AI Visibility Reporting for Agencies: A Monthly Client Template

AI visibility reporting for agencies means giving each client one monthly page that shows how often AI engines name them, how accurately they describe them, who is named instead, and what traffic follows. Use a fixed prompt set, show every denominator, and separate mentions, citations and referrals.

Key takeaways

• Report five blocks every month: coverage, share of voice, framing, sources and referrals. Each answers a different client question.
• Show the denominator on every rate. "Mentioned in 12 of 40 answers" is a finding; "30% visibility" alone is a claim.
• Freeze the prompt set, engines, market and language between reports. Change any of them and you have to restart the trend line.
• Keep referral traffic in its own block. AI referrals are a small share of visits, so they cannot carry the whole report.
• End every report with what the numbers do not prove, then three actions with an owner each.

How do agencies report AI visibility to clients?

Most agencies start by pasting a vendor dashboard screenshot into a deck. Clients then ask what the score means, and nobody can say. A better habit is to treat the report like a lab note: the question set, the method, the result, the limit and the action.

Search Engine Journal published a useful frame on 5 November 2026. In Search Engine Journal, November 2026, Bartek Jazwinski describes three gates: inclusion (is the client named or cited), framing (is the description accurate) and endorsement (is the client recommended for the buyer's need). He also says the three are distinct measures, not a funnel, so a rising citation count beside a falling endorsement share can signal trouble.

The template below keeps those three gates and adds two things the agency reader needs to run a retainer: a share-of-voice block against named competitors, and a referral block that ties back to analytics.

What should an AI visibility report include?

Five blocks, in this order. The order matters: the first blocks are what you measure directly by asking engines questions, and the last block is what you read from the client's own systems. Do not blend them.

Noma by Nurdd's own AI Visibility Measurement Standard uses the same logic: state each numerator and denominator, run the fixed prompt set several times per engine, and call a change only when it exceeds the margin of error.

Block 1: Coverage, with the denominator shown

Search Engine Journal's worked example is simple: if a client appears in 12 of 40 responses, the observed mention rate is 30%, and that describes the tested questions, not every AI search in the market. Track mentions and citations separately. A mention is a name in the text; a citation is a linked source. A September 2026 Search Engine Journal, September 2026 article, drawing on Ahrefs data, makes the same split because a mention without a citation and a citation without a mention have different causes.

On the same sample, a simple binomial calculation gives roughly plus or minus 14 percentage points at 95% confidence for 12 of 40. That is our arithmetic, not a vendor figure, and repeated runs of the same prompt are not fully independent, so treat it as a rough guide. The practical rule: a move from 30% to 35% on 40 answers is noise. Say so in the report.

Block 2: Share of voice against rivals you name

Pick three to five competitors with the client at kickoff and keep them fixed. Count every brand mention in the answers, then divide the client's mentions by the total. Report the rivals' figures beside it. For method detail, see how to measure AI share of voice across engines.

Split by engine. Bluefish data covered in Noma's post on engines disagreeing on the top brand found disagreement on 35% of buying topics, so an all-engine average hides the engine where a client is weakest.

Block 3: Framing and sentiment

Save the exact sentence each engine uses about the client. Compare it with a one-page brand facts sheet the client signs off. Report framing accuracy as the share of evaluated descriptions that are correct, not as a share of all answers. A client that has moved upmarket but is still described as a tool for freelancers needs a page fix, not more citations. The AI brand fact check walks through the claim-by-claim comparison.

Block 4: Sources

List the URLs cited for the prompts where the client is absent. Group them: the client's own pages, review sites and publishers, communities, and competitors' pages. This is the block that turns a report into a work plan, because it says where to publish, correct or pitch.

Block 5: Referrals and crawlers

Referral traffic belongs in the report but not at the top. Conductor's 2026 Benchmarks Report analysed 13,770 enterprise domains across 10 industries, aggregating 3.3 billion sessions, and its release says AI referrals make up only about 1 percent of traffic (Business Wire, November 2025). Pew Research Center's study, reported by Search Engine Journal, July 2025, found users clicked a link inside an AI summary in just 1% of visits to pages with one.

