How to measure your AI visibility: mentions, citations and share of voice

To measure AI visibility, run a fixed set of real buyer questions through ChatGPT, Google AI Overviews and AI Mode, Perplexity, Gemini, Claude and Copilot, several times each. Record whether you're mentioned, whether your website is cited, and how often competitors appear. Add AI referral traffic from GA4 and AI impressions from Search Console. Track weekly, because one answer proves nothing.

Key takeaways

• AI answers change from run to run. A 2026 SparkToro study found less than a 1-in-100 chance that ChatGPT or Google's AI returned the same brand list twice.
• Build 20 to 50 prompts in four groups: best-of, jobs to be done, competitor and branded. Use your customers' words, not your marketing copy.
• Track five numbers: mention rate, citation rate, share of voice, average position and sentiment. Share of voice is the one to watch over time.
• GA4 now has an AI Assistant channel, but it excludes Google's AI Overviews and AI Mode and misses visits with no referrer.
• Search Console's generative AI report shows impressions only, with AI Overviews and AI Mode combined.

Why can't you just ask ChatGPT once?

Because the answer will be different next time. In research published in January 2026, SparkToro's Rand Fishkin and Gumshoe.ai's Patrick O'Donnell had about 600 volunteers run 12 recommendation prompts 2,961 times across ChatGPT, Claude and Google's AI (Search Engine Journal summary). They found:

• Less than a 1-in-100 chance of getting the same list of brands twice from ChatGPT or Google's AI.
• The same list in the same order roughly once in 1,000 runs.
• How often a brand appeared was far more stable than where it ranked. In ChatGPT's answers about cancer-care hospitals, City of Hope appeared in 69 of 71 answers but was listed first in only 25.

Gumshoe.ai sells AI tracking software and the study wasn't peer-reviewed, but the lesson matches what you'll see testing by hand.

Answers also vary by person. OpenAI's help page on ChatGPT search says ChatGPT may use an approximate location from your IP address, and with Memory turned on, it may use saved details about you when it rewrites your question into a search. An HVAC company owner asking from the office gets a different "best HVAC company near me" answer than a homeowner across town. Model updates shift answers too.

Treat AI visibility like polling: many samples, the same questions, the same method, every week.

How do you build a prompt set from real buyer questions?

Start with what customers actually ask: sales call notes, quote forms, emails, reviews, Search Console queries and front-desk questions.

Sort prompts into four groups:

A few rules:

• Aim for 20 to 50 prompts. Fewer and one odd answer swings your numbers. More and you won't keep up by hand.
• Write like a buyer. "Dentist near me who takes Delta Dental and does same-day crowns" beats "dental services."
• Put the location in the prompt for local businesses, so where you test from matters less.
• Freeze the list. Change prompts only once a quarter, or your trend line means nothing.
• List the competitors you'll count, usually three to eight companies.

How do you run the prompts across AI engines?

Cover ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Microsoft Copilot, Claude and Grok. Short on time? Start with ChatGPT and Google.

To keep results comparable:

1. Use a clean session. Log out, or use a separate account with Memory and personalization off.
2. Keep location fixed. Put the city in the prompt and test from the same place each week.
3. Run each prompt three to five times per engine and record every run, not just the best one.
4. Log the details: date, engine, model if shown, whether search or web browsing was on, and every company named and every URL cited.
5. Save a screenshot or copy of answers that mention you, especially wrong ones.

AI Overviews don't appear on every search. Record "no AI Overview" as a result, and use AI Mode for those prompts.

Which metrics should you track?

Mention and citation are different. ChatGPT might name your dental clinic but cite a Yelp page, or cite your blog post without recommending you. Track both.

Share of voice works best for trends. If you appear in 30 answers and all tracked companies together appear in 150, your share is 20%. If the leader appears in 60, you're at half the leader's visibility.

Don't panic over position. Dropping from third to fifth in one week is mostly noise. Going from 10% of answers to 30% over two months is real.

GrowDesk's Marketing Agent runs this kind of tracking automatically, using the same share-of-voice math and a visibility score relative to the leader. A spreadsheet works fine to start.

How do you see AI referral traffic in GA4?

The AI Assistant channel. In May 2026 Google Analytics added an AI Assistant channel to the default channel group. Google's traffic-source documentation describes it as visitors arriving "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok," and says it "excludes Google's AI Overviews and AI Mode," which count as Organic Search. Google hasn't published the full referrer list, so check whether Perplexity and Claude visits land there or under Referral in your property.

