How to report AI visibility to your board: a one-page scorecard.

The short answer.

Report AI visibility to stakeholders on one page with six measures: how often AI answers to your priority buyer questions name you, how often they cite your pages, whether the facts they state about you are right, your impressions in Google’s AI features from Search Console, AI-referred sessions in GA4, and enquiries influenced by AI. Show each with its source, its date range and how confident you are in it.

The first three come from a repeated test of real buyer questions. The last three come from systems you already own. Together they answer the board’s real question: are we present where buyers now ask, and is it turning into business?

What the board is really asking.

When a chair or investor asks “what are we doing about AI search?”, they are rarely asking for a dashboard. Underneath are three questions. Is this changing how our buyers find suppliers? Are we in the answers when it matters? Is it producing enquiries, or is it a cost with no return yet?

A good scorecard answers those in that order and ends with one decision for the board to take: fund the next piece of work, hold, or stop. Anything that doesn’t help with that decision belongs in an appendix.

Agree the question list before the first report

Every answer-based measure depends on which buyer questions you test. Agree 20–50 of them with sales before you report anything, grouped by stage: the problem, the shortlist, the comparison and the risk. Then freeze them for at least two quarters. If the list changes every month, the trend means nothing. Our guide on how to check your AI search visibility sets out how to build that list and run the test.

The one-page scorecard.

This is the shape we recommend. The numbers below are invented for a fictional company, to show the layout. They are not client data.

Three things make this page work. Every number has a source line under it. The sampled measures carry a confidence label and the counted ones say what they miss. And it ends with a decision, not a summary. Keep commentary to one sentence per measure: what changed, and why you think it changed.

Share of answers and citations.

Measure 1, named in AI answers, is the share of answers to your priority questions that name your company. Measure 2, our pages cited, is the share that link to a page on your site. Keep them separate: an answer can quote your guide and recommend a competitor.

How to collect them

• Run every frozen question on the engines your buyers use, in fresh, signed-out sessions, set to the buyer’s country.
• Run each question more than once. AI answers vary from run to run, so a single run is an anecdote.
• Record the full answer text and the cited links, with the date, engine and mode.
• Report the share of all answers, with the denominator: “18% of 270 answers”, not “18%”.

Repeat runs are not optional. In research reported by Search Engine Journal on 30 January 2026, SparkToro and Gumshoe.ai repeated brand-recommendation prompts 60–100 times per platform across ChatGPT, Claude and Google’s AI, and found less than a 1-in-100 chance of the same brand list appearing twice. How often a brand appeared across many runs was more stable than where it ranked.

Our own baseline shows what an early reading looks like. On 6 November 2026 we asked ChatGPT, Perplexity and Google’s AI Overview, once each, which agencies help B2B companies improve their visibility in AI search in the United States. They named 8, 10 and 6 agencies; we were in none of them, and only one agency appeared in all three. On a scorecard that is “0 of 3 single runs”, with low confidence, and it is still worth reporting because it sets the starting line.

Are the facts about you right?

Measure 3 is the one most scorecards miss, and boards understand it immediately. Being named matters less if the answer gets you wrong.

We found this on ourselves. On 6 November 2026 we asked ChatGPT, in a temporary unpersonalised chat, what Prolinkage is and whether it would recommend us. It said we were worth shortlisting. It also quoted a 4.17x figure for one client’s Google AI impressions that our own case study has since replaced with a newer dated figure of 6.7x, and it found conflicting founding years: 2015 on our own profiles, 2011 on older directory listings. A visibility tool would have scored that answer as a success.

How to collect it

List 10–15 facts a buyer would rely on: what you do, who for, where you are, how long you have operated, your headline proof and any credentials. Ask each engine about your company by name, and mark each fact as correct, outdated, wrong or missing. Report “9 of 12 correct” and name the errors. Then trace each error to its likely source, usually an old directory profile, an out-of-date page of your own or a third-party article, and put fixing it on the plan.

Google’s own AI impressions.

Measure 4 is the only first-party number for AI answers. Google announced Search generative AI performance reports in Search Console on 3 June 2026, and added that they had rolled out to all websites worldwide by 31 August 2026. Google’s help page for the report says it shows impressions in AI Overviews and AI Mode: how many times links to your site were shown to a user in a generative AI feature on Google Search. You can break it down by page, country, device and date.

Know its limits before you put it in front of a board. It reports impressions only. Google says traffic from these features is included in the main Performance report under the Web search type, not split out. Several results from your site in one AI feature count as one impression in the chart. And if your site hasn’t had enough impressions, Google says the report may not appear at all.

A real, dated example

For one client, FullTilt Team Development, impressions inside Google’s generative AI features for fulltiltteams.com rose 6.7 times, from 3.45K to 23.1K, comparing the last three months with the previous three months, in a Search Console capture taken on 4 September 2026.

Note how it is worded: the metric’s exact name, both windows, the source and the capture date. That is the standard for every number on the page. Relative windows such as “last three months” move every day, so always record the date you read them, or set a custom date range.

AI-referred sessions in GA4.

Measure 5 counts visits that arrive from AI assistants. GA4’s default channel grouping has no AI channel, so these sessions sit inside “Referral” unless you separate them. Google’s Analytics help page on custom channel groups shows how, using an example channel named “AI assistants”.

