---
title: "The AI Visibility Report Teardown: 11 Line Items, 3 Worth Keeping"
description: "Alexander Perelman dissects a typical monthly AI visibility report, from its headline score to its recommendations, and keeps the three measures that show who gets cited."
url: "https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/dcde722c-dc30-4985-bd1d-a5c554d0fa95.png"
published: "2026-10-17T05:35:16.838Z"
modified: "2026-10-17T05:35:16.913Z"
---

October 17, 2026 · 11 min read

# The AI Visibility Report Teardown: 11 Line Items, 3 Worth Keeping

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

![A glossy report sliced into eleven strips: eight flashy ones with gauges and arrows are discarded, three plain tables are pinned aside as the keepers.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/dcde722c-dc30-4985-bd1d-a5c554d0fa95.png)

In this article

1. [The report on the table: score first, work last](#the-report-on-the-table-score-first-work-last)
2. [The one-page report I would pay to read](#the-one-page-report-i-would-pay-to-read)
3. [Do the work yourself, or pay for the work](#do-the-work-yourself-or-pay-for-the-work)

The monthly AI visibility report has learned the oldest trick in the PPC agency book: put the biggest number at the top and the useful work at the bottom. I built a composite report for this teardown from the familiar format: a visibility score, mention count, green sentiment gauge, share-of-voice bars, hundreds of prompts, an engine split, trend arrows, and a page of recommendations. No single vendor owns this artefact. That is why it is worth taking apart.

My claim is flat: this is the 2015 PPC agency report reborn. Impressions became **mentions**, CTR became **share of voice**, and a sentiment dial fills the page where the Quality Score box used to sit. The [standard Brand Visibility Score formula](https://searchengineland.com/measure-brand-visibility-ai-search-464524) gives the game away: answers mentioning your brand divided by total answers, times 100. I ran PPC when reports could hide weak CPA behind big impression numbers. Here are the three measures I would defend in front of a skeptical CFO: **citation rate per engine on a fixed prompt set, which URL got cited, and who got cited instead**.

## The report on the table: score first, work last

The composite opens with your logo and the month. Then you encounter, in order:

1. Overall AI Visibility Score
2. Total brand mentions
3. Sentiment gauge
4. Share of voice vs competitors
5. Prompts tracked
6. Engine breakdown across ChatGPT, Gemini, and AI Overviews
7. Citation rate
8. Cited URLs
9. Competitors cited instead
10. Week-over-week trend arrows
11. Recommendations

**The order is part of the pitch.** The score gets page one; the evidence of what to fix comes later. I used to write PPC reports this way. Good news up top, homework at the bottom. So I will read this one as a client should have read mine: line by line, asking what each item claims, what produces it, and whether it changes a decision.

### 1. Overall AI Visibility Score: the average hides the question

The score claims to answer whether you appear in AI answers. Using the [standard formula](https://searchengineland.com/measure-brand-visibility-ai-search-464524), 22 mentions across 100 answers prints as 22%. But that one number can fold together branded prompts where you always appear, comparison prompts where you never do, runs on different engines, and whatever prompt list the vendor chose that month. Change the list and the score changes, even if buyer behavior does not.

**Verdict: cut it as the headline.** [Vendors use their own prompts, weighting, and run frequency](https://www.sona.com/blog/how-to-choose-an-aeo-checker-that-measures-more-than-mentions/), so a 42 in one dashboard is not a 42 in another. I used to lead PPC reports with impression share. I was wrong then. I took that habit apart in [_Your AI Visibility Score Is Impression Share With a New Haircut_](https://groas.com/post/your-ai-visibility-score-is-impression-s). Fix the prompts, engines, and locale before you read a trend; do not ask one blended score to explain it.

### 2. Total brand mentions: impressions with a new label

This line claims buyers are hearing about you. It counts answers containing your name, whether or not they link to you. That distinction matters: [Semrush data summarized by Meev puts the share of mentioned brands also cited as a source at 6% to 27%](https://meev.ai/articles/how-to-track-ai-brand-visibility-before-rivals-do). A mention can rise while your cited pages remain at zero. I watched the PPC version when impressions grew and qualified clicks did not.

