A monthly AI visibility report needs eight fixed sections: the score and why it moved, what changed, where the answers get their sources, how the AI tools describe the brand, where it sits against competitors, the opportunities, the top fixes, and what was done. The order is the point. The number and its reason come first, and the work comes last, so the reader can judge in two minutes whether the month went well and whether the money was spent well. This post is the template our own report is generated from, with a labeled sample of every section you can copy.

We used to publish a one-page, four-section version of this template. Once we added the source analysis and the analysis of how the answers describe a brand, and ran it on up to 50 questions, it stopped fitting on one page, so we retired that format. The rules below are the same ones; there are just more sections. This post belongs to the measurement part of our AEO/GEO basics series; the series overview has the reading order.

Who reads this report, and what do they need from it?

The reader is the person paying for the work: a founder, a marketing lead, or an agency owner who has to answer to a client. They are not opening a tool. They read the report in an email, before or after a meeting, and make one call: is this working? Every section exists to make that call faster.

Three rules follow from that:

  • The number opens the report. Citation share and its change from last month are the first thing on the first page, with the reason underneath. Someone who reads only the first screen still gets the answer.
  • Fixed sections force a decision. A dashboard lets you dump findings and leave the reader to interpret them. Eight fixed sections make you turn findings into what changed, why, and what to do next.
  • Same sections every month means you can compare them. This month’s report sits next to last month’s. That is the accountability model: a number that moves, or a number that gets explained. How the number is produced matters as much as the number itself; why proxy-based AI visibility numbers mislead covers what goes wrong.

What are the eight sections?

Sections 1–3 are computed from the month’s captured answers, 4–6 come from an analysis pass over those answers, 7 comes from the fixes list, and 8 is written by a person. Nothing in the first seven sections is typed by hand, which is what makes them reproducible from the raw captures.

#SectionWhat it answersWhere it comes from
1Score + month-over-month changeAre we cited more or less than last month, and why?Computed from the run
2What changedWhich questions and AI tools moved?Diff against last month
3Where the answers get their sourcesWhich sites do the AI tools cite instead of us?Computed from the run
4How the AI tools see the brandWhat do the answers say about us, and is it right?Analysis pass, checked by a person
5Position in the answersWhere do we sit against named competitors?Analysis pass, checked by a person
6Opportunities, rankedWhat could move the number?Analysis pass, checked by a person
7Top fixes, ranked (up to 5)What are we doing about it, and did last month’s fixes work?Fixes list
8What we didWhat actually shipped, and what is waiting on whom?Written by a person

1. Score + month-over-month change

One headline number: citation share, the share of tracked cells where the brand is cited with a link. A cell is one question on one AI tool (ChatGPT, Perplexity, Gemini, Claude); each cell is asked five times because the answers vary between runs. Next to it: mention share, absences, and gaps (captures that failed, kept in the count rather than dropped). Then a scorecard row for every question, and a Why: line. If the number moved, the reason. If it didn’t, the reason for that. “Mixed month” is not a reason.

2. What changed

The ten cells that moved most, each with the old and new status and a short excerpt of the new answer. Excerpts, not screenshots: an excerpt can be copied and searched, and it can’t be cropped to look better than it is.

3. Where the answers get their sources

The domains the AI tools cite in the captured answers: the brand’s own site and every competitor’s site, always listed in full, plus the top five third-party domains (review sites, forums, news, directories). Each with how often it is cited and on which AI tool. This is often the most useful section for planning, because an answer usually cites the pages it took its claims from. If a competitor is named on the strength of a G2 page, you can see that here. If the tools cited no links at all, the section says so.

4. How the AI tools see the brand

A short summary of how the answers describe the brand, the recurring themes with their tone, what is going well, what needs work, and the claims the answers repeat, with how many captures repeat each. This section reports what the AI says, not what the brand would like it to say. A wrong claim repeated in a dozen answers is a finding, and it usually becomes a fix. The entity clarity post covers how to correct one.

5. Position in the answers

A table by question type (comparison, category, alternatives, pain point, trust) showing where the brand appears and which competitors are named, then who the brand is losing ground to and where it is simply missing.

6. Opportunities, ranked

Concrete changes, biggest gap first, each naming the questions it targets. “Add tier pricing to the pricing page for the pricing questions” is an opportunity. “Improve your content” is not.

