Traffic from ChatGPT, Perplexity, and Gemini in GA4 – How Do You Measure It and Assess Its Value?
Traffic from ChatGPT in GA4 has stopped being an analytics curiosity. Since May 2026, Google Analytics 4 has had a separate “AI Assistant” channel that organizes part of the visits from AI assistants. This doesn’t mean, however, that the attribution problem has been solved. Some visits still land in Referral or Direct, and traffic from Google AI Overviews and AI Mode stays in Organic Search. That’s why effective measurement requires combining the native GA4 channel, a custom channel group, landing page analysis, and separate monitoring of pre-click visibility.
TL;DR
- GA4 has a native AI Assistant channel that covers traffic from assistants such as ChatGPT, Gemini, DeepSeek, Copilot, or Grok when the classification conditions are met.
- AI Assistant doesn’t include Google AI Overviews and AI Mode – Google classifies traffic from these modules as Organic Search.
- A custom channel group still makes sense, because it lets you capture LLM traffic more broadly, e.g. from Perplexity, Claude, Copilot, DeepSeek, or other sources, and analyze the data by your own rules.
- GA4 measures only the post-click layer. If a brand appears in an AI answer but the user doesn’t click a link, GA4 won’t show it.
- The value of AI traffic is assessed by session quality, that is, conversions, engagement, landing pages, query types, and the channel’s share in user journeys.
Why does traffic from AI assistants require a separate approach in GA4?
Traffic from ChatGPT, Perplexity, Gemini, or Claude isn’t bot traffic. These are real users who click a link generated or cited in an AI assistant’s answer. The difference is that their path looks different from a classic visit from Google.
In traditional SEO, the user sees a list of results, compares titles and descriptions, and then clicks a chosen link. In conversational search, they first get a synthesized answer. Only later, if they want to verify the source, check the offer, or go deeper, do they click a link. Such a user often reaches the site with a more defined intent than someone clicking a standard organic result.
LLM traffic is visits from users who reached the site after clicking a link in a language model’s or AI assistant’s answer. In GA4 it may appear as AI Assistant, Referral, Organic Search, or Direct – depending on the source, the referrer, and how the link was opened.
For an analyst, this means two things. First, AI traffic should be separated out, or it gets lost in the reports. Second, it shouldn’t be judged by volume alone. Small session counts can carry high value if users reach product pages, comparisons, price lists, contact forms, or BOFU content.
How does GA4 classify traffic from ChatGPT, Gemini, and other assistants?
Google Analytics 4 introduced the AI Assistant channel in the default channel group. Google describes it as the channel through which users reach the site from sources such as ChatGPT, Gemini, DeepSeek, Copilot, or Grok. In the channel definition, the session medium must exactly match the value `ai-assistant`, and the medium and campaign are set automatically when the referrer matches the list of recognized AI assistants.
In practice, the process looks like this:
1. The user clicks a link in a response from ChatGPT, Gemini, Claude, Copilot, or another assistant.
2. The browser passes a referring header to the site, e.g. `chatgpt.com` or `gemini.google.com`.
3. GA4 compares the referrer with the list of recognized sources.
4. If there’s a match, the session may go to the AI Assistant channel, with the medium `ai-assistant` and the campaign `(ai-assistant)`.
This is a good starting point, but not the full picture. GA4 doesn’t let you edit the default channel group. You can, however, create custom channel groups and use them in acquisition reports, explorations, and custom reports. That’s still useful, because AI sources change faster than GA4’s default rules.
AI Assistant vs Organic Search vs Referral vs Direct
| Classification in GA4 | When it may appear | What it means for analysis |
| AI Assistant | GA4 recognizes the AI assistant referrer and assigns the medium `ai-assistant` | The cleanest signal of visits from chatbots and AI assistants |
| Organic Search | The user comes from Google Search, including AI Overviews or AI Mode | Doesn’t automatically separate clicks from Google’s AI modules from classic organic results |
| Referral | The source passes a referrer, but GA4 doesn’t assign it to AI Assistant | Requires your own segment or channel group |
| Direct | No referrer, e.g. a mobile app or manually copying the link | May hide part of the AI traffic, but it can’t be reliably attributed without additional signals |
The most common mistake is treating all AI traffic as a single channel. ChatGPT, Perplexity, Gemini, and Google AI Overviews work differently. Perplexity more often shows sources as footnotes. ChatGPT can generate clicks from the app and the browser. Gemini and AI Mode are tied to the Google ecosystem, but in GA4 traffic from the Google AI Overviews and AI Mode modules is part of Organic Search, not AI Assistant.
How do you set up a custom channel group for AI traffic?
A custom channel group lets you keep more control over reporting. It’s especially useful when some AI visits fall into Referral, or when you want to analyze a broader set of sources than those GA4 recognizes natively.
The path in GA4:
1. Go to Admin.
2. Go to Data display → Channel groups.
3. Create a new channel group based on the default group.
4. Add a channel, e.g. `AI Search & Chatbots`.
5. Set the condition: Session source matches a regular expression.
6. Add a second OR condition: Session medium exactly matches `ai-assistant`.
7. Move the AI channel above Referral. In many configurations it’s good to also place it above Organic Search, but this needs care so you don’t mix classic Google visits with clicks from AI modules.