Both figures say the same thing to a client: most AI influence never shows up as a session. A report that judges AI work on referrals alone will undervalue it, and one that ignores referrals will be unfalsifiable. Show both, and keep them in separate blocks. The GEO referral traffic measurement plan covers branded search lift and assisted conversions as the bridge.

How do you set up the prompt set so reports stay comparable?

The prompt set is the report's instrument. Change it casually and every month-on-month comparison is invalid. Search Engine Journal's advice is to document the prompt set, market, language, platform, dates and number of repeated runs, and always show the denominator. In its words: "Eight mentions across 20 responses is not the same as eight mentions across 100."

1. Write 20 to 30 buyer questions per client. Mix category questions ("best X for Y"), comparison questions and decision-stage questions. The free AI Prompt Finder suggests questions buyers ask about a category, tagged by intent.
2. Leave the brand name out of most prompts. Branded prompts tell you about framing; unbranded ones tell you about discovery. Report them separately.
3. Fix engines, market and language. Record them on the first page of the report.
4. Run each prompt more than once per engine. Answers vary between runs, so a single run is an anecdote.
5. Version the set. If you add prompts, report old and new sets as two series for one month, then merge.

Noma's guide to which prompts to track and the chapter on choosing prompts that measure something go deeper on prompt types.

What does a worked monthly report look like?

The example below uses invented placeholder figures to show layout only. They are not results from any client.

Two lines in that template do most of the client-facing work. The limits line stops the client over-reading a good month. The actions line makes the report a plan: correct the outdated comparison page, pitch the two publishers cited most, add the missing use case to the product page.

Which AI visibility tool should an agency use for client reporting?

Judge a tool against the report above, not against its dashboard. Ask five questions in a trial, and run them on one real client.

For comparison, Semrush's agency reporting page says its reports can combine AI mentions, citations, sentiment and competitor visibility, with branding and traffic, leads and revenue alongside (Semrush, as checked 8 November 2026). The point of any such comparison is the test table above: run it on the client and see which tool answers all five. Noma's guide to AI visibility tools for agencies and its comparison by job set out the wider criteria.

Where does Noma fit in the reporting workflow?

Noma by Nurdd is an AEO and AI visibility platform that asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, a client's buyer questions every day. For each engine it reports visibility, position and sentiment, share of voice against competitors, the sources each answer cites, which AI crawlers fetch each page and which assistants send visitors back. Accuracy checks against a client-signed facts sheet, and the analytics referral and conversion data, stay with the agency.

See it on your own brand: book a demo or start the 3-day free trial. A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged. Plans are listed on the pricing page, and the agency page covers multi-client tracking.

How do you check the client's pages before the first report?

A baseline report is more useful if you first rule out technical blockers. Run the client's key pages through the free AI SEO Checker, which shows whether ChatGPT, Claude, Perplexity and Google AI Mode can read a page, plus its SEO, with the top fixes. Put any failures in the first report as quick wins.

Google's documentation is the reference for what not to promise. Google Search Central, as checked 8 November 2026, says sites appearing in AI Overviews and AI Mode are counted in the overall Search Console Performance report under the Web search type, and that no special schema.org markup is required to appear. So Search Console will not give you a separate AI line, and no markup change should be sold as a guaranteed fix.

What mistakes make AI visibility reports misleading?

• One blended score. It hides whether a change came from mentions, citations or a different engine.
• Missing denominators. Eight mentions means nothing without the number of answers.
• A moving prompt set. Adding easy prompts inflates visibility without any real change.
• Counting citations as endorsements. A model can cite a client's article and recommend someone else.
• Judging by referrals alone. With AI referrals at about 1 percent of traffic in Conductor's sample, the click is the smallest part of the effect.
• Promising positions. Promise work and measurement. Noma's Agency AEO Playbook frames retainers this way.

How we researched this

We reviewed Search Engine Journal's analysis of agency AI visibility measurement, platform documentation from Google Search Central, the Conductor press release on its benchmark, Pew Research coverage, and a vendor's own agency reporting page. Facts and prices were checked on 2026-11-08. We did not run client tests for this post, and the example figures in the template are placeholders for layout, not results.

What to do next

Pick one client, write the 30-prompt set, and run the five-block template for one month before rolling it out. If you want the daily data collected for you across 13 AI engines, book a demo or start the 3-day free trial.