A partner pastes a screenshot into Slack. ChatGPT, asked about firms like yours, named you. Second in the list. Somebody drops a fire reaction. Somebody else types "we're in." The managing partner forwards it to the whole team with three exclamation points.
Run the same question again an hour later and you might be gone.
To measure AI visibility, stop counting single answers and start counting patterns. Run a fixed set of buyer prompts, at least three runs per prompt per engine, across the major AI assistants, and track five numbers: how often you're named, how you stack up against named competitors, whether the AI cites your pages, whether it gets your facts right, and whether any of it turns into pipeline. The rest is how to do it without fooling yourself.
How do you measure AI visibility?
You measure AI visibility by running a fixed prompt set, at least three runs per prompt per engine, across ChatGPT, Claude, and Perplexity (plus Google's AI if your buyers use it), then scoring the answers on mention rate, share of voice, citation rate, and accuracy. Then you connect those answers to real outcomes with AI-referred pipeline. Measure monthly. Watch the trend. Ignore any single answer.
This is the scoreboard half of generative engine optimization. The playbook tells you what to build. This tells you whether it's working.
Why one screenshot proves nothing
AI answers are probabilistic. Ask the same thing twice and you get two different answers. That's the design.
SparkToro and Gumshoe.ai tested this hard. In a study reported by Search Engine Land, the study covered about 3,000 runs of 12 identical prompts across ChatGPT, Claude, and Google's AI. The odds of getting the same brand list twice were under 1 in 100. The same list in the same order? Closer to 1 in 1,000. The researchers called ranking positions "effectively meaningless."
Here's the useful part. Some brands still showed up in 60% to 90% of responses for a given intent. The list shuffles. The strong brands keep appearing. So the question worth asking is how often you appear across many runs. Your rank in one run tells you nothing.
That screenshot in Slack is one roll of the dice. Celebrate it. Just don't report it as a result.
The AI Visibility Scorecard: five metrics that mean something
Five metrics. One spreadsheet. A monthly hour. Small enough for a 15-person firm to actually keep up.
Before any of it works, you need a prompt set. Write 15 to 25 questions your buyers really ask, in their words. Mix three kinds: category questions ("best executive search firms for healthcare boards"), problem questions ("how do I replace a CFO without spooking investors"), and comparison questions ("[your firm] vs [competitor]"). Lock the set. Changing prompts every month makes the trend line worthless.
Then run every prompt at least three times per engine, in a fresh or logged-out session with memory off. Log every answer.
1. Mention rate
Definition: The share of all answers in which your firm is named.
How to calculate it: Answers that name you, divided by total answers. Count every run on every engine.
Cadence: Monthly.
What good looks like: Rising over time, and rising fastest on your category and problem prompts. Those are the questions buyers ask before they know your name, which makes them the ones that build pipeline.
2. Share of voice
Definition: Your mentions as a share of all mentions for you plus three to five named competitors.
How to calculate it: Count total mentions for each firm across all answers. Divide yours by the combined total. Pick competitors you actually lose deals to, and keep the list fixed.
Cadence: Monthly, reviewed quarterly.
What good looks like: Closing the gap on the competitor who shows up most. If your mention rate climbs but a competitor's climbs faster, you're losing ground on a rising tide.
3. Citation rate
Definition: The share of answers that link to or cite one of your own pages, and which pages get cited.
How to calculate it: Answers that cite your domain, divided by total answers. Then list the cited URLs and count them.
Cadence: Monthly.
What good looks like: More citations, spread across your service and proof pages. The "which pages" part matters more than the rate. If the AI only ever cites your About page, your service pages aren't saying anything a model can lift. That's a Claim problem in the CITE Framework, and it's fixable.
4. Accuracy score
Definition: The share of factual statements the AI makes about you that are correct.
How to calculate it: Pull every factual claim about your firm from your identity and proof prompts: services, locations, industries served, leaders' names and titles, client results. Mark each right or wrong. Correct claims divided by total claims.
Cadence: Monthly, plus right after any site rewrite.
What good looks like: Every fact right, and wrong facts trending toward zero. This is the metric firms skip, and it's the one that hurts most. An AI that recommends you and says you have a Dallas office you closed in 2022 is sending a buyer to a firm that doesn't exist. Wrong facts usually trace back to stale pages, old directory listings, or schema that doesn't match the visible text.
5. AI-referred pipeline
Definition: Opportunities that came from someone who found or vetted you through an AI tool.
How to calculate it: Two sources, combined.
First, GA4. Build a custom channel group with an "AI" channel whose source matches chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com. Put it above Referral in the channel order so those sessions get their own line. Then watch which of those sessions turn into form fills and booked calls.
Second, ask. Add "How did you hear about us?" to your intake form as an open text field. Then ask it again, out loud, on the discovery call. Log the answer on the opportunity in your pipeline.
Cadence: Monthly count, quarterly review against closed revenue.
What good looks like: Any number above zero that you can name. Then growth. A lot of AI traffic shows up with no referrer and lands in Direct, so GA4 alone will undercount. The human question fills the gap.
Why pipeline beats traffic
Here's the uncomfortable math. AI answers send fewer clicks.
Pew Research Center found Google users who saw an AI summary clicked a traditional result in 8% of visits, against 15% when there was no summary. They clicked a link inside the summary in just 1% of visits. If you grade AI visibility by traffic, you'll conclude it's tiny and stop paying attention.
