FishingSEO
AI in SEO

How to Track LLM Referral Traffic With AI in 1 Hour

By FishingSEO••13 min read

AI search traffic is still small for most sites, but it is growing too fast to ignore. Adobe Analytics reported that traffic from generative AI sources to U.S. retail websites increased 1,200% in February 2025 compared with July 2024 (Adobe). Semrush later found that outbound referral traffic from ChatGPT to the wider web grew 206% in 2025 (Semrush).

Here is the uncomfortable part: a lot of that traffic is easy to miss.

Some visits from ChatGPT, Perplexity, Gemini, Claude, Copilot, and other AI tools show up in Google Analytics 4 as normal referral traffic. Some appear as direct traffic. Some never pass useful referrer data at all. So if you only look at “Organic Search,” you may be undercounting how often AI tools send people to your site.

The good news: you do not need a huge analytics project to get a useful first view. In about one hour, you can use GA4, a simple regex filter, and AI-assisted analysis to build a working LLM referral traffic report.

What LLM Referral Traffic Means

LLM referral traffic means visits that arrive after someone clicks a link inside or near an AI-generated answer.

Common sources include:

  • ChatGPT
  • Perplexity
  • Gemini
  • Microsoft Copilot
  • Claude
  • You.com
  • Poe
  • SearchGPT-style AI search surfaces
  • AI answer engines that cite or link to websites

This is different from traditional organic search traffic. A user may not type a keyword into Google, scan ten blue links, and click your page. Instead, they ask an AI assistant a question, read a summarized answer, and click a cited source or recommended page.

That changes your SEO measurement in three ways:

  • You may get fewer impressions but more qualified clicks.
  • You often cannot see the exact prompt that caused the visit.
  • Your traffic may be split across referral, direct, and unassigned buckets.

Google’s own GA4 documentation is useful here because it explains why custom grouping matters. Google says, “A channel group is a set of channels” built from rule-based traffic categories (Google Analytics Help). For LLM traffic, you are basically creating your own rule-based AI channel because GA4 does not cleanly separate it for you by default.

Why This Matters for SEO in 2026

AI referrals are not replacing Google organic traffic for most sites yet. They are becoming an early signal of how your content performs in answer engines.

Conductor’s 2026 AEO/GEO benchmark found that ChatGPT represented 87.4% of AI referral traffic across the industries it analyzed (Conductor). Search Engine Land, summarizing the same benchmark, reported that AI referral traffic accounted for just over 1% of website visits across 10 major industries (Search Engine Land).

That combination matters:

  • AI traffic is still a small slice.
  • The growth rate is high.
  • ChatGPT dominates the channel.
  • Referral visibility is messy.
  • Early tracking gives you a baseline before the channel gets more competitive.

There is also a defensive reason to track it. Axios reported that traditional search referral traffic declined sharply for publishers, with Chartbeat data showing a 60% decline for small publishers over two years (Axios). Even if your site is not a publisher, the trend is clear: search journeys are fragmenting.

The 1-Hour Setup

This workflow gives you a practical first version, not a perfect attribution system.

You will build:

  • A GA4 exploration for known AI referrers
  • A custom AI traffic segment or channel group
  • A landing page report
  • An AI-assisted summary of patterns
  • A simple tracking sheet for weekly review

Minute 0-10: List Your AI Referrer Sources

Start with a simple list of known LLM and AI search domains.

Use this starter regex:

.*(chatgpt\.com|chat\.openai\.com|openai\.com|perplexity\.ai|copilot\.microsoft\.com|gemini\.google\.com|bard\.google\.com|claude\.ai|anthropic\.com|you\.com|poe\.com|phind\.com|komo\.ai|writesonic\.com).*

This will not catch everything. That is fine. Your first goal is directional visibility.

Ask AI to help expand the list with a prompt like:

Act as an SEO analytics specialist. Expand this GA4 regex for known LLM and AI search referral sources. Include only domains that could plausibly appear as referrers. Return the regex and a short explanation of each added domain.

Then check the output manually. Do not paste a huge AI-generated regex into GA4 without reviewing it. You want clean tracking, not a junk drawer.

Minute 10-25: Build a GA4 Exploration

In GA4, go to:

Explore -> Free form

Add these dimensions:

  • Session source / medium
  • Session source
  • Page referrer
  • Landing page + query string
  • Page title
  • Date

Add these metrics:

  • Sessions
  • Engaged sessions
  • Engagement rate
  • Average engagement time
  • Key events
  • Total revenue, if relevant

Now apply a filter using your regex.

