Reporting AI search, without implying false precision.

Reporting on AI search is where credibility is usually lost, because the honest picture is noisier than stakeholders expect and the temptation to smooth it into something confident is strong. Smoothed reporting survives until someone asks how the number is produced.

What belongs in the report

• Citation rate with sample size. “Named in 34 of 60 asks” rather than a score out of 100.
• Share of voice against named competitors on the same prompts.
• Sources gained and lost since the last period.
• Accuracy issues found in what assistants say about the business.
• Work completed, especially in early months when outcome metrics have not moved.
• A stated variance caveat, because it prevents most arguments about noise.

A one-page report layout

The reports that get read fit on one page. A layout that works for most businesses:

• Top line: citation rate with sample size, and share of voice, this period and last three periods.
• By assistant: a small table for ChatGPT, Google AI, Gemini, Perplexity and Copilot.
• Competitors: who else is named most, and whether that is changing.
• Sources: gained, lost and pending outreach.
• Accuracy: anything wrong that assistants said, and what is being done.
• Business signals: AI referral sessions, self-reported assistant inquiries, branded search.
• Next period: the three things that will be done.

What to leave out

Composite scores that blend assistants, prompts and metrics into one number. They feel satisfying and they destroy the information needed to act.

Rank-style presentation, which invites comparison with search positions that do not apply.

Anything implying attribution from an assistant answer to revenue. That chain does not exist and claiming it eventually costs you the account.

Presenting a probabilistic metric

Say the sample size every time. Show the trend rather than the point. State plainly that responses vary between sessions and that small movements are not meaningful.

Counterintuitively, reports that acknowledge uncertainty tend to survive scrutiny better than confident ones, because the first question a skeptical stakeholder asks is how the number was produced. Having a good answer ready is worth more than having a bigger number.

Adding the business signals

Visibility numbers alone invite the question “so what?”. Put them next to signals the business already cares about: referral sessions and conversions from assistant domains in analytics, inquiries that mention an assistant at intake, branded search volume in Search Console, and, where available, AI citation counts from Bing Webmaster Tools.

Keep the relationship modest. The honest statement is that these indicators tend to move together over time, not that one causes the other in any given month. Stakeholders who hear that early are far less likely to be surprised later.

Dashboards versus reports

A live dashboard is useful for the person running the program, who needs to see individual answers and sources. It is usually a poor way to inform leadership, who need context and a recommendation more than a chart that moves every day.

Use both for what they are good at: a dashboard for operators, updated as often as data arrives, and a written monthly or quarterly report for decision-makers, with the interpretation done for them.

An example summary paragraph

Reports read better with a plain-language summary at the top. For example: “Across 20 prompts asked five times each in ChatGPT and Google AI, we were named in 38 of 200 answers (19%), up from 27 (14%) last quarter. Competitor A fell from 31% to 26%. Two new directory listings are now cited. One inaccuracy (old hours) was corrected.” That is more useful to a busy stakeholder than any chart.

Frequently asked questions

What should an AI search report include?

Citation rate with sample size, share of voice against named competitors, sources gained and lost, accuracy issues found, work completed, and an explicit note that responses vary between sessions so small movements are not significant.

Should AI visibility be reported as a single score?

No. Composite scores blend assistants, prompts and metrics into a number that feels decisive and cannot be acted on. Segment by assistant and report sample sizes.

How do I explain AI search results to a skeptical stakeholder?

State the sample size, show the trend rather than a point, and say plainly that the metric is probabilistic. Acknowledged uncertainty survives scrutiny better than a confident number nobody can explain.

How often should AI visibility be reported?

Monthly for operators, quarterly for leadership. Monthly leadership reports on a noisy metric tend to produce overreaction to normal variance.

Should AI visibility be in the same report as SEO?

Yes, side by side. Showing rankings, clicks and AI citations together explains patterns, such as stable rankings with falling clicks, that neither shows alone.