A custom check. In Reports, Acquisition, Traffic acquisition, set the dimension to Session source and filter with a regex such as:

chatgpt\.com|openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai

Or build a custom channel group with that rule placed above Referral.

The ChatGPT tag. OpenAI's publisher FAQ says ChatGPT automatically adds utm_source=chatgpt.com to referral links, so those visits show with source chatgpt.com even when the browser sends no referrer. Your site must allow OAI-SearchBot for ChatGPT search to show it at all (see AI crawlers and robots.txt).

What GA4 can't tell you

• Missing referrers. Visits from apps or copied links can arrive with no referrer and show as Direct.
• Tracking glitches. From about October 8 to 15, 2026, inline links in ChatGPT answers lost the utm tag, and OpenAI reportedly called it a bug. Some sites saw ChatGPT visits fall and Direct rise. Annotate that week in your reports.
• Google's AI features. Clicks from AI Overviews and AI Mode look like ordinary organic search.
• Influence without clicks. A homeowner reads ChatGPT's answer, then Googles your name or just calls. Ask new customers how they found you, and add "ChatGPT or another AI tool" as a form option.

What do Search Console and Bing Webmaster Tools show?

Google Search Console. Per Google's AI features documentation, AI Overviews and AI Mode traffic sits in the regular Performance report under the Web search type, with no filter to separate it. In June 2026 Google added a generative AI performance report, rolled out to all sites by August 31, 2026. It shows how often your URLs appeared in AI features in Search and Discover, by page, country, device and date. Limits: impressions only, no clicks, no prompts, and AI Overviews and AI Mode grouped together.

Bing Webmaster Tools. Microsoft's AI Performance report, in public preview since February 2026, shows how often your pages are cited in Microsoft Copilot, Bing's AI summaries and some partner integrations, plus the "grounding queries" behind them. It's free, the data is a sample, and it doesn't cover ChatGPT or Google.

What does a simple weekly routine look like?

By hand, 20 prompts times three runs on three engines is 180 answers, a few hours a week. A tracking tool can automate the runs.

1. Monday: run the prompts. Same list, same engines, same settings. Log every answer.
2. Score them. Calculate mention rate, citation rate and share of voice per engine and overall.
3. Check traffic. In GA4, note AI Assistant sessions, chatgpt.com sessions and, more important, leads and calls from them.
4. Check Search Console monthly. Note generative AI impressions for your key pages.
5. Read the wrong answers. If AI says you don't serve a town you serve, or lists an old price, fix the source: your site, Business Profile or directory listings.
6. Pick one fix a week. Add a missing page, answer a question in plain text, or update a listing. Our guides on getting recommended by ChatGPT and showing up in Google AI Overviews and AI Mode cover what to change.

Sample tracking table

Here is one week for a hypothetical HVAC company tracking 20 prompts, three runs each, across three engines (180 answers in total). The numbers are made up to show the format.

Every cited answer here also named the company, so it appeared in 54 answers. If all six tracked companies together appeared 270 times, its share of voice is 54 ÷ 270 = 20%. The top competitor, at 96, has about 36%.

Keep one row per engine per week in a spreadsheet, and chart share of voice monthly.

FAQ

How many prompts do I need to track AI visibility?

Twenty to 50 is enough for most small and mid-sized businesses. Run each one three to five times per engine so a single unusual answer doesn't skew results.

Does GA4 show traffic from Google AI Overviews?

Not separately. Google's documentation says the AI Assistant channel excludes AI Overviews and AI Mode, and those clicks count as Organic Search. Search Console's generative AI report shows impressions from them, but not clicks.

What is a good AI share of voice?

There's no universal benchmark. Compare yourself to your own tracked competitors and watch the trend. Gaining ground on the leader over a few months matters more than any single number.

Why does ChatGPT give different answers to the same question?

AI models generate answers with some randomness, and ChatGPT may factor in your approximate location and saved memories. That's why you should run prompts several times in a clean session and track averages.

Sources

• New research: AIs are highly inconsistent when recommending brands or products (SparkToro, January 2026)
• AI recommendations change with nearly every query: SparkToro (Search Engine Journal, January 2026)
• ChatGPT search (OpenAI Help Center)
• Publishers and developers FAQ (OpenAI Help Center)
• Traffic-source dimensions and default channel group (Google Analytics Help)
• ChatGPT traffic drop appears to be a tracking problem (Search Engine Journal, October 2026)
• AI features and your website (Google Search Central)
• Introducing Search generative AI performance reports in Search Console (Google Search Central Blog, June 2026)
• Introducing AI Performance in Bing Webmaster Tools public preview (Microsoft Bing, February 2026)