The clicks that do come behave differently. Ahrefs reported on its own site that AI search was 0.5% of visitors yet drove 12.1% of signups, with AI search visitors converting at a 23x higher rate than traditional organic visitors. That's one company's data, a software company with a self-serve signup, so don't paste it into your forecast. The direction is the lesson. The buyer who arrives from an AI answer already got a recommendation. They show up warm.
For a service firm, the unit that matters is the opportunity. A partner meeting with a qualified buyer. Count those.
The proof: "Two from chat, one from Claude."
In 2026 we restructured RE/MAX Commercial Advisors Group's site content for AI answer engines. Afterward, Patrick J. Buckhoff, CCIM, who leads the Houston brokerage, reported this:
"Recently I've gotten three calls inbound that people found us online. Two from chat, one from Claude."
Patrick could name the source of each call because the brokerage tracks where its inbound comes from. All three named an AI tool. That's AI-referred pipeline, and it's worth more than any screenshot in Slack. Read the RE/MAX Commercial Advisors Group case study.
If you only add one metric this month, add the question.
A worked example: a 15-person executive search firm
Here's how the arithmetic plays out. This is a hypothetical example and every number below is invented to show the math.
Say you run a 15-person executive search firm in Charlotte that places CFOs and COOs at healthcare companies. You write a 20-prompt set. You run each prompt 3 times a month on ChatGPT, Claude, and Perplexity.
20 prompts × 3 runs × 3 engines = 180 answers.
Mention rate. Your firm is named in 45 of the 180 answers. 45 ÷ 180 = 25% mention rate.
Share of voice. You track four competitors. Across the same 180 answers, total mentions come out like this:
| Firm | Mentions |
|---|---|
| Competitor A (national) | 90 |
| Your firm | 45 |
| Competitor B | 36 |
| Competitor C | 18 |
| Competitor D | 36 |
| Total | 225 |
Your share of voice is 45 ÷ 225 = 20%. Competitor A holds 40%. That's the gap.
Citation rate. Your pages are cited in 18 of 180 answers, a 10% citation rate. Fourteen of those 18 cite one page: a case study on a hospital system CFO search. Your service pages get cited zero times. That tells you exactly where to work.
Accuracy. Across the identity and proof prompts, the AI makes 40 factual claims about your firm. 30 are correct. Accuracy is 75%. Six of the ten errors list a founding partner who retired in 2021.
Pipeline. Intake answers show two new opportunities this quarter where the buyer said they "asked ChatGPT for healthcare search firms." One more shows up in GA4 from perplexity.ai.
Now you have a plan. Update the leadership page and every directory listing that still names the retired partner. Rewrite the service pages so each one makes a specific, liftable claim, modeled on the case study that's already working. Re-run in 30 days and compare. That's the loop.
Common measurement mistakes
Reporting rank. Position in an AI list moves every run. Report frequency.
Changing the prompt set. Change the prompts and you've reset your baseline.
Testing logged in. Memory and past chats bend the answers toward you. Use a private session or turn memory off.
Only testing ChatGPT. Your buyers use more than one tool. Responsive's 2025 buyer research found generative AI has overtaken traditional search for a quarter of B2B buyers, and nearly two-thirds use it as much as or more than search when researching vendors. Cover the engines they use.
AI visibility trackers exist and can automate the runs if your prompt set grows large. For most founder-led firms, a spreadsheet and one disciplined hour a month gets you most of the value.
Not sure what to put in your prompt set? Start with the three-minute prompt test. If the vocabulary is still fuzzy, GEO vs SEO vs AEO sorts it out. And when you're ready to move the numbers, the generative engine optimization playbook walks through the CITE Framework letter by letter.
Frequently asked questions
How do you measure AI visibility? Run a fixed set of buyer prompts, at least three runs per prompt per engine each month, across ChatGPT, Claude, and Perplexity, then score five things: mention rate, share of voice against named competitors, citation rate, accuracy, and AI-referred pipeline. Track the trend month over month. A single answer tells you almost nothing because AI answers change from run to run.
What is a good AI mention rate for a B2B service firm? There is no reliable industry benchmark yet, so compare yourself to yourself and to your named competitors. A rising mention rate on the prompts your buyers actually ask is the goal. If a competitor shows up in most answers and you show up in a few, that gap is your real benchmark.
How do I track ChatGPT traffic in Google Analytics 4? Create a custom channel group in GA4 with an AI channel that matches sources like chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot. Place it above Referral so those sessions get their own row. Some AI visits arrive with no referrer and land in Direct, so pair GA4 with an intake question.
Do I need a paid tool to measure AI visibility? No. A spreadsheet, a fixed prompt set, and a monthly hour of running prompts will get a small firm a usable scorecard. AI visibility trackers exist and can automate the runs at scale, but the metrics and the discipline matter more than the software you use to collect them.
Why does AI-referred pipeline matter more than AI traffic? Because AI answers send fewer clicks, and the clicks they do send tend to arrive further along. Pew found users clicked a link inside a Google AI summary in just 1% of visits, while Ahrefs reported its small slice of AI visitors converted at a far higher rate than organic. Opportunities are what pay the bills, so tie AI mentions to opportunities.
Want help building your scorecard and reading what it says?