# Why AI Referral Traffic Is Undercounted (and How to Estimate It)

> Why GA4 shows so little AI referral traffic, what Google's docs say about direct traffic and the new AI Assistant channel, and how to estimate the gap.

Source: https://mentionx.ai/learn/measurement/ai-referral-traffic-undercounted  
Author: Dipuck Jones, founder of MentionX  
Updated: 2 November 2026  
Level: Practitioner

**AI referral traffic is undercounted because many AI visits arrive without a referrer and land in (direct) / (none), Google's AI features are counted inside ordinary search traffic, and most AI answers send no click at all. You can't recover the missing visits exactly, but you can estimate them and report a range.** This guide explains each cause and gives an estimation method.

*Method: we read Google Analytics Help on direct traffic, traffic source scopes and "What's new", and Google Search Central's AI features guide, on 31 October 2026 (CET). Quotes are verbatim. Opinions are marked "Our view"; examples are marked "Illustrative".*

> **Key takeaways**
>
> - GA4 undercounts AI driven visits because missing referrers become direct, Google's AI features count as search and many answers get no click.
> - Since 13 May 2026, GA4 labels visits with a recognised AI assistant referrer as the AI Assistant channel, but only when the referrer matches.
> - Estimate the hidden part with deep page direct traffic, a 'How did you hear about us?' question and branded search, and report a range.
> - Lead reports with AI visibility, then show measured sessions and labelled estimates.

*Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.*

## Why does GA4 show so little AI traffic?

### Reason 1: missing referrers become "direct"

Google's definition: "(direct) / (none)" represents "website traffic that doesn't have a clear referral source" ([Google Analytics](https://support.google.com/analytics/answer/15258820?hl=en)). Its listed causes include links that "lack UTM parameters", redirects that "can strip UTM parameters from the URL", and links "from offline documents such as PDFs and Word documents" ([Google Analytics](https://support.google.com/analytics/answer/15258820?hl=en)).

AI visits hit these routes often. *Fictional example:* someone copies a URL out of a chatbot answer and pastes it into a new tab; a desktop app opens a link without passing a referrer; a user asks an assistant for a link and taps it in an in app browser. Each can arrive with no source.

### Reason 2: Google's AI features count as search

Google says "sites appearing in AI features (such as AI Overviews and AI Mode) are included in the overall search traffic in Search Console", reported in the Performance report under the "Web" search type ([Google](https://developers.google.com/search/docs/appearance/ai-features)). In GA4, a click from AI Mode looks like Google organic. There's no separate AI Mode line.

### Reason 3: zero click answers

Many AI answers mention a brand without the user clicking. That influence never appears as a session. *Our view:* this is the biggest gap, and analytics can't close it. Prompt tracking measures it instead.

## What does GA4 detect now?

On 13 May 2026 Google announced "New AI Assistant traffic measurement" ([Google Analytics](https://support.google.com/analytics/answer/9164320?hl=en)). It named "chatbots like ChatGPT, Gemini, and Claude" and listed three changes:

| Dimension | New value |
|---|---|
| Medium | "ai-assistant", assigned "when the referrer matches a recognized AI Assistant" |
| Channel group | "AI Assistant" |
| Campaign | "(ai-assistant)" |

Source: [Google Analytics](https://support.google.com/analytics/answer/9164320?hl=en).

Note the condition: the referrer must match. Visits that arrive without one still go to direct. *Our view:* the new channel improves labelling of AI visits GA4 can see; it doesn't find the ones it can't.

## Which scope should you report?

GA4 has three scopes. "User scoped dimensions show you where your new users are coming from"; "Session scoped dimensions show you where both new and returning users are coming from when they start new sessions"; and "Event scoped dimensions help you attribute credit for a key event" ([Google Analytics](https://support.google.com/analytics/answer/11080067?hl=en)).

*Our view:* use session scoped channel for traffic trends and event scoped for conversions, and state which in every report. Mixing scopes is a common source of mismatched numbers.

## How do you estimate the gap?

You can't measure hidden visits directly, so triangulate. Present a range, never a single number.

### Method 1: direct traffic landing pages

Long, deep URLs rarely get typed. *Fictional example:* if a 3,000 word guide gets 400 direct sessions a month to its exact URL, and you have no email or PDF links to it, some of that is likely AI driven. Compare direct sessions on deep pages before and after you started appearing in AI answers.

