GA4 AI Referral Traffic: Why Filters Reset After Refresh

Date: November 17, 2026Reading time: 6 min readShare:

You open the Traffic Acquisition report to check your ChatGPT traffic tracking numbers, only to find everything buried under a generic ‘Referral’ label. Or, you set a custom filter to isolate AI referral traffic, but refresh the page and the data is gone.

GA4 AI Referral Traffic: Why Filters Reset After Refresh

This is a common frustration. AI platforms are now a distinct discovery layer, yet GA4’s default reporting doesn’t reflect that reality. You are left choosing between a quick, temporary fix and a permanent architectural change.

The 1-minute workaround: Filtering Session source by regex

Use a report filter on Session source for immediate AI referral traffic insights without altering property settings or waiting for data updates. Apply the “matches regex” condition with chatgpt.com|perplexity.ai|gemini.google.com to capture primary AI discovery layers. Note that GA4 enforces a strict 250-character limit, so this concise regex is essential; adding domains like claude.ai or copilot.microsoft.com quickly approaches the ceiling.

This filter is temporary, resetting whenever you refresh the page or navigate to a different report. It does not save to your property configuration, making it unsuitable for ongoing performance monitoring or historical comparisons. Position this method strictly as a diagnostic tool for specific, one-off questions, such as checking the impact of a recent Perplexity analytics spike. For continuous visibility, this method fails because it lacks permanence. If your team reviews ChatGPT traffic tracking data weekly, the effort of re-applying this filter every session becomes inefficient. The workaround answers “what is happening now?” but not “what is the trend over time?”

Building the ‘Artificial intelligence’ channel group

A report filter vanishes on refresh; a custom channel group remains. This configuration is the only method that makes AI referral traffic permanent across all standard GA4 reports, ensuring consistency in your long-term data analysis.

To implement this, navigate to Admin > Data display > Channel groups. You will add a new channel within your selected group. The name matters more than you might think. Use the full phrase “Artificial intelligence” rather than the abbreviation “AI”. While the difference seems minor, the full name makes filtering easier later when you build custom reports or explore specific datasets. If you use “AI”, you risk conflicts with other data points containing those letters, complicating your Perplexity analytics or other deep dives.

Order and Naming Constraints

When adding the condition, select the Source dimension and use the matches regex match type. Paste the regex string covering your target domains (chatgpt.com, perplexity.ai, gemini.google.com, etc.). But the name is only half the battle. Positioning is critical: the new “Artificial intelligence” channel must appear above the default “Referral” channel in the list. GA4 evaluates rules in sequence. If “Referral” is higher, it catches the traffic first, and your new channel records nothing. This order ensures correct attribution logic.

Be strategic with your setup. As of 2026, GA4 imposes a hard limit: you can create only two custom channel groups per property. One slot is effectively reserved for the Default Channel Group logic you are customizing. This constraint means you cannot clutter your property with ad-hoc groups for every new platform. Choose your configuration carefully. If you need a second custom group, reserve it for a truly distinct business unit or campaign type, not a temporary experiment. This discipline ensures your GA4 setup remains scalable as AI referral traffic evolves.

Comparing the three AI traffic reporting paths in GA4

Path Setup Time Persistence Best Use Case
Report Filter 1 min Temporary (resets on refresh) One-off deep dives and ad-hoc queries
Custom Channel Group 5–10 min Permanent Ongoing performance monitoring in standard reports
Custom Report/Exploration 10–15 min Permanent Shareable, in-depth analysis and specific breakdowns

The report filter is a quick diagnostic, but it vanishes the moment you leave the page. For teams that need to share AI traffic breakdowns without granting everyone access to full property settings, the custom report serves as a practical bridge. It isolates the data so stakeholders can see the ChatGPT traffic tracking results without needing admin permissions to modify the underlying channel logic. The channel group provides the foundation, but the custom report controls the presentation layer, ensuring that sensitive property settings remain locked down while the insights remain transparent.

Explorations for Granular Detail

Explorations represent the next level of detail in your analytics stack. When you convert a standard report into an exploration, each report element becomes an editable tab. This structure allows you to create specific views for Perplexity analytics or Gemini performance, separating them from the aggregate AI referral traffic figure. You can build a tab that filters exclusively for the perplexity.ai source, giving you a dedicated workspace to analyze engagement and conversion metrics for that single source without cluttering your main reporting suite.

Choosing the Right Path

The decision between these three methods comes down to frequency and audience. If you look at the data once to answer a specific question, use the report filter. It is fast and requires no configuration beyond a regex string. However, if you check these metrics weekly or need to report on them to stakeholders, the custom channel group is the necessary architectural step. It ensures that the data is consistent across all your standard reports, eliminating the need to re-apply filters every time you open a new view. The custom report is best reserved for when you need to share specific, tailored insights with a non-technical team. By understanding the distinct roles of these three tools, you can build a reporting strategy that balances speed with durability.

Is your AI referral traffic missing from the data?

If you see zero sessions, the issue is rarely the regex but a mismatch between the dimension you filter and GA4’s attribution dimension, specifically filtering on Medium when the condition should target Source. Since these platforms send a source but no medium, a medium-based rule misses every session, so ensure your channel condition targets the source dimension to avoid silent exclusions. Beyond configuration errors, GA4 has a technical ceiling because not every AI interaction ends in a click; if a user gets their answer inside the AI interface, no request reaches your domain, meaning your ChatGPT traffic tracking data represents only the portion of AI exposure that converts into visits. For troubleshooting, verify you selected Matches regex rather than Contains when defining the channel rule, as Contains can fail to parse complex domain strings correctly within the regex engine. Switching to the explicit Matches regex operator ensures the pattern evaluates the full source string as intended. As these platforms mature, the boundary between content exposure and trackable visits will tighten, with more granular signals like viewable impressions within AI answers becoming standard. For now, keep your GA4 bot filtering rules strict and interpret your AEO traffic metrics as a lower bound on actual AI-driven influence. Pick your tool by how often you need the data, choosing between the report filter for one-off checks, the custom channel group for weekly monitoring, or the custom report for sharing granular breakdowns.

Now consider how your team values that source. Do you treat AI-driven referrals with the same weight as organic search, or do they remain a curiosity on the periphery? If the answer is unclear, your current data setup likely mirrors that hesitation. Reviewing your attribution logic now will determine whether you can see the shift before it becomes the baseline.

AEO/GEO

Win visibility in AI-generated answers

Create, optimize, host and distribute AI-ready content at scale with AEO/GEO.

Explore the platform

Show up in AI answers

Explore our service

See how it works