7. Top fixes, ranked

Up to five fixes, open ones first, ranked by how many cells on their target questions are still uncited. Each names its target questions and whether it is waiting for the next run or was verified in a past one. Five, not fifteen: a list of fifteen has no priority, and nobody ships fifteen fixes in a month.

8. What we did

The work done this month and the ship status of every active fix: shipped and verified, shipped incorrectly, not shipped, or blocked, with who it is waiting on. When unshipped fixes explain a flat number, this section says so in those words. If no work was logged, it says “No work was logged this month.” It never invents work. This is the section that separates a service from a tool, because a tool has no “what we did” (the cost comparison of tools and a managed service is here).

What does a filled-in report look like?

Below is every section in the shape our report renders it. This is SAMPLE data: illustrative rows, not live results for any real brand. Questions are shown in category form, and the sample uses six questions to stay readable; a real report has a row for every tracked question.

1. Score (SAMPLE)

Citation share: 33%, cited in 40 of 120 tracked cells. ▲ up 6.7 pts.

Mention share 15% · absent 61 · gaps 1 (failed captures stay in the count, shown as gaps).

QuestionChatGPTPerplexityGeminiClaudeChange
“best [category] for [use case]”Cited #2AbsentCited #4 (2/5)Mentioned, no link▲ improved
“[your product] vs [competitor]”AbsentCited #1Cited #2 (4/5)Cited #3 (4/5)— flat
“[competitor] alternatives”Cited #3 (2/5)Cited #2AbsentAbsent▼ declined
“[product] pricing”Cited #1Absent (1 gap)Cited #2 (3/5)Cited #1 (4/5)▲ improved
“is [category] worth it for [use case]”Mentioned, no link (2/5)AbsentAbsentMentioned, no link (1/5)— flat
“[category] security / compliance”AbsentCited #3 (1/5)AbsentAbsent— flat

Why: share moved up to 33%; largest movers: “[product] pricing” on Gemini (improved); “best [category] for [use case]” on ChatGPT (improved); “[competitor] alternatives” on Gemini (declined).

2. What changed (SAMPLE)

  • "[product] pricing" (Gemini): ▲ absent → cited 3/5. Excerpt: "…[Brand] publishes three tiers, starting at…"
  • “best [category] for [use case]” (ChatGPT): ▲ cited 3/5 → cited 5/5. Excerpt: "…for teams in [use case], [Brand] is often listed alongside [Competitor] for its [differentiator]…"
  • "[competitor] alternatives" (Gemini): ▼ cited 2/5 → absent. The competitor’s new comparison page is now cited in its place.

3. Where the answers get their sources (SAMPLE)

DomainKindCitationsCited cellsShareBy AI tool
yourbrand.comown4640/11921%ChatGPT 14, Perplexity 15, Gemini 9, Claude 8
competitor-a.comcompetitor5244/11924%ChatGPT 12, Perplexity 18, Gemini 13, Claude 9
g2.comthird-party3127/11914%ChatGPT 9, Perplexity 12, Gemini 10
reddit.comthird-party2219/11910%Perplexity 14, Gemini 8
(other: 14 third-party domains)third-party6731%

Counts cover the 119 completed answers only; share = the domain’s citations ÷ all 218 citations captured.

4. How the AI tools see the brand (SAMPLE)

The answers describe [Brand] as a capable option for [use case] with clear pricing, but rarely as a fit for larger teams.

  • ✓ Transparent pricing (from 9 captures).
  • · Usually listed second, after [Competitor] (from 14 captures).
  • ⚠ Described as lacking [feature] (from 6 captures). The feature shipped in [month]; the answers cite an older review.

Going well: pricing questions now cite the brand’s own pricing page. Needs work: the [feature] claim is out of date and repeated across three AI tools.

5. Position in the answers (SAMPLE)

Question typeBrandCompetitors named
ComparisonNamed in most answers, usually second[Competitor A], [Competitor B]
AlternativesNamed in under half[Competitor A], [Competitor C]
Trust / securityMissing[Competitor A]

Losing ground to: [Competitor A] on alternatives questions, where its comparison page is cited (5 captures). Gaps: no answer to a security or compliance question names the brand.

