# AI Search Analytics: AI Referrals to Pipeline

> AI search analytics is the practice of measuring what AI assistants do for your revenue, not only the clicks they send. Referral traffic shows a fraction of it. Join three records (Search Console, product analytics, a dated log of what you published), watch six signals, and set up GA4 and HubSpot so AI visits land in their own channel.

By Othmane Khadri, founder of Earleads. Published 2026-10-15. Canonical: https://www.earleads.com/blog/ai-search-analytics/

## What is AI search analytics?

AI search analytics is how you measure what ChatGPT, Gemini, Claude, Perplexity and Google's AI features do for your revenue. It covers the visits they send, how often their answers name you, and the effects a referral report never shows, like branded searches, direct visits and what new customers say when you ask how they found you.

Most teams start with the visits, because that's what their analytics shows. Open GA4, filter for chatgpt.com, see a thin line near zero and conclude that AI search isn't worth the effort. We think that conclusion is usually wrong, and this guide shows why, then walks through the setup that gets AI traffic analytics right in GA4 and HubSpot.

A word on who's writing. [Earleads](https://www.earleads.com/) runs AI search visibility programmes for B2B and B2C brands, and the measurement model below is the one we use with clients, starting at the [audit](/#audit). The setup steps don't depend on hiring anyone. Everything here works with the free GA4 and whatever HubSpot tier you already have.

## Why AI referral traffic undercounts what AI does for you

AI sends citations, not clicks. A buyer asks ChatGPT for a shortlist, reads three names, and maybe clicks one. Often they don't click at all. They open a new tab and type the brand, or they come back two days later from a bookmark, or a colleague forwards the link in Slack.

The public numbers show how small the click share is. Conductor's 2026 benchmarks, built on 13,770 domains, found AI referral traffic "accounts for 1.08% of all website traffic," with ChatGPT sending 87.4% of it ([Conductor, July 2026](https://conductor.com/academy/aeo-geo-benchmarks-report/)). SE Ranking's study of 101,574 websites puts AI platforms at 0.32% of website traffic in 2026, up from 0.02% in 2024 ([SE Ranking, June 2026](https://seranking.com/blog/ai-traffic-research-study/)). Different samples, same picture. AI referral traffic is a rounding error next to direct and organic.

Three leaks explain the gap between that rounding error and what AI really influences.

1. **Google's AI features hide inside organic.** Google says sites appearing in AI Overviews and AI Mode "are included in the overall search traffic in Search Console" under the Web search type ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). In GA4 those clicks look like any other google / organic session. GA4's own AI Assistant channel excludes them by design.
2. **Visits without a referrer land in Direct.** A link tapped inside a mobile app often opens in the phone's browser with no referrer attached. A URL copied out of an answer and pasted into the address bar never had one. Both show up as Direct.
3. **The answer works, the click happens later.** The visitor ChatGPT told about you, who then typed your domain, is invisible to every channel report. Only asking them catches that visit.

![Diagram showing a buyer who reads an AI answer that names you taking four paths, a click in ChatGPT or Perplexity tracked as an AI referral, a click from Google AI Overviews counted as Organic, a link opened from an app or pasted counted as Direct, and a search for your name days later left untagged](/blog/ai-search-analytics/diagram-1.svg)
*One AI answer, four ways the visit arrives. A referral report only sees the first one.*

So if you judge AI search by referrals, you'll cancel the channel buyers use to build their shortlists. We make the same argument in our guide on [how to choose a GEO agency](/blog/how-to-choose-a-geo-agency/), because an agency reporting AI referral traffic alone is reporting the part that barely moves.

## The three records and six signals of AI search analytics

Good AI search analytics doesn't rely on one report. It joins three records that already exist in most companies, then reads six signals across them.

**The three records:**

1. **Search Console.** Branded search impressions and clicks over time. When AI answers start naming you, more people search for you by name.
2. **Product analytics or your CRM.** Landing page, then signup, then paid customer or closed deal. GA4 plus HubSpot covers this for most B2B teams. Shopify or your store platform does it for ecommerce.
3. **An action log.** Every page you published or refreshed and every forum thread or community post, each with the date it went live. Without this log you can't tell a result from a coincidence.

**The six signals:**

1. **Landing page.** Which page AI-referred and branded visitors land on first.
2. **Branded search lift,** read against the weeks your citations landed.
3. **Direct traffic lift** over the same weeks.
4. **Citation share.** How often AI answers name you on the questions buyers ask, measured on a fixed question list (our [AI visibility audit guide](/blog/what-an-ai-visibility-audit-should-show-you/) covers how).
5. **Tagged links** where you control the link, like a utm on a link in your own directory listing or community profile.
6. **A "how did you hear about us" field** at signup or checkout. It's the only way to catch the visitor who heard about you in ChatGPT and typed your domain.

![Diagram showing three records, Search Console branded search, analytics and CRM from landing to deal, and an action log of what went live and when, joined by week into six signals that end in a pipeline you can defend](/blog/ai-search-analytics/diagram-2.svg)
*Three records you already own, joined by date. The action log is the one most teams skip.*

No single signal proves anything. A branded search lift in a week when you also ran a webinar proves nothing. A branded search lift in the three weeks after five threads and two comparison pages started getting cited, with no other campaign live, is a pattern worth acting on. The steps below set up each record.