**Verdict: demote mentions to a diagnostic.** They can tell you where your name appears. They cannot, alone, tell you which page a buyer can open. Never pay for growth in mentions alone.

### 3. Sentiment gauge: green is not a recommendation

The dial claims AI feels good about your brand. It scores the tone around mentions and prints a reassuring number, say 87% positive, in green. But a pleasant sentence is not a citation. [A study of 9 million AI answers found that brands with more negative sentiment were slightly more likely to be cited across the engines it examined; positive sentiment did not reliably raise citation likelihood](https://dev.to/alifar/study-of-9-million-ai-answers-challenges-the-link-between-sentiment-and-citations-16mg). Critical comparisons and problem threads can contain the specifics an answer cites.

**Verdict: delete the gauge from the monthly business report.** You can have a glowing sentiment score while a competitor takes the citation. Tone and visibility are different outcomes; the dial should not pretend otherwise.

### 4. Share of voice: inspect the denominator

This bar chart claims you own a slice of the conversation. A citation-based version divides [citations to your domain by total citations across the prompt set, times 100](https://www.mentionlytics.com/blog/ai-citation-tracking/). That denominator does the work. Add ten branded prompts where you already get cited and your share can jump without a change on the buying questions you care about.

Raw counts can mislead in the other direction. Your citations can rise from 40 to 50 while your share falls from 8% to 5% because total citations doubled from 500 to 1,000. The chart says your count grew; it does not say you won more of the conversation.

**Verdict: keep share of voice only as context**, with a fixed prompt set and a visible denominator. Without both, the bars are decoration.

### 5. Prompts tracked: a big list is not a buying list

This line claims coverage. A report tracking 500 prompts looks industrious, even if near-duplicates count similar answers again and informational questions crowd out buying questions. The size of the list tells me little until I can read it.

Prompt volume also has a cost. The [plan comparison cited here](https://www.sona.com/blog/how-to-choose-an-aeo-checker-that-measures-more-than-mentions/) lists Otterly Lite at $29 a month for 15 prompts, Peec Starter near $80 to $95 for 50 across three models, and Scrunch Core at $250 for 125. Running more prompts across engines takes resources; a large headline count does not tell you how often each was run. I used to pad PPC reports with keyword counts. I was wrong.

**Verdict: ask to see the list and the run frequency.** Thirty buying prompts you can defend beat 500 you cannot.

### 6. Engine breakdown: one run is not a pattern

Now the report splits results across ChatGPT, Gemini, and AI Overviews. Good: an aggregate score can conceal an engine where you never appear. But the split depends on sampling. [The same prompt can return different answers by location, model version, and hour](https://www.sona.com/blog/how-to-choose-an-aeo-checker-that-measures-more-than-mentions/). One monthly run per engine is a sample of one dressed as coverage.

The alternative is repeated measurement. One [tracking approach runs each prompt three times per surface](https://meev.ai/articles/how-to-track-ai-brand-visibility-before-rivals-do) across ChatGPT, Perplexity, Gemini, Claude, Grok, AI Overviews, and AI Mode. That does not eliminate variation. It lets you see it instead of mistaking one answer for the whole month.

**Verdict: keep the engine split, but repeat the fixed prompts.** If the report cannot tell you how often it ran them, do not treat a gap on Gemini as settled fact.

### 7. Citation rate: the first line I would defend

Here is the measure the opening score was trying to be. On a defined set of buying prompts, report how often your site gets cited **per engine**, using the same method each month. Hold the prompts and measurement rules still; then a change in rate has a meaning you can investigate.

There is a reason to measure citations rather than infer them from rankings. [An analysis of 4 million URLs found that 37.9% of URLs cited in AI Overviews also ranked in the top 10, down from around 76% in July 2025, while 94% of Overviews cited at least one top-20 URL](https://aeovision.ai/articles/google-ai-overviews-geo-statistics-2026/). Ranking remains useful context. It is not a record of which URL the answer sourced.

