AI Visibility Tracking: What to Measure and How to Do It

On top of that, you add the data Google Search Console, Bing Webmaster Tools, and your web analytics already provide.

There's no "rank" in an AI answer, and the same question can get two different answers back to back. So what counts isn't a screenshot; it's a data series over time. This guide covers what to measure, where each number comes from, and a weekly method you can start today.

What to measure: the five core metrics

Two more worth logging:

• Cited sources. Which domains the engine uses to answer questions in your space. If it keeps citing the same directory or publication, that's where you need to show up.
• Accuracy. If the AI says something false about you (old prices, services you don't offer), that's a problem to fix at the source.

Mentions and citations aren't the same thing. ChatGPT may recommend your brand without linking to you, or link to your article without naming you in the text. Track both.

Where the data comes from

1. Asking the engines (your primary data)

This is the only way to know what ChatGPT, Perplexity, or Gemini actually say, because none of the three currently gives publishers a citation report like Google's or Bing's. You can do it by hand with a spreadsheet, or with a tool that sends the questions through each API and stores the results.

2. Google Search Console: the generative AI report

Search Console has a Generative AI performance report showing your site's impressions in AI Overviews and AI Mode, by page, country, date, and device. Google says it had rolled out to all websites as of August 31, 2026¹.

Its limits: it measures link impressions, not unlinked brand mentions; sites with too few impressions may not see it; and it tells you nothing about ChatGPT or Perplexity. Google points to this report in its official guide to generative AI in Search, and warns that no third-party tool has access to its internal ranking systems².

3. Bing Webmaster Tools: AI Performance

In February 2026 Bing launched an AI Performance report in public preview, covering your citations in Copilot and Bing's AI summaries: total citations, cited pages, and the queries behind them. In June 2026 it added citation share against other sites. Bing notes the data is sampled and that a citation doesn't equal a visit³⁴.

4. Web analytics: traffic from AI

Visits from AI assistants show up in Google Analytics 4 as referrals. ChatGPT also usually adds utm_source=chatgpt.com to its links (not always: it stopped for a week in October 2026), per OpenAI's publisher FAQ. To see it all in one place, create a custom channel group in GA4 called "AI" with a source rule that matches a regex like this⁵:

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

One catch: clicks from AI Overviews and AI Mode can't be separated this way. They arrive as Google organic traffic and are counted in Search Console's regular performance report.

5. Server logs

Logs show which AI crawlers visit your pages and how often. They don't measure visibility, but they warn you if an engine stops crawling you. More in our guide to AI crawlers and robots.txt.

A weekly method, step by step

1. Write 20 to 50 real questions. Phrase them the way a customer would, with context: "best invoicing software for freelancers in the US," "cheaper alternatives to [competitor]." Pull them from sales, support, Search Console, and forums.
2. Group them by intent. Discovery ("what is…"), comparison ("X vs Y"), purchase ("best… for…"), and brand ("reviews of…").
3. Pick 3 to 5 competitors. Without them you can't calculate share of voice.
4. Run each question in each engine, several times. Answers vary; run it once and you'll mistake noise for signal.
5. Log it in a table: date, engine, question, mentioned or not, cited or not, position, sentiment, brands that appear, and sources cited.
6. Repeat weekly with the same questions. Changing the questions breaks the series.
7. Look at 4- to 8-week trends, not a single day's number. Cross-check against what you've published or changed.

A sample log

Here's what one week might look like for a single question, run three times in each engine (made-up numbers, just to show the format):

Read this way, the priority is obvious: figure out why Gemini never names you and which sources it uses to name your competitors.

How to read the results

• Small ups and downs are normal. Models get updated, indexes change, and answers have a random element. Look for changes that hold for several weeks.
• Compare engines separately. You can do well in Perplexity and badly in ChatGPT. The causes are usually different: a blocked crawler, a different index, different sources.
• Questions a competitor wins are your to-do list. Look at which sources the engine cites in those answers and what those pages have that yours don't.
• An AI error about your brand is urgent. Fix the information on your site and in the sources it cites, then measure again.

Common measurement mistakes

• Testing from your personal account and history, which personalize the answer. Use clean sessions or the API.
• Blending different engines into a single number.
• Trusting tools that promise "rankings" in ChatGPT. There are no rankings, only odds of showing up.
• Ignoring competitors: a 10% gain doesn't mean much if your competitor doubled.

Skip the spreadsheet

The method above works by hand, but it's a lot of questions, engines, and weeks. That's why a GEO Plus Academy subscription includes an AI visibility tool: you enter your brand, your competitors, and up to 25 questions, and every week it checks ChatGPT, Perplexity, and Gemini and shows you mentions, citations, and trends by engine. It's the same tool we use in the course's measurement module.

If you'd first like to see whether a page is ready to be cited, start with our free GEO audit: it checks robots.txt, llms.txt, structured data, and content in a single pass.