Useful filter combinations:

  • Session source matches regex
  • Page referrer matches regex
  • Session source / medium matches regex

Google defines “Page referrer” as “the referring URL” in its GA4 dimensions documentation (Google Analytics Help). That makes it especially useful for spotting AI tools that pass a referring domain.

Export the result as CSV once you have data.

Minute 25-35: Create a Custom AI Channel Group

If you have access to custom channel groups, create a new channel called something like:

AI Referral

Rules can include:

  • Source matches your AI regex
  • Page referrer matches your AI regex
  • Medium equals referral and source matches known AI domains

Place this channel above broad referral rules where possible. Channel order can affect classification, so your AI rule should not sit below a generic bucket that catches everything first.

Keep the name simple. Avoid trendy labels like “GEO Magic” or “Answer Engine Goldmine.” Your reports should be understandable six months from now.

Minute 35-45: Use AI to Analyze the Export

Now use AI where it is actually useful: pattern recognition.

Paste your CSV into a secure AI workspace if your company policy allows it. If the data contains personal information, remove it first.

Use a prompt like:

You are analyzing GA4 referral traffic from AI and LLM sources.

Find:
1. Which AI sources sent the most sessions
2. Which landing pages received the most AI traffic
3. Which pages had unusually high or low engagement
4. Any obvious content themes
5. Three SEO actions based on the data

Do not invent missing data. If the sample is too small, say so.

This can quickly tell you whether AI tools are sending people to:

  • Product pages
  • Blog posts
  • Comparison pages
  • Definitions
  • How-to guides
  • Pricing pages
  • Homepages
  • Support docs

That matters because LLM referrals often reveal what your content is being used for. If AI tools mostly send users to your definitions and guides, your informational authority may be working. If they send users to comparison or pricing pages, AI may already be helping bottom-funnel research.

Minute 45-55: Build a Weekly Tracking Sheet

Create a simple sheet with these columns:

  • Week
  • AI sessions
  • AI engaged sessions
  • AI engagement rate
  • Top AI source
  • Top landing page
  • Key events
  • Revenue or leads
  • Notes
  • Content action

You are looking for trends, not perfection.

Good weekly questions include:

  • Did AI traffic rise after a content refresh?
  • Which pages keep getting cited or clicked?
  • Are AI visitors more engaged than organic visitors?
  • Do AI referrals assist conversions, even if they are not last-click winners?
  • Are there sudden spikes from one platform?

This also pairs well with content refresh work. If you already update older pages, you can connect this workflow with an AI-assisted refresh process like 9 Ways to Use AI for Content Refreshes That Recover Rankings.

Minute 55-60: Write Down the Limitations

This final step is important. Add a note directly inside your report:

LLM referral tracking is directional. Some AI traffic may appear as direct, referral, unassigned, or may not pass referrer data.

That note prevents bad decisions later.

What AI Can and Cannot Do Here

AI can speed up the analysis, but it cannot magically recover hidden attribution data.

AI is useful for:

  • Expanding your source list
  • Cleaning exported GA4 rows
  • Grouping landing pages by topic
  • Summarizing engagement patterns
  • Spotting anomalies
  • Drafting weekly commentary
  • Suggesting content updates based on page performance

AI is not reliable for:

  • Guessing exact user prompts
  • Identifying hidden direct traffic with certainty
  • Proving that a page was cited in an AI answer
  • Replacing GA4, server logs, or analytics governance
  • Making attribution claims from tiny samples

Use AI as an analyst, not as a source of truth.

Pros and Cons of Tracking LLM Referral Traffic

Pros

You get earlier visibility into AI search behavior. Even small numbers can show which pages answer questions well enough to earn clicks.

You can find new content opportunities. If Perplexity sends traffic to one comparison page, related comparison pages may deserve updates.

You can separate AI traffic from generic referral traffic. This keeps your reporting cleaner.

You can improve executive reporting. “AI traffic is growing 18% month over month” is more useful than “referral traffic changed.”

You can connect SEO with brand visibility. LLM traffic often reflects whether your brand, entities, and content are being surfaced in AI-assisted journeys.

For deeper entity and topical relevance work, the ideas in The Simple Secret to Entity SEO With AI are a useful next layer.

Cons

The data is incomplete. Some AI apps and browsers strip referrer data.

The volume may be tiny. For many sites, AI traffic is still too small for strong conclusions.

GA4 classification can be messy. Traffic may appear under referral, direct, unassigned, or source-specific rows.

You usually cannot see the prompt. That makes intent analysis more indirect than keyword reporting.

Regex lists need maintenance. New AI products and domains appear often.

Attribution can be misleading. An AI tool may influence a user who later returns through Google, email, or direct traffic.