How to set it up

1. In GA4 Admin, open Data display, then Channel groups, and create a new channel group from the default one.
2. Add a channel called “AI assistants”, with a “matches regex” condition listing assistant domains. Google’s help page gives an example pattern covering popular assistants.
3. Move the new channel above “Referral”. GA4 assigns traffic to the first channel it matches, so order matters.
4. Save it. Google says custom channel groups can be applied to your reports retroactively.

Two practical notes. A standard GA4 property allows two custom channel groups, so agree with whoever owns analytics before using one. And the pattern needs updating as assistants change their domains.

Why this number undercounts

Most AI influence does not produce a click. A Pew Research Center study published on 22 July 2025, tracking 900 US adults in March 2025, found that people who saw a Google AI summary clicked a traditional result on 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary itself on 1%. Report AI-referred sessions as a floor, and say so on the page.

Enquiries influenced.

Measure 6 is the one the board cares about most, and the hardest to count. Use two signals and report them together.

• Arrived from AI: enquiries whose session came through your “AI assistants” channel. This needs the source saved on the form entry or in your CRM, not only in analytics.
• Said AI: enquiries where the buyer told you. Add an optional “how did you hear about us?” field with an AI assistant option, and ask on the first call.

Count an enquiry once, even if both signals apply, and count real buyers only, not spam or vendor pitches. Self-reported answers are imperfect, but a buyer saying “ChatGPT suggested you” is evidence no tool can give you. Our guide on tracking website leads from first visit to CRM covers saving the source on every entry, and turning AI visibility into qualified enquiries covers the pages that make an enquiry more likely.

If the number is small, say so. “Two enquiries this quarter mentioned AI; one became a proposal” is more useful to a board than a large, vague influence estimate.

Cadence and uncertainty.

Collect often, report rarely. Sampled measures need repeated runs to settle; counted measures need whole periods to compare.

Report to the board quarterly, with the monthly points visible as small dots, not a smooth line. For sampled measures, use three plain labels. Low: fewer than about 50 answers, or the question list changed. Medium: a stable list, repeated runs, one engine set. High: the same list for two or more quarters with repeated runs. If a change is smaller than the spread between runs, call it “no clear change”. That sentence protects your credibility more than any chart.

What not to report.

• A single screenshot. One answer, good or bad, is an anecdote. Use it as an illustration beside the measure, never as the measure.
• “Rank” in an AI answer. Position in a list changes too much between runs to trend.
• A tool score without its formula. If you can’t say what the numerator and denominator are, the board can’t trust it, and you can’t defend it when it falls.
• Undated numbers. “Up 6.7x” means nothing without both windows and the capture date.
• Totals of mixed things. Don’t add mentions, citations and impressions into one “AI reach” figure.
• Revenue you can’t trace. Report enquiries influenced, and let the pipeline report carry revenue.

Our article on cost per lead versus cost per qualified opportunity explains why boards should judge every channel, AI included, on qualified opportunities rather than raw counts.

Build your scorecard.

Enter what you have; leave blank what you don’t. The builder writes a one-page summary you can copy into a board pack. Nothing you type is sent anywhere.

The rough range assumes each answer is independent, which flatters the precision, so treat it as a minimum. Paste the summary into your board pack, then add one sentence per measure on what changed and why.

Questions owners ask.

How do I share AI visibility data with stakeholders?

Use one page with six measures: named in AI answers, pages cited, brand facts correct, Google AI feature impressions, AI-referred sessions and enquiries influenced. Give each a source line, its date range and a confidence label, and end with one decision you need. Keep raw answers and tool exports in an appendix.

Which AI search KPIs belong in a board pack?

Those that lead to a decision: whether buyers see you when they ask priority questions, whether the facts are right, and whether visits and enquiries follow. Leave out AI “rank”, single screenshots and tool scores whose formula you cannot explain. Boards should judge AI search on qualified enquiries, like any other channel.

Does Google Search Console show AI Overview data?

Yes. Google’s generative AI performance report in Search Console shows impressions in AI Overviews and AI Mode by page, country, device and date. Google said it had rolled out to all websites worldwide by 31 August 2026. It shows impressions only; clicks stay inside the main Performance report.

How do I track ChatGPT traffic in GA4?

Create a custom channel group with a channel such as “AI assistants” that matches assistant domains with a regular expression, and place it above Referral so sessions are assigned to it first. Google says custom channel groups can be applied to reports retroactively. Treat the result as a floor, because most AI influence produces no click.

How often should we report AI visibility?

Collect monthly and report quarterly. Sampled measures need repeated runs to settle, and month-to-month movement is often noise. Freeze your question list for at least two quarters so the trend means something, and say “no clear change” when a movement is smaller than the spread between runs.

Sources: Prolinkage AI self-test on ChatGPT, Perplexity and Google AI Overview, 6 November 2026. Client result: FullTilt Team Development, Google Search Console Generative AI features impressions, last 3 months vs previous 3 months, captured 4 September 2026. Google Search Central blog, 3 June 2026; Google Search Console Help, generative AI performance report; Google Analytics Help, custom channel groups (read 7 November 2026). Search Engine Journal, 30 January 2026, on SparkToro and Gumshoe.ai research. Pew Research Center, 22 July 2025.