**Verdict: keep citation rate per engine on a fixed prompt set.** Use [_The 30-Prompt Test for AI Search Visibility_](https://groas.com/post/the-30-prompt-test-measure-your-brand-s) to build the set, run it consistently, and put this rate where the overall score used to be.

### 8. Cited URLs: the line that points to the work

This table lists the pages the engines actually linked. Unlike a mention count, it ties an answer to a URL a buyer can open. It also gives you a page to inspect: does it answer the buying question directly, or is the useful detail buried?

A cited-domain list can be sorted into [three useful buckets](https://www.mentionlytics.com/blog/ai-citation-tracking/):

- Pages that already recommend you
- Pages that mention you but favor rivals
- Frequently cited pages that never mention you

That third bucket shows a gap the score cannot. Your own uncited money pages deserve a look too. An edit is not a guaranteed citation, but at least the table tells you what to change and what to check next cycle.

**Verdict: keep cited URLs and move them to page one.** A page-level work ticket beats another dial.

### 9. Competitors cited instead: the uncomfortable useful line

This row names who got cited on the buying prompt where you did not. On the same question, the engine linked to their page instead of yours. Their page may answer more directly, offer fresher specifics, or sit on a domain the model reaches for. The report alone cannot prove which explanation won. It can show you the page to examine.

That is why I would read it before the sentiment gauge. In PPC, the search-term report showing where competitors ate my budget was the page that changed what I did Monday morning. A rival’s cited URL can serve the same purpose here.

**Verdict: keep competitors cited instead, prompt by prompt.** Name the rival, show the URL, and give the person doing the page work something concrete to beat.

### 10. Week-over-week arrows: weather sold as climate

Red and green arrows sit beside every metric: mentions up 12%, share down two points. But answers can shift between runs without a page edit. On a small prompt set, a weekly delta may measure sampling noise rather than a durable change.

The boring discipline is to establish a baseline: [at least four weeks of consistent daily tracking, then weekly and monthly averages](https://www.mentionlytics.com/blog/ai-citation-tracking/). Week-to-week fluctuation is normal. An arrow does not become an explanation because it is colored red.

**Verdict: strip the weekly arrows from the headline.** Keep the monthly comparison per engine, with the method unchanged, and investigate movement rather than celebrating a one-week bounce.

### 11. Recommendations: homework at the bottom of the invoice

The final page says to add an FAQ, build comparison content, or earn mentions on review sites. Those may be sensible tasks. In this report, though, they are often detached from the prompt you lost and the page that lost it. Worse, the client is handed the work.

I recognize this page. In 2015, I wrote it at the end of PPC reports when CPA had not moved: next steps for the client, which meant work I had not done. Monitoring identifies a problem. It does not become management because the tool prints a task list.

**Verdict: reject recommendations without ownership and execution.** A monthly fee should buy fixes in the account and on the page, not just a longer to-do list. That is the same vanity-reporting grievance behind [_The 7 Google Ads Metrics That Actually Matter in 2026_](https://groas.com/post/google-ads-metrics-that-matter-2026-vanity-metrics-agencies-hide).

## The one-page report I would pay to read

I would put **three measures** on one screen: citation rate per engine, the URL cited for each prompt, and the competitor cited instead when we lose. A prior-month comparison and a log of work done belong beside them as context, not as two more visibility scores. Use the same 30 buying prompts, locale, engines, and three-run method each cycle. State how those runs roll up into the rate.

For illustration, the page could read:

- **ChatGPT:** cited on 9 of 30 prompts, 30%, versus 26% last month; winning URL listed by prompt.
- **Gemini:** cited on 6 of 30 prompts, 20%, versus 23% last month; two competitor domains named on lost prompts.
- **AI Overviews:** cited on 11 of 30 prompts, about 37%, versus 34% last month; two money pages still uncited.
- **Work done:** pages rewritten, newly cited pages, and prompts where a competitor citation gave way to ours.

Those are illustrative entries, not a client result. The point is the reading order: see the rate, open the cited pages, inspect the losses, then check what work followed. The eleven-line version makes you dig through a score, a dial, and arrows to get there.

