TL;DR
- There is no single number for AI visibility yet. No tool can tell you with full accuracy how often ChatGPT, Gemini or Perplexity mentions your business.
- “Zero AI traffic” in your reports is almost never true. Part of it is hidden in direct traffic, and part never reaches your website at all.
- What works today is a combination: your own prompt tests from neutral accounts, filters in Search Console and Google Analytics, the new AI reports from Google and Microsoft, and paid tools as a second opinion.
- The biggest shift is in thinking: from keyword research to prompt research, built on a real understanding of the business.
- Below: the six methods I use for the brands I work with, ready-to-copy filters and a setup checklist.
Almost every brand I work with asks me the same thing sooner or later: “Search Console shows no traffic from ChatGPT. Does that mean AI doesn’t recommend us?”
My answer is always the same: no. For a large company it is statistically impossible that nobody has asked ChatGPT about it in recent weeks, or bought something after that conversation. The problem isn’t that the traffic doesn’t exist. The problem is that our tools were built for a different kind of search.
The short answer
Why your reports say zero
| What the report shows | What is really happening |
|---|---|
| No traffic from ChatGPT in Analytics | Some visits arrive without a referrer and are counted as direct traffic |
| No traffic before mid-2025 | ChatGPT started tagging more of its links for analytics in June 2025, so older visits are harder to separate |
| Fewer sessions than you expect | In the EU, consent rules mean part of the visits are never recorded in full |
| A tool says you are not visible in AI | It ran a limited set of prompts, from its own locations, on its own accounts |
| Organic traffic is falling | The answer is often given directly in AI Overviews or a chatbot, without a click |
A user can also read a recommendation in ChatGPT, open a new tab and type your name. Your analytics will never connect that visit to the assistant.
Why AI visibility is hard to measure
With classic SEO we aimed for the top three positions for a keyword. Even that was a simplification of a complex ranking system, but at least there was a ranking.
With AI there is no fixed ranking. Every answer is generated for a specific prompt, and prompts are almost never the same. Two customers looking for the same thing will describe it in completely different ways. The assistant also knows their context from previous conversations, so the answer is personalized.
This is why optimizing from a checklist no longer works. To predict how your customers will ask, you have to understand the business itself: its operations, finances, processes, goals and marketing.
The six methods I use
Open each one. No method is perfect on its own, which is why I combine them.
1. Manual prompt tests from neutral accounts
Run your customers' questions from devices and profiles not connected to you, from the right locationUse private windows, separate accounts, colleagues in the target country or a VPN for foreign markets
Most reliable
Especially useful for smaller markets like BulgariaPaid tools often have little data for local-language prompts, so your own tests fill the gap
Free
2. Paid AI visibility tools
Tools such as the Semrush AI Visibility Toolkit or Ahrefs Brand Radar simulate prompts and estimate how often you are mentionedRead how each tool collects its data before you trust the numbers
Second opinion
Bigger studies usually mean paying per prompt runResults are an average across many unknowns: location, account, model version
Paid
3. Search Console: long, prompt-like queries
Filter queries by length to see the conversational onesMany of them come from AI Overviews and AI Mode
Free
The new generative AI report, if your site already has itGoogle announced it in June 2026 and is rolling it out gradually. It shows impressions, not clicks.
Check now
4. Google Analytics: AI referral traffic
Filter session source by AI platforms in the Traffic acquisition reportBetter still, create a custom channel group for AI so it shows up in every report
Free
Remember that this is the minimum, not the full numberDirect traffic and consent rules hide part of it
Undercounts
5. Bing Webmaster Tools: AI Performance
Shows how often your pages are cited in Copilot and Bing's AI answers, and the queries behind the citationsLaunched as a beta in February 2026. No clicks, only citations.
Free
6. Microsoft Clarity: AI Visibility
Citations, share of authority and AI referral traffic in one dashboardCitations became generally available in May 2026. AI bot activity needs a server or CDN integration.
Free
Filters you can copy
Search Console. Go to Performance, add a filter for Query, choose Custom (regex) and paste this. It shows queries of roughly ten or more words:
^(\S+\s+){9,}\S+$
Google Analytics. In Reports, open Acquisition, then Traffic acquisition. Switch the dimension to Session source / medium and filter with a regex that matches the main AI platforms:
chatgpt|openai|perplexity|gemini|copilot|claude
Build a prompt library
Keyword research isn’t enough anymore. In my client work I now add prompt research and keep a prompt library, a simple table that grows every month.
| Step | What to do |
|---|---|
| 1. Choose platforms | Start with two or three: ChatGPT, Gemini and Perplexity, plus Claude or Copilot if your audience uses them |
| 2. Write 20 to 30 prompts | The questions your customers would really ask, in their own words and language |
| 3. Mix the types | Direct, comparison, problem-based and location-based questions |
| 4. Test cleanly | A new chat for every prompt, a private window or neutral account, the right location |
| 5. Record everything | Date, platform, whether you were mentioned, who was mentioned instead, which sources were cited |
| 6. Repeat monthly | Compare the results over time, not single answers |
Don’t limit yourself to the classic search intents. One prompt often covers three or four of them at once: it starts as a question for information, and the answer ends with a recommendation and a link to buy.
For where to find real prompts, the columns to track and a free template, read What Is Prompt Research? How to Find the Questions Customers Ask AI.
How to read the results
One answer means nothing. The same brand appearing in 12 out of 30 prompts, month after month, means a lot.
Note which websites the assistant cites. They show you where you need to be mentioned.
Your share of mentions next to theirs is more useful than any absolute number.
Ask new clients how they found you. More and more of them will say they asked an AI assistant.
Your checklist
Six things to do this week
My take
As someone responsible for the visibility of the brands I work with, my job is to stay as current as the market. That is why I keep testing and comparing methods, and why I don’t trust any single number.
The tools are getting better quickly. Google, Microsoft and the big SEO platforms are all building AI reports, because without them their products lose their purpose. Until the data becomes reliable, the combination above is the most honest picture you can get.
If you measure AI visibility in a different way, I’d like to hear how.
For a neutral overview with data and sources, read How to Measure AI Traffic and AI Visibility, Explained on Egorithms.
About The Author
