Short answer, as of October 2026: a monthly AI visibility report needs five things on one page — mention rate by prompt group, share of voice against named competitors, the sources assistants cited, what changed and why, and next month's actions. Leave out single-run screenshots, composite scores nobody can explain, and long per-prompt tables. Those three fill pages and start arguments without helping anyone decide anything.
Reporting is where a lot of AI search work quietly fails. The tracking is sound and the content work is real, but the report is either a pasted chat window ("look, it named us") or an export of every prompt and engine. The first proves nothing, because one answer is one sample. The second buries the two facts that matter under two hundred that do not.
Who reads the report, and what they want to know
Write for the reader's three questions, in this order.
Are we showing up more or less than last month? One number per topic, compared with the previous period.
Are we gaining or losing against the businesses we compete with? A relative figure. An absolute rate cannot answer this.
What are you doing about it? A short list with owners and dates.
If a line in the report does not help answer one of those, it belongs in an appendix or nowhere. This is the same discipline as any client reporting, covered more broadly in white-label SEO reporting for agencies. The difference with AI search is that the underlying data is noisier, so the temptation to over-read it is stronger.
The five things worth showing
1. Mention rate by prompt group
Group your prompts by intent: cost questions, comparison questions, problem questions, "who should I use" questions. For each group, show the share of runs in which you were named this month and last month.
Groups are steadier than single prompts because they average over more runs. They also map to how a client thinks. "We are named in about half of the pricing questions and almost none of the comparison questions" is a sentence that leads to a decision. The reasons a single answer bounces are set out in why AI answers change between runs.
2. Share of voice against named competitors
Pick two or three competitors and keep them fixed. For the same prompts, show each brand's share of all mentions.
This line does the most work in the report. If your rate fell and theirs fell too, an assistant changed and there is nothing to fix. If yours fell while theirs rose, something displaced you. Without the comparison you cannot tell those apart. Setting the first baseline is covered in AI visibility benchmarks.
3. The sources assistants cited
List the domains cited most often across your prompt set, and mark which are yours, which are competitors' and which are third parties such as directories, review sites and forums.
This turns the report from a scoreboard into a plan. If a review platform is cited in a third of your category's answers and your profile there is thin, that is next month's first task. The pattern is explained in third-party sources and AI citations.
4. What changed, and the likely cause
Three to five bullets. Each names a movement and gives the most likely explanation with an honest confidence level: "pricing prompts up from contested to owned after the cost page was rewritten" or "comparison prompts down across all tracked brands, consistent with an assistant update."
Use "likely" and "consistent with" when that is the truth. Clients forgive uncertainty. They do not forgive a confident explanation that reverses the following month.
5. Actions for next month
No more than five, each with an owner and a date. Tie each one to a line above it. An action that does not trace back to something in the report is a sign the report is missing a section, or the action is habit.
What to leave out
| Often included | Why it misleads | Show this instead |
|---|---|---|
| A screenshot of one answer naming the client | One run is a sample; it may not repeat tomorrow | Mention rate across repeated runs |
| A single composite "AI score" with decimals | Nobody can say what moved it or what to do | Rates by prompt group, in bands |
| A full table of every prompt and engine | Invites debate over readings inside normal variance | Group totals plus two or three example answers |
| Week-over-week percentage changes | Weekly movement is mostly noise | Month against month |
| Traffic from assistants as the headline | Most AI influence produces no click | Visibility first, traffic as a separate floor |
The composite score deserves a word. Scores are useful inside a tool for sorting and prioritising. In a client report, a score moving from 61 to 58 produces a conversation about three points that mean nothing. If you show a score, show the components next to it.
Bands, not decimals
Report positions in bands. A workable set:
- Owned — named in most runs
- Contested — named in roughly half
- Marginal — named occasionally
- Absent — not named
"We own four of eight cost prompts, contest three and are absent from one" survives being repeated in a meeting the author is not in. "Visibility index 41.3" does not. Bands also stop a client from reading meaning into a two-point move.
