Guide

AI visibility report template for clients

Updated · 11 min read · By the SearchCrow team

A monthly AI visibility report tells a client whether ChatGPT, Perplexity, Gemini, Claude and Google's AI search name their business when customers ask, how that compares with last month and with named competitors, and what you did about it. The core number is a visibility score: the share of tracked answers that named the client. Around it sit a by-assistant table, a by-question table, share of voice, the sources AI cited, the facts it got wrong, and the actions done and planned. This page gives you each section, how to measure it, and a copyable Markdown skeleton.

Key takeaways

  • Report the same fixed set of customer questions every month, so the score is comparable.
  • Define the visibility score plainly: answers naming the client divided by total answers.
  • Show the by-question table. It is where clients see why the score moved and what to do next.
  • Always include a caveats section. Google and OpenAI both say AI answers and placement vary.
  • Collecting the data by hand takes about an hour per client per month; a tool removes that hour, not the thinking.

What should a monthly AI visibility report contain?

Eleven sections, in the order a client reads them: the headline first, the evidence in the middle, the plan and the caveats at the end.

SectionWhat it answers
1. SummaryAre we winning or losing, and why?
2. Questions trackedWhat did we ask, and why those?
3. Visibility scoreWhat share of answers named us, vs last month?
4. By assistantWhich assistants name us and which don't?
5. By questionWhere are we named, and who is named instead?
6. Share of voiceHow do we compare with named competitors?
7. Sources citedWhich sites do the assistants cite, and are we on them?
8. Facts AI got wrongIs AI misdescribing us?
9. Actions done and effectWhat did we do, and did it work?
10. Actions for next monthWhat's next?
11. CaveatsHow much should we trust these numbers?

The rest of this page goes through each section with an illustrative example row. The example numbers are made up to show the format; they are not measurements.

How do I write the summary?

Three sentences: the score and its direction, the single biggest reason, and the one thing you want the client to notice or decide.

Write it after everything else; the by-question table usually tells you which rows moved, and that is your reason.

Example (illustrative): "Visibility rose from 28% to 36% this month, mostly because Perplexity now names you for the three emergency-callout questions after the Checkatrade listing went live. Gemini still names none of your pages. Next month we publish the comparison page and chase the two directories you are missing from."

Which questions should the report track, and why list them?

List every question tracked, grouped by intent, with a one-line reason for each group, so the client can see the score is built on questions their customers plausibly ask.

Phrase questions the way a customer would, not as keywords, and cover the intents that matter: urgent, comparison, local, price and trust. Keep the list fixed for at least a quarter; if you change a question, say so and do not compare its row with last month.

Example rows (illustrative questions, not search volumes):

  • "Who is a reliable emergency plumber in Leeds?" (urgent, local)
  • "Is it worth paying more for a Gas Safe registered plumber?" (trust)
  • "What does a boiler service cost in Leeds?" (price)

For help choosing questions, our guide on how AI assistants choose businesses explains how assistants rewrite a question into several searches, which is why plain customer phrasing works better than keyword lists.

How do I define and report the visibility score?

The visibility score is the share of answers that named the client: answers naming the business divided by all answers collected, shown as a percentage, this month next to last.

State the definition every month, because "visibility" means different things in different tools, and show the raw counts next to the percentage so the client can see how small the sample is.

MetricLast monthThis monthChange
Answers collected60 (20 questions × 3 assistants)60–
Answers naming the client1722+5
Visibility score28%37%+9 points

(Illustrative.)

If you run each question more than once, count every run as an answer and say how many. For a client-friendly explanation, point them to what an AI visibility score is.

How do I break the score down by assistant?

Show the same score split by assistant, because clients are usually invisible on one or two assistants and fine on the others, and the fixes differ.

An assistant that never names the client is often a crawler or indexing problem rather than a content problem. Our AI visibility checklist covers the robots.txt and host settings to check first.

AssistantAnswersNamed the clientScoreLast month
ChatGPT20945%35%
Perplexity201155%40%
Gemini20210%10%

(Illustrative.)

What goes in the by-question table?

One row per question: whether the client was named by each assistant, and which competitor was named when the client wasn't.

This is the most-read table in the report: it turns a percentage into "ChatGPT recommends Smith & Co for your boiler-service question". Record the first one or two competitors named when the client wasn't.

Question (illustrative)ChatGPTPerplexityGeminiNamed instead
Who is a reliable emergency plumber in Leeds?ClientClientSmith & CoSmith & Co (Gemini)
What does a boiler service cost in Leeds?Smith & CoClientSmith & CoSmith & Co
Is it worth paying more for a Gas Safe registered plumber?No business namedNo business namedNo business named–

(Illustrative.)

Rows where no business is named at all are worth keeping: they are questions the client could own with a page that answers them directly.

How do I report share of voice against competitors?

Share of voice is each business's share of all business mentions across the tracked answers, for the client and the competitors the client has named.

Count every business mention in every answer, then divide. The score asks "were we named"; share of voice asks "of everyone named, how often was it us". A client named in half of answers can still have a small share if each answer lists five businesses.