8. Save the group and use it in acquisition reports and explorations.
A sample regex to test:
.*chatgpt\.com.*|.*openai\.com.*|.*perplexity.*|.*gemini\.google\.com.*|
.*claude\.ai.*|.*copilot\.microsoft\.com.*|.*edgepilot.*|.*edgeservices.*|
.*deepseek\.com.*|.*grok\.x\.com.*|.*writesonic\.com.*|.*copy\.ai.*|
.*iask\.ai.*|.*nimble\.ai.*
This regex isn’t a closed list. AI sources change domains, apps, redirect mechanisms, and the way they pass the referrer. That’s why once a month it’s good to check the Session source / medium report and add new domains that look like traffic from AI assistants.
Custom Channel Group is your own rule-based channel group in GA4. It lets you create traffic categories tailored to your company’s analysis, e.g. a separate channel for ChatGPT, Perplexity, Gemini, Claude, and other assistants.
Why you still won’t see the full AI traffic?
GA4 measures clicks but doesn’t see the brand’s full exposure in AI. That’s a fundamental limitation. If a user asks ChatGPT about the best tools, reads the answer, and doesn’t click any link, GA4 won’t record the interaction. Similarly, if a user copies an address from an AI answer and pastes it manually into the browser, the session may land in Direct.
The most important measurement limitations:
- no referrer – mobile apps and in-app browsers may strip the source information,
- no prompt data – GA4 won’t show the exact question the user typed into ChatGPT or Perplexity,
- no pre-click visibility – GA4 doesn’t measure whether the brand was mentioned in an AI answer without a click,
- different source classifications – the same tool may appear once as AI Assistant and another time as Referral or Direct,
- Google AI Overviews and AI Mode – in GA4 they are classified as Organic Search, so they need to be interpreted together with Google Search Console.
This doesn’t rule out GA4. It only means that GA4 shows the post-click layer, that is, the user’s behavior after reaching the site. To measure pre-click visibility, you need separate GEO/AEO tools or manual prompt tests.
How do you prepare an AI traffic report in GA4?
The most practical place for analysis is Explore. Standard reports show the general picture, but explorations let you combine source, channel, landing page, and conversions.
Report configuration:
1. Go to Explore and create a Free form report.
2. Add dimensions:
– Session default channel group,
– custom channel group, e.g. `AI Search & Chatbots`,
– Session source / medium,
– Landing page,
– Session campaign.
3. Add metrics:
– Sessions,
– Active sessions,
– Engagement rate,
– Average engagement time,
– Key events,
– Session conversion rate,
– Revenue, if e-commerce is implemented.
4. Set a filter: channel group contains `AI Assistant` or the custom channel `AI Search & Chatbots`.
5. Compare the data with Organic Search, Referral, and Direct.
It’s worth adding a second view based on landing pages. That’s what shows which content AI assistants most often cite as sources: guides, definitions, rankings, case studies, service pages, product categories, or blog posts.
How do you assess the business value of traffic from ChatGPT, Perplexity, and Gemini?
AI traffic is usually smaller than organic traffic from Google, so the session count alone can be misleading. It’s better to treat it as a quality channel: fewer visits, but often with stronger intent. The assessment should combine data on volume, engagement, and conversions.
| Metric | What it measures | How/where to measure | How to interpret |
| AI sessions | The volume of visits from AI assistants | GA4: Traffic acquisition / Exploration, the AI Assistant channel or a custom channel group | Shows the scale of the channel, but isn’t enough to assess value |
| AI share of total traffic | The channel’s importance relative to other sources | AI sessions / all sessions | Worth watching the month-to-month trend, not just the baseline level |
| Engagement rate | The quality of visits after a click from AI | GA4: Engagement rate, average engagement time | Higher engagement may suggest a better match of content to intent |
| Conversions / key events | The real business effect | GA4: key events, forms, phone/email clicks, purchases | The most important metric when assessing ROI from GEO and content |
| AI landing pages | The topics most often recommended by assistants | GA4: Landing page + the AI channel | Helps identify content that AI considers citable or useful |
| Revenue / leads from AI | The financial value of visits from assistants | GA4 e-commerce, CRM, lead scoring | Lets you compare AI with SEO, Ads, and quality referrals |
In practice, a good report doesn’t say: “we have 80 sessions from ChatGPT.” A better report says: “AI traffic accounts for 0.7% of sessions but generates 2.4% of leads, most often leads to three BOFU guides, and has a higher engagement rate than standard Organic Search.”
How do you combine GA4 with pre-click visibility measurement?
GA4 answers the question: what did the user do after reaching the site? It doesn’t answer the question: how often does the brand appear in answers from ChatGPT, Gemini, or Perplexity. For that, you need a second measurement layer.
You can build it in three ways:
1. A manual prompt set – regularly check 20-50 branded, comparative, and commercial queries. Record whether the brand appears, in what position, with what context, and with what sources.
2. GEO/AEO tools – use solutions that monitor brand visibility in AI, citations, share in answers, and sentiment. This layer can later be cross-referenced with GA4 data.
