In short: traffic from language models (ChatGPT with browsing, Perplexity, Gemini) shows up in GA4 as referral from specific domains (chatgpt.com, perplexity.ai, gemini.google.com). The base setup is a custom segment plus a dedicated dashboard. On the Google side the most solid data point is the click collapse when an AI Overview appears: -34.5% CTR on the top result (Ahrefs, April 2025), up to -58% in the February 2026 update. Absolute click volume from assistants is still lower than SEO, but it is growing fast. For tracking in custom GPTs and MCP, dedicated UTM parameters (utm_source=chatgpt, utm_medium=ai_assistant) enable precise attribution.
How does language-model traffic arrive?
There are three patterns of assistant traffic to websites in 2026.
Clicks on the Sources panel (Perplexity, Gemini). The user sees the generated answer and clicks the links of the cited sources. HTTP source visible in analytics: perplexity.ai, gemini.google.com.
Clicks on links in a ChatGPT browsing answer. ChatGPT with web browsing active shows links inside the answer. Source: chatgpt.com.
Deep links from custom GPTs or MCP. Specialised custom GPTs can include links to specific domains as part of a structured answer, and tracking requires dedicated UTMs.
On relative volume in 2026, Perplexity generates the largest share of assistant-to-site traffic, followed by Gemini and ChatGPT with browsing. Absolute numbers depend on the query: for high-volume informational queries with a high citation rate, tens to hundreds of clicks per month are common; for niche queries, tens.
GA4: how do you set the base source filter?
Step 1: identify the language-model source domains. 2026 list.
Source: Google Analytics 4 default classification (2026).
Step 2: in GA4, create a custom audience segment with the filter Source contains the domain list, and save it as "LLM Traffic". Step 3: create a custom report with primary dimension Source/Medium and the segment you just built. Step 4: add the report to the collection visible in the sidebar for quick access.
How do you define a custom "AI Search" channel?
The more advanced step creates a custom channel in GA4 channel groupings, under Admin, Data display, Channel groups.
The rule to define is: Source matches the regex (chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|you\.com|poe\.com|claude\.ai), with Channel "AI Search". The effect is that in GA4 standard reports assistant traffic appears as a separate channel, distinct from Organic Search and Direct. The step-by-step procedure is in the Google Analytics 4 Help.
UTM strategy for custom GPTs and MCP
When you build a custom GPT with knowledge files that include links to your site, or MCP servers that return URLs, use dedicated UTMs to track them separately.
https://www.example.com/blog/article
?utm_source=chatgpt
&utm_medium=ai_assistant
&utm_campaign=custom_gpt_marketing_helper
&utm_content=article_linkIn GA4 the traffic from the custom GPT with these UTMs appears as Source chatgpt, Medium ai_assistant, Campaign custom_gpt_marketing_helper, and lets you distinguish organic traffic from directly controlled custom GPTs. Note: UTMs work even if added after the knowledge file is indexed, because the custom GPT serves the links it has indexed and updating the file updates the links served afterward.
Plausible and Matomo: the alternatives
Privacy-friendly tools have a simpler setup than GA4 because they see the source natively. In Plausible the Sources section automatically shows perplexity.ai, chatgpt.com and the like, with no custom filters. In Matomo a segment for Source containing "perplexity.ai" or equivalent is enough. The advantage is no sampling, complete cookieless data and first-party attribution; the limit is less cross-channel attribution with a complex paid and organic mix.
How much is a language-model click worth?
The most solid data point is on the Google side: when an AI Overview appears, the click on the top result collapses. Ahrefs (April 2025) measured -34.5% CTR on the top result; the February 2026 update takes the drop up to 58%. This is the phenomenon Ahrefs calls the great decoupling, impressions up and clicks down.
On the assistant side public click-rate data is still fragmentary. The figures below are industry estimates, not a certified number, and should be taken as an order of magnitude to verify against your own data. We say this from experience: when a client asks what a Perplexity click is worth, the honest answer is that it depends on the sector and has to be measured, not copied from a benchmark.
Source: Ahrefs (2025-2026) for the Google figure; industry estimates for assistants.
On post-click conversion the data is limited, but some analyses suggest Perplexity traffic converts better than Google organic, probably because the user arrives with intent already refined by the assistant's answer. It is a reasonable hypothesis, not a law, and should be verified against your own funnel. On absolute value, for most brands in 2026 assistant traffic is still 1-5% of total organic traffic, with fast growth but a volume that stays below classic SEO. On why it is worth covering from the start, see the pillar how to appear on ChatGPT, Perplexity and Gemini and the reference academic research (GEO: Generative Engine Optimization, Aggarwal et al., Princeton/IIT, KDD 2024).
Custom GA4 dashboard: what to include
A useful layout for weekly monitoring keeps a few readable cards together. The first shows LLM Traffic sessions over the last 30 days against the previous 30. The second breaks traffic down by source (Perplexity, ChatGPT, Gemini). The third lists the top landing pages from assistant traffic. The fourth compares conversions between LLM traffic and Organic Search. The fifth watches bounce rate and engagement rate against the site average, while the sixth follows the monthly trend over the last six months to tell growth from stagnation.
Common mistakes
Confusing assistant traffic with Organic Search. By default GA4 classifies chatgpt.com as referral, not organic search, so without a custom channel grouping the traffic stays indistinct from other referrers.
Ignoring low-volume data. Sessions at 1-5% of the total seem ignorable, but with strong growth the data will be relevant in 12-18 months, and tracking it from the start builds a baseline.
Not tracking separate conversions. If assistant traffic converts differently from SEO, the difference does not emerge without a per-source breakdown and dilutes into the mix.
Expecting precise attribution. ChatGPT with memory active can open a link without a source, making the traffic direct instead of assistant, with a 10-30% under-count of the real total.
FAQ
Can I track assistant traffic in Search Console?
No, not directly. Search Console tracks only Google search traffic. For assistant traffic you need GA4, Plausible, Matomo or a dedicated analytics tool.
Do custom GPTs with web browsing see my UTMs?
Yes. ChatGPT with browsing reproduces exactly the link the user clicked, so if the link in the knowledge file contains UTMs the click inherits them. To check, monitor Source chatgpt and your Campaign.
Does Perplexity always report the correct source?
Yes in most browsers. The exceptions are users with privacy extensions that strip the source, and in those cases the traffic appears as direct, with an estimated 5-10% under-count.
Should I create separate conversion goals for assistant traffic?
Not separate. Use the site's existing goals (lead form, signup, purchase) but compare them by channel and source, so the conversion-rate difference emerges naturally in the breakdown.
How much assistant traffic is reasonable to expect in 2026?
For a brand with an optimised GEO presence, 1-5% of total organic traffic is a realistic reference; for a non-optimised brand, under 1%. The range is wide and depends on niche and baseline SEO maturity.
Sources and references
- Aggarwal, P. et al. (2024). GEO: Generative Engine Optimization. KDD 2024 (Princeton/IIT).
- Ahrefs (2025). AI Overviews Reduce Clicks by 34.5%. Ahrefs.
- Ahrefs (2026). Update: AI Overviews Reduce Clicks by 58%. Ahrefs.
- Ahrefs. The Great Decoupling. Ahrefs.
- Google Analytics 4. Help (channel groupings and custom dimensions). Google.
- Plausible. Referrer tracking documentation. Plausible.
- Matomo. Visitor segmentation and source tracking. Matomo.
