Measuring AI traffic: spotting visitors and leads from ChatGPT, Perplexity & co.

More and more people no longer ask Google but ChatGPT, Perplexity or Gemini: “Which agency can help me with …?” According to Bitkom (September 2026), 57 percent of German companies now use AI – so whether your website shows up in these answers is no longer a gimmick. The good news: you can measure a large part of it with built-in tools. The bad news: not all of it.

This guide shows you step by step how to make AI traffic visible in Google Analytics 4, attribute enquiries to the right source and find out whether AI systems read your pages at all.

Can you measure how much traffic comes via ChatGPT and other AI assistants?

Yes – with limitations. When someone in ChatGPT, Perplexity, Gemini, Microsoft Copilot or Claude clicks a link to your website, the browser passes on the origin in most cases. In Google Analytics 4 the visit then shows up as a referral, for example with the source chatgpt.com or perplexity.ai. ChatGPT also appends the parameter utm_source=chatgpt.com to many links, which makes attribution even more reliable.

What you do not see directly:

• Clicks from apps without a referrer: If the visit comes from a desktop or smartphone app, the source is sometimes missing. The visit then lands under “Direct”.
• Google AI Overviews and AI Mode: Google Analytics counts clicks from these as perfectly normal organic Google traffic.
• Mentions without a click: If ChatGPT recommends your agency and the user later googles you directly, it looks like a brand search.
• Visitors without consent: Anyone who declines in the cookie banner does not show up in GA4 at all.

So the GA4 numbers are a lower bound. For trends – is the AI share growing, which pages get linked – they are still entirely sufficient.

How to set up measurement in Google Analytics 4

By default, GA4 assigns AI visits to the “Referral” channel, where they get lost between business directories and partner sites. A dedicated channel is better.

Step 1: Create a custom channel group “AI assistants”

1. In GA4, bottom left, go to Admin → Data display → Channel groups.
2. Create new channel group (the default group cannot be edited, but a copy can).
3. Add new channel, name it, e.g. KI-Assistenten or simply “AI assistants”.
4. Condition: Source → matches regex →

.*(chatgpt|openai|perplexity|gemini\.google|copilot|claude\.ai|chat\.mistral|deepseek).*

5. Move the new channel above “Referral” in the order – GA4 checks the channels from top to bottom, otherwise “Referral” matches first.
6. Save. Custom channel groups also apply retroactively, so you see historical data right away.

Then, in Reports → Acquisition → Traffic acquisition, select your new channel group from the dropdown in the first column.

Step 2: Which pages get linked from AI answers?

The more interesting question is often not “how much” but “which pages”. Under Explore → Free form, create a table for this: rows = Landing page, columns = Session source, filter = session source matches the same regex. This shows you which guides or service pages ChatGPT & co. use as a source – and where it pays to add more.

How many leads come via ChatGPT?

Visitors are nice, enquiries are what counts. For that you need two building blocks:

1. Key events in GA4. Mark the form submission, the appointment booking or the click on your phone number as a key event (formerly “conversion”). In the “Traffic acquisition” report with your AI channel group you then see per channel how many enquiries were generated.

2. Store the source directly with the enquiry. GA4 shows you totals, but not which enquiry came from ChatGPT. That is why the website stores the origin (referrer and UTM parameters) on the first visit and passes it on with the enquiry via a hidden form field. Your CRM or the enquiry email then says “Source: chatgpt.com” right away. Technically this takes little effort and also works when someone only gets in touch on their third visit.

A thoroughly analogue trick helps as well: a required field “How did you hear about us?” with the option “ChatGPT / AI assistant”. It catches exactly the cases no software sees – for example when someone types in the recommendation from ChatGPT by hand.

Checking AI crawlers and AI visibility

Before an AI can recommend you, it has to read your pages. You can see this in your host’s server logs by the user agent. The difference between the bot types matters:

• OpenAI: GPTBot (training), OAI-SearchBot (search index for ChatGPT), ChatGPT-User (fetch for a user question)
• Anthropic: ClaudeBot (training), Claude-SearchBot (search index), Claude-User (fetch for a user question)
• Perplexity: PerplexityBot (search index), Perplexity-User (fetch for a user question)

The fetches for a user question are the most revealing: a fetch by ChatGPT-User means that someone has just asked a question and ChatGPT is reading your page live to answer it. Also check your robots.txt – if you block OAI-SearchBot or PerplexityBot, you will not appear in their search results.

For visibility without a click there are now two official sources:

• Google Search Console: Appearances in AI Overviews and AI Mode are included in the normal performance report, without a separate filter. Since June 2026 Google has also been rolling out a separate report for generative AI, initially only for some websites. Check whether it already shows up for you.
• Bing Webmaster Tools: The “AI Performance” report shows how often your pages are cited as a source in Copilot and Bing’s AI answers, including the triggering questions. Click numbers are missing there, though.

What we observe on softstorm.de

For a few months now, our own Search Console has been showing strikingly long, fully worded questions, such as a German query along the lines of “how do agencies use AI for status updates and project reports?” – with up to 14 impressions and 0 clicks. Hardly any human types like that into Google. Most likely these are sub-questions that Google’s AI search derives from a user question, or prompts from tools that measure AI visibility. For us these queries are a topic list: they show for which questions an AI is already considering us as a possible source. You can filter them in the performance report via Query → Custom (regex) with ^(wie|was|warum|welche|wann) for German queries, or the English equivalent ^(how|what|why|which|when) for English ones.

My conclusion

Start small: create the channel group in GA4, set a key event for your enquiries, add the hidden source to your form. That takes one afternoon and after four weeks gives you a reliable answer to the question of whether content for AI search is worth it. Where your website stands on loading time and SEO basics is shown by our free website check. If you would rather have the measurement including form tracking implemented for you, take a look at our web development or book a free initial consultation.

Note: For Google Analytics and comparable tracking services you will usually require your visitors’ consent. This is a technical overview. It is not legal advice.