3. Server log analysis – check the activity of AI crawlers and search bots, distinguishing bot traffic from real user sessions.
Only combining this data gives a fuller picture. A brand may have good visibility in AI answers but a low CTR. It may also have few citations but AI traffic that converts very well. Each of these scenarios requires a different response.
What should you optimize if AI traffic is to grow?
Measurement is only the first step. If GA4 shows that AI assistants direct users to specific topics, it’s worth expanding those subpages. If traffic comes mainly from one tool, you can check which sources and content types it cites. If traffic falls into Direct, it helps to strengthen indirect analysis, e.g. through landing pages, logs, and trend comparisons.
Actions that most often improve the quality of AI traffic:
- creating Q&A sections with unambiguous answers,
- updating dates, examples, price lists, data, and definitions,
- adding comparison tables and short summaries,
- strengthening authorship and content credibility,
- organizing structured data,
- improving the accessibility of key subpages for crawlers,
- building external mentions and citations that models can treat as confirmation.
The `llms.txt` file can be treated as an extra layer for organizing information for selected AI systems, but not as a guarantee of traffic growth in GA4. For Google, the standard fundamentals remain the most important: indexability, content quality, site structure, organized data, and alignment with user intent.
How do you measure the effectiveness of AI traffic?
The best approach is to create a simple monthly dashboard. It doesn’t have to be elaborate. What matters is that it consistently compares AI Assistant, your custom AI channel, Organic Search, Referral, and Direct.
| KPI | What it measures | Where to measure | Target value format |
| AI Sessions | The number of sessions from the AI Assistant channel and your custom AI group | GA4 Traffic acquisition / Exploration | m/m trend and share of total traffic |
| AI Conversion Rate | The effectiveness of AI traffic in generating leads or purchases | GA4 key events / e-commerce | comparison with Organic Search and Referral |
| AI Landing Page Value | Which subpages generate AI visits and conversions | GA4: Landing page + the AI channel | list of top URLs + conversions |
| AI Engagement Quality | The quality of sessions after a click from an assistant | Engagement rate, average engagement time | difference relative to the site average |
| AI Visibility vs AI Traffic | The relationship between visibility in AI answers and clicks | prompt monitoring + GA4 | growth in mentions, citations, and sessions |
A report should end with a decision, not just a table. For example: “expand the three pages that already generate traffic from Perplexity,” “investigate the drop in clicks despite a rise in citations,” “add a separate segment for Copilot,” “compare AI Assistant with Direct traffic on the pages AI most often recommends.”
The most common mistakes in measuring AI traffic
The first mistake is assuming that after the AI Assistant channel was introduced, your own rules are no longer needed. They are needed, because the default classification doesn’t always cover all sources and scenarios.
The second mistake is mixing AI Overviews with chatbot traffic. In GA4, AI Overviews and AI Mode belong to Organic Search, so analyzing them also requires Google Search Console and observing conversational queries.
The third mistake is judging the channel solely by the number of sessions. In AI traffic, quality often matters more: conversions, engagement, and the type of landing pages.
The fourth mistake is not updating the regex. The list of sources should evolve with the market. If a new tool, domain, or referrer variant appears, the rules in GA4 need to be updated.
The fifth mistake is confusing AI crawlers with users. `GPTBot`, `Google-Extended`, and other bots aren’t user sessions in GA4. They’re visible rather in server logs. Traffic in GA4 appears only when a person clicks a link and reaches the site.
Traffic from Google Search is classified as Organic Search. Traffic from ChatGPT may be classified as AI Assistant if GA4 recognizes the source and assigns the medium `ai-assistant`. If the referrer isn’t passed or isn’t recognized, such a session may end up in Referral or Direct.
No. Google classifies clicks from AI Overviews and AI Mode as Organic Search, not AI Assistant. That’s why analyzing visibility in Google’s AI modules requires combining GA4 data with Google Search Console and query monitoring.
Yes. The native AI Assistant channel organizes part of the data, but a custom channel group lets you add a broader set of sources, test regexes, analyze Perplexity or other tools, and apply your own reporting logic.
The most common reasons are a missing referrer, a mobile app, an in-app browser, manual copying of the link, or the visit being classified as Direct. GA4 also won’t show brand exposure in an AI answer if the user didn’t click a link.
Not directly in GA4. You can analyze landing pages, content types, traffic sources, and your own prompt monitoring, but the specific questions users type into ChatGPT or Perplexity aren’t passed to GA4.
At least once a month, or whenever there are larger changes in the traffic source reports. A good habit is to review new values in the `Session source / medium` dimension and add domains that look like AI tools or chatbots.
Sources
- Google Analytics Help – Default channel group: https://support.google.com/analytics/answer/9756891
- Google Analytics Help – Custom channel groups: https://support.google.com/analytics/answer/13051316
- Google Analytics Help – What’s new in Google Analytics, May 13, 2026: https://support.google.com/analytics/answer/9164320
- Google Search Central – Optimizing your website for generative AI features on Google Search: https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- OpenAI Platform Docs – OpenAI crawlers and user agents: https://platform.openai.com/docs/bots