6. Opportunities, ranked (SAMPLE)

  1. Correct the [feature] claim at its source. Update the review profile the answers cite. Targets: “[product] vs [competitor]”, “[competitor] alternatives”.
  2. Publish a security and compliance page with named certifications. Targets: “[category] security / compliance”.

7. Top fixes, ranked (SAMPLE)

  1. Security and compliance page. Targets: “[category] security / compliance”, “is [category] worth it for [use case]”. Verify against the next monthly run.
  2. Alternatives page depth. Targets: “[competitor] alternatives”. Verify against the next monthly run.
  3. Pricing page tier detail. Targets: “[product] pricing”. Verified in run [month].

8. What we did (SAMPLE)

  • Pricing page tier detail: shipped, verified.
  • Review profile correction for [feature]: blocked, waiting on review-site support since [date].
  • Alternatives page: not shipped, waiting on your developer. This is why “[competitor] alternatives” is flat.

Copy the shape, not the content. The sections are the contract; the rows are yours.

Copy the blank template

The blank version below has the eight headings and the fields each one needs. Paste it into a doc and fill it from your own monthly run. Keep the headings and their order the same every month, so each report can be compared with the last one.

# AI visibility report: [Brand], [YYYY-MM]

## 1. Score + month-over-month change
Citation share: [X]% (cited in [n] of [total] cells). Change: [▲/▼/—] [pts] vs last month.
Mention share [X]% · absent [n] · gaps [n]
| Question | ChatGPT | Perplexity | Gemini | Claude | Change |
|---|---|---|---|---|---|
| "[question]" | [Cited #pos (n/5) / Mentioned / Absent / Gap] | … | … | … | [▲/▼/—] |
Why: [the reason it moved, or the reason it didn't]

## 2. What changed
- "[question]" ([AI tool]): [old status] → [new status]. Excerpt: "[new answer, quoted]"

## 3. Where the answers get their sources
| Domain | Kind (own / competitor / third-party) | Citations | Cited cells | Share | By AI tool |

## 4. How the AI tools see the brand
Summary: [one or two sentences]
Themes: [theme] ([tone], from [n] captures)
Going well: … Needs work: …
Claims repeated: "[claim]" ([n] captures), correct or wrong?

## 5. Position in the answers
| Question type | Brand | Competitors named |
Losing ground to: [competitor]: [why] · Gaps: [where the brand is missing]

## 6. Opportunities, ranked
1. [Change]. Targets: "[question]", "[question]"

## 7. Top fixes, ranked (up to 5)
1. [Fix]. Targets: "[question]". [Verify against next run / Verified in run …]

## 8. What we did
- [Fix or task]: [shipped, verified / shipped incorrectly / not shipped / blocked], waiting on [who]

Method: [questions] × [AI tools] × [repeats] = [cells]; capture dates [from → to]; raw captures retained.

What never goes in the report?

Four things: traffic promises, screenshots presented as proof, numbers nobody can trace, and lists without priorities. Each one makes a report look busier and makes it harder to trust.

  • Traffic promises. No “projected AI traffic”, no “on track to 10x”. AI referral traffic is small for most sites even while it grows (the published numbers are here), and promising it is how a report loses credibility. Traffic can appear as an observed row, never as a target.
  • Screenshots as proof. A dashboard screenshot or a cropped “we got cited!” chat verifies nothing. Use excerpts tied to a question, an AI tool and a date.
  • Untraceable numbers. Every outside statistic names its source and date. Every internal number reproduces from the question list and the raw captures. The method line at the bottom of our report states questions × AI tools × repeats, the models used and the capture dates.
  • Lists without priorities. “15 things we noticed” is a dashboard dump. Up to five fixes, ranked, or the section has failed.

The same rules run through the whole site: source and date on every number, sample data labeled as sample, and no promises the data doesn’t support. The report is where those rules become a deliverable. The mechanics, including the human review every run gets before a report goes out, are in the methodology. The numbers behind sections 1–3 are explained in the AI visibility metrics post.

Next step: if you would rather we run this for you every month (the measurement, the diffs, the fixes and the report), the monthly service page has the prices and what’s included. If you sell this report to your own clients, the white-label ops post covers how it runs with our name out.

Updated 2026-11-02: replaced the one-page, four-section template with the eight-section report we now send, added a sample of every section and a blank copy; moved from /posts/ai-visibility-report-template/.