How to track your brand in ChatGPT, Gemini and AI Overviews

01Why one report is not enough

AI platforms hide a lot of data. Some do not pass referrer information, so visits show up as "direct". Mentions without links send no traffic at all. And training-data visibility has no analytics. You need several imperfect signals rather than one perfect one.

Ahrefs' testing found Claude passes referrer data properly, ChatGPT passes it for source links but not for some in-answer links, Perplexity passes it on the web but not from its desktop app, Copilot on the web but not in Windows, and Grok does not pass it at all. Treat your AI traffic numbers as an undercount.

02Signal 1: AI referral traffic in GA4

Create a custom channel group in GA4 called "AI" that matches referrers containing chatgpt, openai, perplexity, gemini, copilot, claude and deepseek. Report it alongside Organic Search each month.

Google documents custom channel groups in Analytics Help; the rule is a source condition that matches those domains, placed above Referral so AI visits are not swallowed by it. Expect small numbers: Conductor's 2026 benchmarks put AI referrals at roughly 1% of all web traffic, with ChatGPT responsible for the large majority of it. Then look at conversion rate, not volume. Ahrefs saw AI visitors convert at about 23 times its organic rate.

03Signal 2: AI crawler activity in your logs

Your server or CDN logs show which AI bots read which pages. Training bots (GPTBot, Google-Extended, ClaudeBot) tell you what the models may learn from. Live-fetch bots (ChatGPT-User, OAI-SearchBot, PerplexityBot) tell you which pages are being used to answer questions right now.

OpenAI, Anthropic and Perplexity each publish their crawler names and how to control them. Pages that live-fetch bots request repeatedly are very likely being cited. Pages that no AI bot ever requests are invisible to AI, which is often a robots.txt or JavaScript rendering problem rather than a content problem.

04Signal 3: ask your customers

Add "How did you hear about us?" to your enquiry and checkout forms, with ChatGPT, Gemini, Perplexity and Copilot as options. It captures visits analytics cannot see, including people who were recommended by AI and then typed your name into Google.

Ahrefs reported that about 3% of its conversions selected AI in this question, a share its analytics alone would never have shown. Keep the question optional and short, and report the AI share monthly next to the GA4 number.

05Signal 4: monthly share of voice

Keep a fixed list of 20–50 buyer questions and run them monthly in ChatGPT, Gemini, Perplexity and Google (for AI Overviews). For each, record whether you were mentioned, whether you were cited with a link, whether the facts were correct, and which competitors appeared.

1. 1Mentioned? Your brand is named in the answer.
2. 2Cited? Your site is linked as a source.
3. 3Accurate? Prices, locations, claims and people are correct.
4. 4Who else? Competitors named alongside or instead of you.

Results vary run to run, so use the same prompts, the same accounts and a consistent day of the month. Ahrefs found about 45% of AI Overview citations change on each refresh, so month-to-month movement is normal; look at the trend over a quarter. Dedicated tools can automate this at scale, but a spreadsheet is enough to start.

06Reading the numbers without fooling yourself

AI traffic is small and noisy, so look at rates and trends rather than raw counts. A page that gets 40 AI visits and 6 enquiries is more interesting than one with 4,000 organic visits and 6 enquiries.

Watch for three traps. First, "direct" traffic that is really AI, because some apps strip referrers; a rise in direct visits to deep pages after an AI mention is a clue. Second, mentions without links, which send no traffic but still shape decisions; only the form question and share-of-voice check will catch them. Third, month-to-month volatility; Ahrefs found roughly 45% of AI Overview citations change on refresh, so compare quarters, not months.

07Tools: what you need and what you do not

You can run the whole setup with GA4, your hosting or CDN logs, your existing form tool and a spreadsheet. Dedicated AI visibility tools add scale and automation, which becomes worth paying for once you track more than a few dozen prompts or several competitors.

Start manual. A spreadsheet with your prompt list, a column per platform and a cell per month is enough to see share of voice move, and doing it by hand for a quarter teaches you how volatile the answers are and which prompts actually matter. Once the routine is stable, automate it. Whatever tool you use, keep the method fixed: same prompts, same accounts, same cadence. A change in method looks exactly like a change in visibility.

One more free source: Google Search Console. It does not separate AI Overview impressions from normal ones yet, but a page whose impressions rise while clicks fall is very often being shown inside Overviews. Cross-check those pages with your share-of-voice log.

08Setting a baseline before you change anything

Measure for one full month before the first fix. Without a baseline you cannot show what moved, and the first weeks of GEO work (crawler access, tracking) often produce the clearest jumps.

The baseline month also tells you which platforms matter for your market. If Perplexity sends real enquiries and Copilot sends none, prioritise accordingly. If the form question shows AI influence but GA4 shows almost no AI sessions, you have a mentions-without-links situation and earned coverage should lead the plan. Keep the baseline report; it is the "before" picture every later report is compared with.

09What goes in the monthly report

One page: AI sessions and conversions from GA4, the share of leads who said "AI" on the form, share of voice versus three to five competitors, any wrong facts found and fixed, and what we are doing next month. Readable in five minutes.