BusinessMentionsShare of voiceLast month
Client2231%24%
Smith & Co2535%38%
Northside Heating1420%21%
Others1014%17%

(Illustrative.)

Our guide to AI competitor analysis covers how to pick the competitors to name and what to do when AI recommends a competitor instead.

Which sources did the assistants cite, and why does that matter?

List the domains the assistants cited most across all answers, and mark which ones mention the client, because being on the pages AI reads is the most direct route into its answers.

Assistants mostly answer by searching the web and citing what they find. Google says its AI features rely on its Search index and can show "what's being said about products and services across the web, including in blogs, videos, and forum discussions" (Google Search Central, 2026). So the directories, review sites and local publications that keep getting cited are a target list.

Cited source (illustrative)Times citedClient present?Action
checkatrade.com14Yes (listed in October)Keep reviews coming
yell.com9NoClaim listing
leeds-live.co.uk5NoPitch a local story
smithandco.co.uk4n/a (competitor site)Note which pages are cited

(Illustrative.)

Our list of directories AI assistants cite in the UK is a starting point for the "claim listing" rows.

How do I report facts AI got wrong?

List each wrong or outdated fact in an answer, the assistant that said it, the correct fact and the fix, because a client named with the wrong price or hours can lose the enquiry anyway.

Agree a short "key facts" sheet with the client (hours, prices, areas covered, services, accreditations) and check each answer against it. OpenAI warns that ChatGPT's "search results and citations can be incomplete, outdated, or incorrect" (OpenAI Help Center, 2026), so assume some errors every month.

AssistantWhat it said (illustrative)Correct factLikely sourceFix
ChatGPT"Closed on Saturdays"Open Saturdays 8–1Old Yell listingUpdate listing; add hours to site in plain text
Perplexity"Call-out fee £120"£852024 blog postUpdate the post; put the price on the service page

(Illustrative.)

See what to do when ChatGPT has wrong information about your business for the fix routes by source.

How do I report the actions done this month and their effect?

List each action with the date it went live and the rows in the by-question or sources tables that changed afterwards, and be honest when nothing changed yet.

Crawling and indexing take time, so an action done late in the month usually shows nothing until the next report. Say so rather than claiming credit early.

Action (illustrative)Live onGap it addressedEffect seen so far
Checkatrade listing with 12 reviews3 OctMost-cited source, client absentPerplexity names client on 3 emergency questions
Boiler-service page with price24 OctPrice question, competitor namedNone yet; page indexed 29 Oct

(Illustrative.)

What goes in the actions for next month?

Three to five actions, each tied to a specific gap in this report, with an owner and what you expect to change.

Typical action types: get listed on a cited source, publish a comparison or FAQ page for a question nobody owns, fix a wrong fact at its source, fix a crawling problem for a low-scoring assistant, and ask for reviews on the sites being cited. Our guides on comparison pages AI quotes and FAQ pages AI assistants quote cover the content ones.

Example (illustrative):

  1. Claim and complete the Yell listing (agency, by 15 Nov) — target: client present on the second most-cited source.
  2. Publish "Boiler service cost in Leeds" FAQ page (agency draft, client approves) — target: client named on the price question.
  3. Check Gemini: confirm Googlebot can fetch the site and the Business Profile hours match the website (agency) — target: Gemini score above 10%.

What caveats should every report carry?

State that AI answers vary, that the sample is small, that location and account affect results, and that no tool sees inside the assistants' ranking.

The platforms say this themselves. Google: "AI Mode and AI Overviews may use different models and techniques, so the set of responses and links they show will vary", and AI Overviews "often don't trigger" (Google Search Central, 2026). OpenAI, on ChatGPT search: "Placement is not guaranteed" (OpenAI Help Center, 2026). Google also warns against tools that "claim to use 'internal' Google metrics", adding that "No third-party tool has access to our internal ranking or AI systems" (Google Search Central, 2026).

Practical caveats to include:

  • Variation. The same question asked twice can name different businesses. Report trends over months, not single moves.
  • Sample size. 60 answers is a small sample; a few answers are several percentage points.
  • Location and personalisation. ChatGPT "may use an approximate location based on your IP address to provide relevant local results" and, if memory is on, may use saved memories when rewriting a search (OpenAI Help Center, 2026). Collect from one location, in fresh conversations, with memory off.
  • Google's own data. The client's Search Console has a Generative AI performance report showing impressions: how many times links to the site were shown in a generative AI feature on Google Search (Google Search Console Help, 2026). Put that figure in the report next to your counts; it is the only first-party number available.

Copyable Markdown skeleton

Paste this into a document, a Notion page or an email. Replace everything in square brackets.

