How do I share AI visibility data with my board?

The short answer.

A board does not need a dashboard of AI metrics. It needs to know three things: are buyers seeing us in AI answers, are those buyers arriving on our website, and is any of it turning into pipeline? Each question has its own data source, its own error and its own owner, so each gets its own row on the page.

The mistake we see most often is a single blended number, such as an “AI visibility score”, sitting next to a revenue figure as though one caused the other. A board member will ask how the score is calculated and what it has to do with the revenue line. If the honest answer is “we are not sure”, trust in every other number on the page goes with it.

If you need the reasoning behind each measure first, our guide on reporting AI visibility to your board walks through the six measures one by one. This page is the working kit: the template, the scorecard and the setup.

The one-page board summary.

Use one table, four sections and no more than nine rows. The worked example below uses made-up numbers for a fictional company, so you can see the shape; the downloadable template has the same rows with blank cells.

Why these rows and not others

Visibility is measured by sampling: you ask a fixed set of buyer questions, several times per engine, and count. Answers change from run to run, so a share of runs is steadier than a position in one list. Website and pipeline numbers come from systems you already own, which makes them more trustworthy but also narrower, because many AI-influenced buyers arrive with no referrer at all.

That last point belongs on the page in plain words. A board that is told “this undercounts, and here is why” will trust the trend. A board that later discovers the undercount on its own will not.

Cadence

Monthly for visibility and website rows, quarterly for closed-won value. Deals take longer than a month to close, so a monthly revenue row mostly reports noise. Keep the same question panel between months and note any change to it in the footnote, or the trend is meaningless.

Write your summary sentence.

Most boards read the first line and the decision row. Put the whole story in one sentence, with the evidence in it. Fill in the boxes and copy the result. Nothing you type leaves this page.

The last clause matters. It tells the board you know the difference between two numbers moving together and one causing the other, which is the question a careful director will ask.

What to leave off the board page.

Some numbers look impressive and mean little. Leave these in the working file, not the board pack.

• A position in a single AI answer. Answers vary between runs, so “we were third in ChatGPT” is one observation, not a fact about the market.
• A vendor’s blended score without its formula. If you cannot write the numerator and denominator, you cannot defend the number.
• Prompt counts. “We track 1,000 prompts” describes effort, not outcome.
• Revenue “from AI” with no source field behind it. If the CRM cannot show which deals had an AI original source or a self-reported AI mention, the figure is an estimate. Label it as one, or leave it out.
• Every engine at once. Pick the two or three your buyers use. A wall of engines hides the one that matters.

If a number fails the test “could a sceptical director recalculate this from the source?”, it belongs in an appendix.

Twenty questions to ask any AI-visibility agency or tool.

When a buying committee compares providers, the same few questions separate a measured programme from a sales pitch. Use these in a first call or an RFP. Score each answer 0 (no or vague), 1 (partly) or 2 (clear, with evidence), and multiply by the weight. The editable version is in the kit. Our separate article on questions to ask before choosing an SEO or AI agency covers the wider relationship; this list is specific to AI visibility.

The maximum is 90 points. We would not hire a provider that scores 0 on questions 5, 9 or 11, whatever its total. Those three are about whether you can check the work, whether the promises are honest, and whether you keep what you paid for.

Red flags.

Each of these, on its own, is a reason to ask harder questions. Two or more is a reason to walk away.

• Guaranteed rankings or placements. Nobody outside OpenAI, Google or Perplexity controls what their engines say, so a guaranteed placement is a promise the seller cannot keep.
• “#1 in ChatGPT” or “number one AI agency”. There is no single ChatGPT ranking to be first in. Answers differ by wording, memory, location and run.
• Undated numbers. “Visibility up 400%” means nothing without a period, a tool and a capture date. Ask for the screenshot.
• A fixed timeline to appear. “In ChatGPT within 30 days” cannot be known in advance.
• A score you cannot recalculate. If the vendor will not show the formula, you cannot report it upward.
• Revenue claimed from visibility alone. Visibility and revenue can both rise for other reasons. A provider worth hiring will say so before you do.
• Data you lose when you leave. Your question panel and answer archive are your baseline. Without them, the next provider starts from zero.

Prolinkage holds itself to the same list. Every client figure we publish carries its tool, period and capture date, and we scope and price work only once we both know what is worth fixing.

Tool, agency or both?

A monitoring tool tells you what the engines say. It does not change it. An agency changes things on and off your site, but you still need a way to check its work. Most mid-sized B2B teams end up with one of three set-ups. Walk the questions in order.

For a side-by-side of the monitoring tools themselves, see our comparison of AI visibility tools. For what a programme typically involves and how pricing is usually structured, see what AI search optimisation costs.

The KPI dictionary.

Agree these definitions before the first report, and write them in the footnote of every board page. Changing a definition later breaks the trend, so if you must change one, restart the series and say so.

Two rules keep this honest. Never add the tracked and self-reported enquiry counts together, because the same buyer can appear in both. And show the count next to every rate when the base is small: “3%” of 40 sessions is one visit.

GA4: an AI-referral channel.