## Do the work yourself, or pay for the work

If you have four hours a month and an operator who likes spreadsheets, run the fixed set yourself: the same 30 buying prompts, three runs per engine, the same locale, logged consistently. The work is boring, and it stays boring. I stopped treating monitoring on its own as the prize. Tracking tells you which page lost. It does not rewrite that page.

If you pay a vendor, ask for the one-page view and an action log: pages edited, sections added, and prompts where citations changed. If the deliverable ends with recommendations for your team to carry out, you bought a to-do list.

That is where I land on groas, as someone who used to bill for the long report. groas runs tracking continuously across ChatGPT, Gemini, and AI Overviews, does the page work, and logs actions with reasoning under a named strategist. **My overall verdict on the standard report: cut the decoration and keep the evidence.** Keep citation rate per engine on fixed prompts, the URLs cited, and who got cited instead. If your current report cannot put those on one page and show the work behind them, replace the report.

## Frequently Asked Questions

### Why shouldn't I trust the overall AI Visibility Score as the headline number in my monthly report?

The score averages branded prompts where you always appear, comparison prompts where you may never appear, and different engines into one number. Vendors use their own prompt lists, weighting and run frequency, so a 42 in one dashboard is not a 42 in another. Fix the prompts, engines and locale before reading any trend.

### Does a rise in brand mentions mean buyers can actually find my pages in AI answers?

No. A mention counts answers containing your name whether or not they link to you, and data cited in the article puts the share of mentioned brands also cited as a source at only 6% to 27%. Treat mentions as a diagnostic of where your name appears, not as evidence a buyer can open your page.

### Is a positive sentiment score in my AI visibility report a sign I'm getting cited more?

No, tone and visibility are different outcomes. A study of 9 million AI answers found brands with more negative sentiment were slightly more likely to be cited, while positive sentiment did not reliably raise citation likelihood. You can have a glowing sentiment score while a competitor takes the citation.

### How can share of voice in an AI visibility report be misleading?

The denominator does the work: adding ten branded prompts where you already get cited can inflate your share without any change on buying questions. Counts can mislead in reverse too, since your citations can rise while your share falls because total citations grew. Keep it only as context with a fixed prompt set and a visible denominator.

### Is a report that tracks 500 prompts better than one that tracks 30?

Not necessarily. Large lists can contain near-duplicates and informational questions that crowd out buying questions, and a headline count does not tell you how often each prompt was actually run. Thirty buying prompts you can defend beat 500 you cannot, so ask to see the list and the run frequency.

### Why do AI answers differ between engine runs and how often should prompts be tested?

The same prompt can return different answers by location, model version and hour, so one monthly run per engine is a sample of one dressed as coverage. One tracking approach runs each prompt three times per surface across ChatGPT, Perplexity, Gemini, Claude, Grok, AI Overviews and AI Mode. Repeated runs do not eliminate variation but let you see it instead of mistaking one answer for the whole month.

### Which measures in an AI visibility report are actually worth keeping?

Keep citation rate per engine on a fixed prompt set of the same buying prompts run consistently, the URLs the engines actually cited for each prompt, and the competitor cited instead when you lose. A prior-month comparison and a log of work done belong beside them as context. Cut the headline score, sentiment dial and weekly arrows.

### Should I act on week-over-week trend arrows in my AI visibility report?

Be cautious. On a small prompt set, a weekly delta may measure sampling noise rather than a durable change, since answers can shift between runs without any page edit. The recommended discipline is at least four weeks of consistent daily tracking to establish a baseline, then weekly and monthly averages, and investigating movement rather than celebrating a one-week bounce.