# AI visibility report: [Client] — [Month Year]

## 1. Summary
[Score and direction.] [Biggest reason.] [What we want you to notice or decide.]

## 2. Questions tracked ([N] questions, [N] assistants, [N] runs each)
Illustrative customer questions, agreed on [date]. Not search volumes.
- Urgent: [question], [question]
- Comparison: [question]
- Local: [question]
- Price: [question]
- Trust: [question]

## 3. Visibility score
Share of answers that named [Client].
| Metric | Last month | This month | Change |
|---|---|---|---|
| Answers collected | | | |
| Answers naming [Client] | | | |
| **Visibility score** | | | |

## 4. By assistant
| Assistant | Answers | Named [Client] | Score | Last month |
|---|---|---|---|---|
| ChatGPT | | | | |
| Perplexity | | | | |
| Gemini | | | | |
| Claude | | | | |
| Google AI Overviews / AI Mode | | | | |

## 5. By question
| Question | ChatGPT | Perplexity | Gemini | Named instead |
|---|---|---|---|---|
| | | | | |

## 6. Share of voice
| Business | Mentions | Share of voice | Last month |
|---|---|---|---|
| [Client] | | | |
| [Competitor 1] | | | |
| [Competitor 2] | | | |
| Others | | | |

## 7. Sources cited
| Source | Times cited | [Client] present? | Action |
|---|---|---|---|
| | | | |

## 8. Facts AI got wrong
| Assistant | What it said | Correct fact | Likely source | Fix |
|---|---|---|---|---|
| | | | | |

## 9. Actions done this month and effect
| Action | Live on | Gap it addressed | Effect seen so far |
|---|---|---|---|
| | | | |

## 10. Actions for next month
1. [Action] ([owner], by [date]) — target: [row that should change]
2. 
3. 

## 11. Caveats
- AI answers vary between runs; read the trend over several months.
- [N] answers is a small sample; a few answers move the score by several points.
- Collected from [location], in fresh conversations, memory off, on [dates].
- Google's Search Console Generative AI performance report: [impressions] this month.

How do I collect the data by hand?

Ask each assistant each tracked question in a fresh conversation, and record the answer in a spreadsheet with one row per question per assistant per run.

  1. Set up the sheet. Columns: date, assistant, question, run, client named (yes/no), businesses named in order, sources cited (domains), wrong facts.
  2. Fix the conditions. Same device and location, memory off, web search on.
  3. Ask the questions. Paste each into a new chat, open the sources list where the assistant offers one (ChatGPT has a Sources view for answers that used web search (OpenAI Help Center, 2026)), and fill in the row.
  4. Repeat the run on another day and count both runs as answers.
  5. Compute. Score = rows with client named ÷ all rows. Share of voice = each business's mentions ÷ all business mentions. Sources = count of each domain.
  6. Check the facts against the key facts sheet, then fill the skeleton, summary last.

For 20 questions across three assistants, two runs, expect about an hour per client per month to collect, plus half an hour to write up. With five assistants and more questions it grows in proportion, which is usually when agencies look for a tool. If you are setting up the service itself, see how to offer AI visibility as a service.

How SearchCrow helps

SearchCrow does the collection on a schedule, and the dashboard maps onto this template: visibility score, share of voice against named competitors, by-assistant and by-question tables showing who was named instead, top cited sources with the ones you are missing from, a "fix facts" check against the key facts you enter in Settings, and an action plan with guides and drafts. Paid plans send a weekly report email; the Agency plan (£199/month, 150 questions, 25 competitors, see pricing) covers ChatGPT, Gemini, Perplexity, Claude, Copilot, Google AI Overviews and Google AI Mode. Its report page saves as a PDF with your own name, logo and colour (white-label) for every client business. Run a free check on a client's site first. The template works without it.

FAQ

What should an AI visibility report include? A three-sentence summary, the list of customer questions you track, a visibility score (the share of answers that named the client) this month against last, a breakdown by assistant and by question, share of voice against named competitors, the sources AI cited and which the client is missing from, facts AI got wrong, the actions done and their effect, next month's actions, and caveats about variation.

How do you measure AI visibility? Ask each assistant a fixed set of questions customers would plausibly ask, record whether the client's business is named in each answer, and divide the number of answers naming the client by the total number of answers. Repeat with the same questions each month. Record which competitors were named and which websites were cited so you can report share of voice and source gaps as well.

How often should I send an AI visibility report to clients? Monthly is the practical cadence for most small-business clients. Answers vary from one run to the next, so weekly numbers on a small question set can swing without anything having changed. A month gives time for an action, such as a new comparison page or a directory listing, to be crawled and to show up in answers, which makes the actions section meaningful.

Why does the AI visibility score change when nothing has changed? Because the assistants' answers are not fixed. Google says the responses and links shown by AI Overviews and AI Mode will vary, and OpenAI says placement in ChatGPT search results is not guaranteed. Track more questions, run them more than once, and report the trend over several months rather than reacting to a single month's movement.

Can I collect AI visibility data by hand? Yes. Open each assistant, ask the tracked questions in fresh conversations, and record in a spreadsheet whether the client was named, who else was named and which sources were cited. For 20 questions across three assistants, with a second run to smooth out variation, expect about an hour per client per month, plus the time to write the report.

Sources

Free check

Does ChatGPT recommend your business?

Free, in under a minute. You’ll see whether AI assistants mention you, who they recommend instead, and your top 3 fixes.

We’ll read your website for your business details, so there’s less to type.

Free, no card needed. We ask ChatGPT and Perplexity 5 questions your customers might ask, and check your website. We use your email to send you your results and a few short follow-ups about them, which you can stop with one click; see our privacy notice.