Google Analytics now has a default channel called “AI Assistant”. Google’s channel documentation describes it as “the channel by which users arrive at your site from sources like ChatGPT, Gemini”, and states that it excludes Google’s AI Overviews and AI Mode. Search Engine Journal reported its arrival on 14 May 2026 and noted that Google has not published the full list of assistants it recognises. Reports disagree on whether Perplexity and Claude are included, so check your own Source report rather than assuming.

A custom channel group is still worth adding, because it lets you see exactly which domains are counted. Google’s custom channel group documentation (accessed 8 November 2026) sets out the steps:

1. Admin, then Data display, then Channel groups. You need Editor access or above.
2. Create new channel group. This copies the default group.
3. Add a channel called “AI assistants”. Condition: Session source matches regex, using the line below. Add a second condition with OR: Session medium exactly matches ai-assistant, so anything GA4’s own list catches is kept.
4. Reorder so “AI assistants” sits above the default “AI Assistant” and “Referral” channels. Google says traffic goes to “the first channel whose definition it matches”.
5. Save. Google says custom channel groups “can be applied to your reports retroactively”. Standard properties allow two custom groups.

The domains come from current analytics guides by Terminus (updated September 2026) and Orbit Media (updated July 2026), plus HubSpot’s own examples; the AI vendors themselves do not publish referrer lists. GA4 uses RE2 syntax and matches the whole string, which is why the pattern is anchored.

What this channel cannot see

Clicks from AI mobile and desktop apps often arrive with no referrer and land in Direct. Clicks from Google’s AI Overviews and AI Mode arrive as Google organic; use Search Console’s generative AI report for those. Do not add bing.com to the pattern, because most of it is ordinary search. Then make sure your enquiry form is marked as a GA4 key event; our guide to why GA4 lead numbers are wrong covers the usual faults.

CRM: original source and self-report.

GA4 counts visits. The board wants to know about buyers. Two CRM fields close that gap, and they should sit side by side, never merged.

1. The tracked field: original source

HubSpot’s Original Traffic Source property now includes an “AI Referrals” value, with the AI platform domain (for example chatgpt.com or claude.ai) in the first drill-down, according to HubSpot’s knowledge base (last updated 4 August 2026). For revenue, build a deals report grouped by the associated contact’s Original Traffic Source, filtered to closed-won, with close date as the period. HubSpot’s dedicated revenue attribution reports need Marketing Hub Enterprise; the original-source deals report works on lower tiers through the custom report builder. Other CRMs have an equivalent first-touch field; if yours does not, pass the GA4 session source into a hidden form field.

2. The self-reported field

Add one required question to your main enquiry form: “How did you first hear about us?” Make it free text, not a dropdown, so you do not suggest an answer. Buyers who found you through an AI answer and then searched your name will show as Direct or Organic in tracking but often write “ChatGPT” here. Code the answers monthly.

If your deals close offline or through sales calls, the original source still works as long as the contact was created from a tracked form. Our guide to offline conversion tracking for B2B covers the rest.

Download the kit.

Four plain files that open in Excel, Google Sheets or any text editor. No sign-up. Edit them freely and drop them into your own reporting.

Questions owners ask.

How do I share AI visibility data with stakeholders?

Use one page with three views kept apart: sampled AI answers (how often engines name and describe you correctly), website behaviour from GA4 (AI-referred sessions and enquiry key events), and pipeline from your CRM (enquiries and closed-won value by original source). Show the numerator, denominator and period for every rate, and end with one decision.

What AI search KPIs should a board see?

No more than six: panel coverage per engine, description accuracy, Google AI impressions, AI-referred sessions, AI-sourced accepted enquiries (tracked and self-reported, side by side), and quarterly closed-won value by original source. Everything else belongs in the working file.

Can I connect AI visibility to revenue?

Partly. A CRM original-source field and a “How did you first hear about us?” question show which deals began with an AI referral. Neither proves that visibility caused the deal, and many AI-influenced buyers arrive with no referrer. Report the numbers side by side and say what they cannot show.

Should we buy an AI visibility tool or hire an agency?

A tool reports what engines say; it does not change it. If you have someone with time to act on the data and you already know what to fix, a tool may be enough. If the gap is on other people’s sites or you lack the time, you need hands as well. Many teams use an agency working from a tool they own.

Does GA4 track ChatGPT traffic automatically?

GA4 added a default “AI Assistant” channel in May 2026 that covers sources such as ChatGPT and Gemini, but Google has not published its full list. Add a custom channel with the regex on this page to see exactly what is counted, and remember that app traffic often lands in Direct.

Sources (accessed 8 November 2026 unless stated): Google Analytics Help, default channel group, custom channel groups and regular expressions; Search Engine Journal, 14 May 2026; Terminus, updated 16 Sep 2026; Orbit Media, updated 7 Jul 2026; HubSpot Knowledge Base, traffic source properties (updated 4 Aug 2026) and attribution reporting (updated 21 Jun 2026). Client result: HubSpot, closed-won deal value by original traffic source, Jul–Sep vs Apr–Jun 2026, client name withheld. The worked example in chapter 02 uses fictional numbers.