Steps:

1. Filter to Session default channel group = Direct.
2. Add landing page as a dimension.
3. Exclude the homepage and short URLs people might type.
4. Check your campaign links, so known untagged sources (emails, PDFs) are ruled out first, as Google's guide suggests ([Google Analytics](https://support.google.com/analytics/answer/15258820?hl=en)).
5. Track the remaining deep page direct sessions monthly.

### Method 2: ask people

Add "How did you hear about us?" with an "AI assistant (ChatGPT, Gemini, etc.)" option to signup or checkout. Self reported data is imperfect but captures zero click influence that analytics never sees.

### Method 3: branded search lift

If AI answers mention you, some people search your brand name later. Watch branded queries in Search Console against your AI visibility over time. *Our view:* correlation only, but useful alongside the other two.

### Putting it together

*Fictional* (made up numbers to show layout):

| Signal | Monthly value |
|---|---|
| AI Assistant channel sessions | 300 |
| Unexplained direct on deep pages | 500 to 900 |
| Survey: "AI assistant" share of signups | 6% |
| Estimated AI influenced sessions | 300 (measured) + 200 to 600 (estimated) |

State which part is measured and which is estimated, with the method.

## What should you fix first?

- **Tag what you control.** Add UTMs to links you place yourself, such as your own GPT descriptions, app listings or partner pages.
- **Check redirects.** Google warns redirects can strip UTMs, including https to http hops ([Google Analytics](https://support.google.com/analytics/answer/15258820?hl=en)).
- **Build a custom channel group** if you need AI sources the default group doesn't recognise yet.
- **Track prompts.** Measure mentions and citations in AI answers directly. See [GA4 custom channel group for AI referrals](https://mentionx.ai/learn/measurement/ga4-custom-channel-group-ai-referrals) and [self reported attribution](https://mentionx.ai/learn/measurement/self-reported-attribution-adding-chatgpt-forms).

## How do you report it to leadership?

*Our view:*

1. Lead with visibility (share of prompts where you're mentioned), not clicks.
2. Show measured AI sessions, then the estimated range, clearly labelled.
3. Explain the three reasons for undercounting in one line each.
4. Don't promise precision you can't deliver.

## From MentionX data: why clicks understate AI influence

**Method.** Read only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

Referral traffic only captures answers that send a click. Our sample shows how much of AI influence happens in the answer itself:

| Engine | Brand mentioned | Answers with no citations |
| --- | --- | --- |
| Gemini | 47.3% | 50.7% |
| OpenAI | 42.9% | 30.2% |
| Perplexity | 31.5% | 4.1% |
| AI Mode | 31.3% | 12.6% |
| Copilot | 26.0% | 47.0% |
| DeepSeek | 24.0% | 99.0% |

Engines that rarely show sources give users little to click, even when they name brands. How often brands are named also depends on the question:

| Intent | Brand mentioned | Share of voice |
| --- | --- | --- |
| Brand specific | 100% | 64.3 |
| Comparison | 91.9% | 44.6 |
| Recommendation | 29.2% | 26.8 |
| Best of | 26.6% | 22.7 |
| Problem | 5.7% | 7.7 |

*Our view:* comparison and brand questions name brands often, and users who see your name there may come later by typing your URL or searching your brand. That's influence analytics records as direct or organic, if at all.

**Limits.** A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

## Worked example: Ledgerline sizes its AI traffic

*Ledgerline is a fictional accounts payable tool. Figures are illustrative.*

Ledgerline's monthly report shows a few hundred sessions in the new AI Assistant channel. Leadership asks whether AI matters if it drives so little.

The analyst filters direct traffic to deep guide URLs, excludes pages linked from emails and PDFs and finds a steady base of unexplained direct sessions that rose after the guides began appearing in AI answers. The signup form's "How did you hear about us?" question shows a small but growing share choosing "AI assistant".

The report presents measured AI Assistant sessions, an estimated range for hidden AI visits with the method stated, the survey share and prompt tracking results showing how often Ledgerline is mentioned. Leadership sees the measured number and the estimate side by side, with no single invented total.

> **Pro move:** Mark 13 May 2026 on every trend chart. Visits GA4 called Referral before may be AI Assistant after, and an unmarked jump will be misread as growth.