## 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-of-a-typical-monthly-ai-visibil#webpage","url":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil","name":"The AI Visibility Report Teardown: 11 Line Items, 3 Worth Keeping","description":"Alexander Perelman dissects a typical monthly AI visibility report, from its headline score to its recommendations, and keeps the three measures that show who gets cited.","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/dcde722c-dc30-4985-bd1d-a5c554d0fa95.png"},"breadcrumb":{"@id":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil#breadcrumb"},"dateModified":"2026-10-17T05:35:16.913Z","mainEntity":{"@id":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil#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: 11 Line Items, 3 Worth Keeping","item":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil#article","url":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil","mainEntityOfPage":{"@id":"https://groas.com/post/teardown-of-a-typical-monthly-ai-visibil#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"The AI Visibility Report Teardown: 11 Line Items, 3 Worth Keeping","description":"Alexander Perelman dissects a typical monthly AI visibility report, from its headline score to its recommendations, and keeps the three measures that show who gets cited.","datePublished":"2026-10-17T05:35:16.838Z","dateModified":"2026-10-17T05:35:16.913Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/dcde722c-dc30-4985-bd1d-a5c554d0fa95.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"cited, report, score, visibility, page, prompts, visibility report, line","wordCount":2220,"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-of-a-typical-monthly-ai-visibil#faq","mainEntity":[{"@type":"Question","name":"Why shouldn't I trust the overall AI Visibility Score as the headline number in my monthly report?","acceptedAnswer":{"@type":"Answer","text":"The score averages branded prompts where you always appear, comparison prompts where you may never appear, and different engines into one number. Vendors use their own prompt lists, weighting and run frequency, so a 42 in one dashboard is not a 42 in another. Fix the prompts, engines and locale before reading any trend."}},{"@type":"Question","name":"Does a rise in brand mentions mean buyers can actually find my pages in AI answers?","acceptedAnswer":{"@type":"Answer","text":"No. A mention counts answers containing your name whether or not they link to you, and data cited in the article puts the share of mentioned brands also cited as a source at only 6% to 27%. Treat mentions as a diagnostic of where your name appears, not as evidence a buyer can open your page."}},{"@type":"Question","name":"Is a positive sentiment score in my AI visibility report a sign I'm getting cited more?","acceptedAnswer":{"@type":"Answer","text":"No, tone and visibility are different outcomes. A study of 9 million AI answers found brands with more negative sentiment were slightly more likely to be cited, while positive sentiment did not reliably raise citation likelihood. You can have a glowing sentiment score while a competitor takes the citation."}},{"@type":"Question","name":"How can share of voice in an AI visibility report be misleading?","acceptedAnswer":{"@type":"Answer","text":"The denominator does the work: adding ten branded prompts where you already get cited can inflate your share without any change on buying questions. Counts can mislead in reverse too, since your citations can rise while your share falls because total citations grew. Keep it only as context with a fixed prompt set and a visible denominator."}},{"@type":"Question","name":"Is a report that tracks 500 prompts better than one that tracks 30?","acceptedAnswer":{"@type":"Answer","text":"Not necessarily. Large lists can contain near-duplicates and informational questions that crowd out buying questions, and a headline count does not tell you how often each prompt was actually run. Thirty buying prompts you can defend beat 500 you cannot, so ask to see the list and the run frequency."}},{"@type":"Question","name":"Why do AI answers differ between engine runs and how often should prompts be tested?","acceptedAnswer":{"@type":"Answer","text":"The same prompt can return different answers by location, model version and hour, so one monthly run per engine is a sample of one dressed as coverage. One tracking approach runs each prompt three times per surface across ChatGPT, Perplexity, Gemini, Claude, Grok, AI Overviews and AI Mode. Repeated runs do not eliminate variation but let you see it instead of mistaking one answer for the whole month."}},{"@type":"Question","name":"Which measures in an AI visibility report are actually worth keeping?","acceptedAnswer":{"@type":"Answer","text":"Keep citation rate per engine on a fixed prompt set of the same buying prompts run consistently, the URLs the engines actually cited for each prompt, and the competitor cited instead when you lose. A prior-month comparison and a log of work done belong beside them as context. Cut the headline score, sentiment dial and weekly arrows."}},{"@type":"Question","name":"Should I act on week-over-week trend arrows in my AI visibility report?","acceptedAnswer":{"@type":"Answer","text":"Be cautious. On a small prompt set, a weekly delta may measure sampling noise rather than a durable change, since answers can shift between runs without any page edit. The recommended discipline is at least four weeks of consistent daily tracking to establish a baseline, then weekly and monthly averages, and investigating movement rather than celebrating a one-week bounce."}}]}]}
```