AI visibility report: a template you can copy, section by section

What an AI visibility report should contain

An AI visibility report should tell a reader how often AI assistants name you when buyers ask, who they name instead, and what they say, with every figure traceable to the answers behind it. Everything else in the report supports those three points.

Think of it like a school report. A single grade on the front is easy to read, but it hides which subjects went well and which did not. A good report gives the grade, then a line per subject, then the teacher's notes, then what the grade does not cover. This template does the same for AI answers.

The rules for making the numbers honest are covered in our guide on reporting AI visibility honestly. This post is the template that puts those rules on a page. It assumes you already have the answers, either from a tool or from tracking by hand as our guide on tracking brand mentions in AI search describes.

The AI visibility report template, section by section

The template has seven sections, in the order a busy reader needs them. Each row below says what goes in the section and gives a made-up example line.

Sections one and two: the scope and the headline

Open with the scope, so that every figure after it has a frame. Then give one headline figure, non-branded visibility with both counts, beside the same figure for the last period.

The scope line says how many questions were asked, which assistants answered, over what dates, and how often. It also says whether the questions changed. If they did, the headline is a new baseline, not a trend, and the report should say so in words.

The headline should be visibility over non-branded questions, meaning the questions a buyer asks without using your name. Visibility here means the answers that named you over the answers asked. Branded questions go in their own line, because an assistant asked about you by name names you nearly every time.

Resist a single composite score. A score built from weighted signals cannot be recomputed by the reader, and it hides the denominator. If a tool gives you one, put the counts behind it in the report as well.

If a tool supplies the figures, name it and quote its definition in the scope line, because tools count differently. Peec AI, for example, defines visibility as the percentage of AI responses where your brand appears (Peec AI docs). Semrush says the Brand Performance data in its AI Visibility Toolkit is updated weekly (Semrush knowledge base). A reader comparing your report with another team's needs to know both.

Write every figure as a count over a count. Named in 14 of 480 answers can be checked. A bare 2.9% cannot.

Section three: results per assistant

Break the headline out by assistant, because a healthy total can hide an assistant that never names you. Each assistant searches different sources, so each one can need a different fix.

Write the change in answers, not only in percentage points. Up 1 answer out of 120 sounds as small as it is. Up 0.8 points sounds like a trend. Our engine pages explain how each assistant finds its sources, which helps when one row is far below the rest.

Sections four and five: who was named instead, and what was said

Section four lists the brands the assistants named instead of you, each with its count over the same answers. Section five lists what the answers claimed about you, favourable and unfavourable, each with the sentence it came from.

For rivals, include brands you did not expect. A tool that only searches for names on a list you gave it cannot find a new rival, so read a sample of the answers yourself each month and add any brand that keeps appearing.

For claims, quote the sentence rather than a score. A line such as two answers called our pricing unclear, with both sentences pasted below it, gives the reader something to act on. Our guide to AI brand sentiment analysis covers how to collect these.

Include the claims made about your main rivals as well, in the same format. If three answers praise Brand A's onboarding and none mention yours, that gap is often the most useful line in the report. It tells the reader what buyers are hearing in your place, in the assistants' own words.

Keep both sections short. Five rivals and the five most common claims are usually enough. Link to the full list rather than pasting it, so the report stays readable and the detail is still there for anyone who wants to check it.

Section six: AI traffic, as a floor

Include traffic from AI assistants as a separate section, and label it as a floor. Analytics can only count the visits that arrive labelled, and some never are.

Google Analytics 4 now has an AI Assistant channel, which Google defines as visits from sources like ChatGPT, Gemini, DeepSeek, Copilot or Grok, excluding Google's AI Overviews and AI Mode (Google Analytics help). Google also says AI Overviews and AI Mode are included in the overall search traffic in Search Console (Google Search Central), so those clicks cannot be split out. Our guide to tracking AI traffic in GA4 covers the setup.

Keep traffic and visibility in separate sections. They measure different things. Visibility is what the assistants said. Traffic is the small part of their readers who clicked.

Section seven: what the report cannot tell you

End with a short, plain list of limits. It protects the reader from over-reading the figures, and it protects you when a number moves for no reason.

The biggest limit is that answers change between runs. A SparkToro and Gumshoe study of 2,961 prompt runs found less than a 1 in 100 chance that ChatGPT or Google's AI gave the same list of brands in any two responses (SparkToro). The same study found that how often a brand appears across many runs is steadier, which is why the report counts appearances and never a rank.

• Answers vary between runs, so a small change can be chance. Say how big a change has to be before you call it real.
• The questions are a sample of what buyers ask, not every question they ask.
• Rivals not on the tracked list can be undercounted.
• Traffic figures exclude visits with no referrer, and AI Overviews clicks.
• Assistants answered through an API can differ from the app a buyer uses.

A worked example of the summary paragraph

The summary paragraph at the top of the report should say what happened in three or four plain sentences, using only figures from the sections below it. Here is a made-up example for a made-up brand, written the way the template asks.

In September we asked the same 30 buyer questions of four assistants each week. Across the 480 answers to questions without our name, we were named in 14, against 11 of 480 in August. Most of the gain came from Perplexity, and Gemini named us in none of its 120 answers in either month. Brand A was named in 61 of the same 480 answers, and two answers described our pricing as unclear.

Notice what the example does not say. It does not call a rise of 3 answers a trend, because that is within the swing a month of answers can show. It does not blend in branded questions. It does not turn the traffic floor into a total. A reader can check every number in it against the tables below.

If the month really did change something, say why you believe it. For example, you might write that the gain is concentrated in the four pricing questions, where a new pricing page went live on 9 September. Give the reason as a reason, not as proof, because one month of answers cannot prove cause.

How to build the report each month

Collect answers weekly and report monthly. Weekly data gives four readings per question per assistant, which smooths some of the swing, and a monthly report is often enough for a reader to act on.

1. Check that the questions and assistants were the same all month, and note any change in the scope line.
2. Pull non-branded and branded counts separately, per assistant and in total.
3. Fill the headline and the per assistant table, with both counts in every cell.
4. List the rivals named instead of you, with their counts over the same answers.
5. Paste the favourable and unfavourable sentences about you, with the assistant and the date.
6. Add AI traffic from analytics, labelled as sessions from labelled AI sources.
7. Write the limits, and say which changes you think are real and why.

Agentcites collects most of what sections two to five need. It asks ChatGPT, Perplexity, Gemini and DeepSeek up to 50 buyer questions every week, keeps every answer word for word, reports branded and non-branded apart, and keeps the sentence behind each claim about you. It does not read your analytics, so section six still comes from GA4. It costs $149 a month per website, as the pricing page shows.