## How do you set up reporting in GA4?

Start with the default channel group and check whether the AI Assistant channel appears in your property. Build an exploration with session default channel group, session source and landing page, filtered to AI Assistant, to see which pages receive these visits. If you need AI sources the default group doesn't recognise, create a custom channel group with rules based on source; see [GA4 custom channel group for AI referrals](https://mentionx.ai/learn/measurement/ga4-custom-channel-group-ai-referrals).

Decide your scope before you share numbers. Session scoped channel suits traffic trends, while event scoped attribution suits conversions ([Google Analytics](https://support.google.com/analytics/answer/11080067?hl=en)). Write the scope in the chart title so nobody compares a session figure with a first user figure.

## What can you learn from landing pages?

The pages AI visitors land on tell you which content engines cite and which answers send clicks. Compare AI Assistant landing pages with your organic search landing pages. Pages that appear in both are doing double duty. Pages that get AI visits but little organic traffic may be answering questions people now ask assistants rather than search engines, which is worth knowing when you plan updates.

## How should you handle conversions?

AI visitors may convert on a later visit through another channel. Look at conversion paths rather than last click alone, and treat any figure as a lower bound. If your form asks how people heard about you, compare the survey answers with the recorded channel for the same conversions. *Our view:* disagreements between the two are informative: a signup recorded as direct but self reported as "AI assistant" is exactly the hidden visit this article is about.

## How often should you review the numbers?

Monthly is enough for most teams. Look at the trend in AI Assistant sessions, the unexplained deep page direct estimate, the survey share and your prompt tracking visibility together. Quarterly, revisit your method: new assistants appear, GA4's recognition changes and your own campaign tagging improves, which all shift what's measured versus estimated.

## Where does AI traffic reporting go wrong?

- **Reporting AI referrals as "AI traffic".** They're the visible minority of AI influenced visits.
- **Adding all direct traffic to the AI total.** Most direct traffic has other causes; only the unexplained deep page portion is a candidate.
- **Comparing periods across the 13 May 2026 change.** Sessions that GA4 labelled Referral before may be labelled AI Assistant after. Note the date on any trend chart.
- **Mixing scopes.** First user, session and event figures differ by design.
- **Presenting an estimate as a measurement.** Always label ranges and show the method.

## What should you tell stakeholders about uncertainty?

Be direct: some AI influence can't be measured, and any total includes an estimate. Explain the method for the estimate in one sentence, give a range rather than a point and say what would change it. Stakeholders tend to trust a clearly labelled range more than a precise number that later turns out to be wrong. *Our view:* the goal is a consistent method you repeat each month, so the direction of change is reliable even if the absolute level isn't.

## Common questions

**Does GA4 now show all AI traffic?** No. It labels visits whose referrer matches a recognised AI assistant. Visits without a referrer still go to direct.

**Are AI Overviews clicks in the AI Assistant channel?** No. Google counts AI features as search traffic in Search Console's Web search type.

**Can I add AI tools GA4 doesn't recognise?** Yes, with a custom channel group based on source rules.

## Summary

| Cause | Fixable? | What to do |
|---|---|---|
| Missing referrer → direct | Partly | Deep page direct analysis, UTMs where you control links |
| AI Mode counted as search | No (not split out) | Report Search Console "Web" alongside AI visibility |
| Zero click answers | No | Prompt tracking, surveys, branded search |

## Sources

All accessed 31 Oct 2026.

- Google Analytics Help, [Understand (direct) / (none) traffic](https://support.google.com/analytics/answer/15258820?hl=en); [Scopes of traffic-source dimensions](https://support.google.com/analytics/answer/11080067?hl=en); [What's new in Google Analytics](https://support.google.com/analytics/answer/9164320?hl=en) (entry dated 13 May 2026)
- Google Search Central, [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features)

## Frequently asked questions

### Does GA4 track ChatGPT traffic now?

Google's 'What's new' page (13 May 2026) announced an AI Assistant channel that assigns the medium 'ai-assistant' when the referrer matches a recognised AI assistant, naming ChatGPT, Gemini and Claude.

### Where does Google AI Mode traffic show up?

Google says sites appearing in AI Overviews and AI Mode are included in overall search traffic in Search Console, in the Performance report under the 'Web' search type. It isn't broken out separately.

### Can I recover lost AI visits exactly?

No. Visits without a referrer or campaign tag become (direct) / (none) and can't be reassigned reliably. You can only estimate the gap.

---

Written and reviewed by Dipuck Jones for MentionX Academy. The HTML version of this page: https://mentionx.ai/learn/measurement/ai-referral-traffic-undercounted