Practical Tips to Improve Your LLM Traffic Report

Track Landing Pages, Not Just Sources

Do not stop at “ChatGPT sent 47 sessions.”

Look at the pages.

Ask:

  • Which pages attract AI clicks?
  • Are they clear, current, and trustworthy?
  • Do they include original examples or data?
  • Do they answer the question quickly?
  • Do they link to the next useful page?

If a post gets AI referral traffic but has weak internal links, improve the path. A related guide like How to Build AI-Driven Internal Links in 30 Minutes can help you turn isolated visits into deeper sessions.

Compare AI Visitors With Organic Visitors

For your top AI landing pages, compare:

  • AI referral engagement rate vs organic search engagement rate
  • AI average engagement time vs organic search
  • AI key event rate vs organic search
  • AI landing page mix vs organic search landing page mix

Do not assume AI traffic is better or worse. Measure it.

Adobe found that visitors from generative AI sources showed stronger engagement in some verticals; for example, banking-site visitors from generative AI sources spent 45% more time browsing than non-AI visitors (Adobe). Your site may behave differently.

Add Server Log Checks if You Can

GA4 is enough for a one-hour setup, but server logs can reveal more.

Look for:

  • Referrers from AI domains
  • AI crawler user agents
  • Sudden spikes to specific URLs
  • Pages requested by bots before traffic increases

This is especially useful for larger sites, ecommerce sites, and publishers.

Watch for Homepage-Heavy Direct Traffic

Some AI-influenced visits may land as direct traffic, especially from apps or privacy-restricted environments.

You cannot label all unexplained direct traffic as AI traffic. But you can watch for patterns:

  • Direct traffic spikes after AI visibility improves
  • Direct visits landing on deep informational URLs
  • Direct sessions increasing without matching brand search growth
  • Assisted conversions where the first visible source is unclear

Treat this as a clue, not proof.

Keep Your Regex Versioned

Save your regex with a date.

Example:

AI referral regex v1 - created 2026-04-29

When you add new domains, document them. This makes reporting changes easier to explain later.

Content Trends Behind LLM Referral Traffic

LLM referral tracking is not just analytics housekeeping. It reflects a bigger shift in SEO.

AI systems tend to reward content that is easy to parse, cite, and summarize. That does not mean you should write robotic content. It means your pages should make expertise obvious.

Content that often performs well in AI-assisted discovery includes:

  • Clear definitions
  • Step-by-step tutorials
  • Comparison pages
  • Original research
  • Expert commentary
  • Updated statistics
  • FAQs with direct answers
  • Strong author and source signals
  • Well-structured internal links
  • Pages that explain entities, relationships, and context

This overlaps with E-E-A-T work. If you use AI to draft or refresh content, make sure the final page includes real expertise, examples, and verification. The guide How to Turn AI Drafts into E-E-A-T Content in 7 Days is a good companion for that process.

A Simple AI Prompt for Weekly Reporting

Use this after exporting your GA4 AI referral report:

You are my SEO analytics assistant.

Analyze this weekly LLM referral traffic export.

Return:
- 5 key observations
- 3 landing pages worth improving
- 3 possible reasons traffic changed
- 3 recommended SEO actions
- Any data quality warnings

Rules:
- Use only the data provided.
- Do not invent causes.
- Mark weak conclusions as hypotheses.
- Keep the summary clear enough for a marketing manager.

This gives you a fast weekly readout without pretending the data is more precise than it is.

Common Mistakes to Avoid

Do not call all direct traffic “AI traffic.” That will make your reporting useless.

Do not track only ChatGPT. It dominates many benchmarks, but Perplexity, Gemini, Copilot, Claude, and niche answer engines can matter depending on your audience.

Do not ignore tiny numbers. A small number of AI visits to high-intent pages can still be meaningful.

Do not over-optimize for one AI platform. Build content that is clear, useful, and well-sourced across search and answer engines.

Do not publish AI-written updates without quality checks. If your pages become generic, they are less likely to earn citations, links, or trust. Use a review process like Stop Publishing AI Content Without These SEO Checks before shipping changes.

Conclusion

LLM referral traffic tracking is imperfect, but it is already useful. In one hour, you can create a GA4 exploration, group known AI sources, export the data, and use AI to summarize patterns across sources, landing pages, and engagement.

The key is to treat the report as directional intelligence. It will not show every AI-influenced visit, and it will not reveal every prompt. But it will help you see which pages are already earning clicks from AI-assisted discovery, which content deserves improvement, and how your SEO strategy is adapting as search behavior changes.

Back to all posts