps reveal—and what they cannot prove.","datePublished":"2026-10-16T05:38:46.268Z","dateModified":"2026-10-16T05:38:46.349Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ab6cb7f7-20f7-4a9d-8358-85c30f932096.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"report, page, visibility, visibility report, mentions, prompt, teardown, share","wordCount":1996,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"Ex Goldman Sachs and Ex Stanford Computer Science","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/teardown-the-typical-ai-visibility-repor#faq","mainEntity":[{"@type":"Question","name":"Should I use the overall visibility score in an AI visibility report as a measure of progress?","acceptedAnswer":{"@type":"Answer","text":"No. A composite score like 68/100 compresses different situations that call for different work, and AI responses vary even when the prompt does not—one study reports 10.56% within-model output divergence under identical conditions. Keep the underlying prompt-level results instead and discard the headline dial as a performance verdict."}},{"@type":"Question","name":"What does Share of Voice actually measure in AI visibility reports?","acceptedAnswer":{"@type":"Answer","text":"It divides your brand mentions by mentions of all tracked brands, so it measures mentions in tracked responses rather than the proportion of buyers who saw you. The prompt mix and whether a mention is a recommendation or a cautionary example are not reflected, so the pie chart can give very different situations the same slice."}},{"@type":"Question","name":"Are engine-by-engine mention counts in AI reports worth keeping?","acceptedAnswer":{"@type":"Answer","text":"Yes, keep the engine breakdown, but split the outcomes. A clickable citation to your product page sends traffic to your site, while an unlinked mention in a list of twelve products does not, so they should not share one bar. Check referral visits before calling a mention a pipeline win."}},{"@type":"Question","name":"Can a positive sentiment score in an AI visibility report hide a serious problem?","acceptedAnswer":{"@type":"Answer","text":"Yes. A polite AI response can contain a serious buying objection, such as users noting pricing friction, which a sentiment parser may register as positive. The source of the objection, such as third-party reviews, and its substance matter far more than the gauge's colour."}},{"@type":"Question","name":"How should I use the content gap list from an AI visibility report?","acceptedAnswer":{"@type":"Answer","text":"Keep the gaps only when they point to evidence and an asset. A useful row names the prompt, the cited competitor source, the relevant claim and the page your team should inspect. A prompt list alone is a question, not a fix, and can lead to a queue of generic posts."}},{"@type":"Question","name":"Why can't an AI visibility report tell me whether my work caused the changes in my metrics?","acceptedAnswer":{"@type":"Answer","text":"These reports usually lack a logged timeline placing page edits, content updates and technical work beside changes in citation patterns, and they often omit citation lineage. Without a timeline, a rise in mentions could follow your work, a competitor's publication, or ordinary variation in responses—the chart does not establish causation."}},{"@type":"Question","name":"What should I check before buying an AI visibility dashboard?","acceptedAnswer":{"@type":"Answer","text":"Ask for a sample export and check five things: fixed, unbranded commercial buyer prompts run at least three times daily; per-prompt citation lineage distinguishing primary recommendations from footnotes; engine-by-engine outcomes separating clickable links from unlinked mentions; connection of visits and qualified leads to CRM outcomes; and a logged timeline of work beside dated citation movement."}}]}]}
```