The revenue question
Someone will ask what this is worth. Answer in two parts and keep them apart on the page.
The measurable part is small and real: sessions referred by assistants in analytics, and enquiries where the customer said an assistant recommended you. Add a "how did you hear about us" option for AI assistants to your forms if you have not, because it is the cheapest attribution you will ever get. The mechanics are in AI search traffic attribution.
The unmeasurable part is larger: people who read an answer, saw your name and searched for you later, or simply remembered. Say that this exists, say that you cannot count it, and do not put a figure on it. A report that claims precise revenue from AI visibility will be challenged, and it should be. The fuller argument for a budget holder is in the GEO business case.
A one-page template
Header. Client, month, engines tracked, number of prompts, runs per prompt. That last item matters: it tells the reader how much weight the numbers can bear.
Line 1 — Summary. Two sentences. What moved, and whether it is ahead of or behind competitors.
Block A — Mention rate by group. Four to six rows, this month and last, in bands.
Block B — Share of voice. You and two or three competitors, this month and last.
Block C — Top cited sources. Ten domains, labelled yours, competitor or third party.
Block D — Changes and causes. Three to five bullets.
Block E — Next month. Up to five actions with owners and dates.
Appendix. Per-prompt detail and two or three example answers, for the reader who wants to check.
That is one page plus an appendix. It takes about twenty minutes to assemble once the tracking is in place, and most of that time is writing Block D.
When the month is flat
Flat months are normal, particularly after the first quarter. Report them straight. Show that the rate held, show the competitor gap, list the work completed, and name what you expect it to affect and when. Citation changes after content work usually take weeks to show, a timeline covered in how long GEO takes.
The failure to avoid is manufacturing movement. Switching the comparison period, swapping in a friendlier prompt group, or leading with a single good screenshot will all work once.
Producing it without a spreadsheet
The report depends on repeated runs across several engines with the answers stored, which is impractical by hand beyond a handful of prompts. DigiRank Expert's AI Visibility Tracker runs a prompt set across ChatGPT, Claude, Gemini, Perplexity, Grok and Google AI Mode, keeps the answer behind each reading, charts visibility over time from a daily snapshot job, and lists the prompts you are missing from in an Opportunities tab. Tracking starts on the $99/mo Starter plan with 200 prompt-checks a month. The Agency plan at $249/mo adds white-label share pages on a custom domain, which is the practical way to send a client a live version of this report. The full module list is on the features page.
Frequently asked questions
What should a monthly AI visibility report include? Five things: mention rate by prompt group, share of voice against two or three named competitors, the sources assistants cited most often, a short list of what changed and the likely cause, and the next month's actions. That fits on one page and answers the questions a client or manager actually asks.
How often should AI visibility be reported? Monthly is the right cadence for judging progress. AI answers vary from run to run, so weekly figures mostly show noise. Check weekly for breakage, such as a sudden drop to zero, but hold conclusions for the monthly comparison.
Should I show individual prompts in the report? Only as examples. Report mention rate by group of prompts, then show two or three individual answers that illustrate the trend. A table of forty prompts invites debate about single readings that are inside normal variance.
How do I report AI visibility when nothing changed? Say so plainly, show that the rate held against competitors, and report the work that was done. A flat month with a stable competitor gap is a result. Inventing movement to fill the page costs trust the first time a client checks.
Can I connect AI visibility to revenue in a report? Partly. You can show referral sessions from assistants in analytics and any leads that named an assistant as the source, but most AI influence produces no click. Present those figures as a floor, state that clearly, and keep them separate from the visibility numbers.
What is the difference between mention rate and share of voice? Mention rate is how often your brand is named across repeated runs of your prompts. Share of voice is your mentions as a proportion of all tracked brands' mentions for the same prompts. The first shows your own trend. The second shows whether you are gaining or losing ground.
Do clients need to see which engine cited them? A per-engine line is useful when the engines disagree, because the fix differs. If one assistant never names you while the others do, that points to an access or retrieval issue specific to it. When the engines agree, one combined figure is enough.