## Step 1: Start the action log before you change anything

The action log is the cheapest record and the one teams skip. Start it before any tracking work, because you can't rebuild it later.

A spreadsheet is enough. One row per action, with these columns:

- Date live (the day it was published, not the day it was written)
- Type (new page, refresh, comparison page, forum thread, directory listing, press mention)
- URL
- Buyer question it targets
- First date an AI answer cited it, filled in when you see it

Give it one owner. When someone asks in month three why demos went up, this log is how you answer with something better than a guess. It also protects you from the opposite mistake, crediting AI search for a lift that came from a sales push or a price change.

## Step 2: Track AI traffic in GA4 with a custom channel group

GA4 got better at this in 2026. On 13 May 2026, Google added an AI Assistant channel to the default channel group. Visits from "chatbots like ChatGPT, Gemini, and Claude" now get the medium ai-assistant and the campaign (ai-assistant) "when the referrer matches a recognized AI Assistant" ([Google Analytics, what's new](https://support.google.com/analytics/answer/9164320)). The channel definition names ChatGPT, Gemini, DeepSeek, Copilot and Grok, and excludes AI Overviews and AI Mode ([GA4 default channel group](https://support.google.com/analytics/answer/9756891)).

That's a good start, with two limits. Google doesn't publish the full list of assistants it recognises, so check where Perplexity and smaller assistants land in your own data. And the channel only fires when a referrer is present, so it can't see the Direct leak above.

A custom channel group fixes the first limit and gives you a channel you control. Here's the setup, from [Google's channel group documentation](https://support.google.com/analytics/answer/13051316):

1. **Check your role.** You need Editor or above on the property.
2. **Open channel groups.** Go to Admin, then Data display, then Channel groups, then Create new channel group. If you haven't made one before, GA4 starts you with a copy of the default group.
3. **Name it.** Something like "Channels with AI". Standard properties get two custom groups, 360 properties get five, so don't waste one on a test.
4. **Add a channel called AI assistants.** Set two conditions joined with OR. Medium exactly matches ai-assistant, which catches everything Google's own list recognises. Source matches regex, with a pattern for the assistants you want named.
5. **Use a regex that covers the rest.** This one matches the nine domains HubSpot recognises plus DeepSeek:

```
.*(chatgpt\.com|openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|meta\.ai|mistral\.ai|poe\.com|grok\.com|deepseek\.com).*
```

6. **Move the channel above Referral.** Google is explicit that "traffic is included in the first channel whose definition it matches given the current order of channels in the group." If AI assistants sits below Referral, every AI visit is claimed by Referral first.
7. **Save and apply it to your reports.** Google says custom channel groups "can be applied to your reports retroactively," so you get AI traffic analytics for months before the native channel existed.

![Diagram comparing GA4 channel order, where a session from perplexity.ai is lost inside Referral when Referral is listed first, and counted as AI traffic when the AI assistants channel sits above Referral](/blog/ai-search-analytics/diagram-3.svg)
*GA4 stops at the first channel that matches. Order is the whole trick.*

Then make it useful. In Reports, open Traffic acquisition, switch the primary dimension to your new channel group, and add your key events (sign up, demo request, purchase) as columns. Compare engagement and conversion rate for AI assistants against Organic Search. Treat the absolute numbers as a floor, never a total, for the reasons in the section above. A word of caution on deletions too. Google warns that deleting a custom channel group is irreversible and breaks every report and audience built on it.

## Step 3: Set up HubSpot AI Referrals reporting

If HubSpot is your CRM, most of the work is done for you. HubSpot has a native AI Referrals traffic source. In its words, "traffic categorized under AI Referrals comes from AI assistants and chatbots like ChatGPT, Claude, Perplexity, and others" ([HubSpot Knowledge Base](https://knowledge.hubspot.com/reports/understand-hubspots-traffic-sources-in-the-traffic-analytics-tool)).

Two details matter for accuracy:

- **What counts.** HubSpot recognises chatgpt.com, claude.ai, perplexity.ai, copilot.microsoft.com, gemini.google.com, meta.ai, mistral.ai, poe.com and grok.com. A visit is classed as AI Referrals when a utm_source, utm_medium, utm_campaign or source parameter matches one of those domains, or the referring domain itself is a recognised AI platform.
- **What doesn't.** Regular Google and Bing search stay in Organic search. Only AI specific domains like gemini.google.com count as AI Referrals. So clicks from AI Overviews are invisible here too.

The bigger win is on the contact record. AI Referrals is a value of the Original Traffic Source and Latest Traffic Source properties, with the AI platform domain as Drill-Down 1 and the utm_campaign as Drill-Down 2 ([HubSpot traffic source properties](https://knowledge.hubspot.com/properties/understand-traffic-source-properties)). That means you can follow an AI visit all the way to revenue.

Build three things:

1. **A saved contact view** where Original Traffic Source is AI Referrals, with Drill-Down 1 as a column so you see which assistant sent each contact.
2. **A deal report** on deals whose associated contacts have Original Traffic Source equal to AI Referrals, showing amount, stage and close date. HubSpot's prebuilt closed revenue by source report does a version of this, counting only deals with an associated contact.
3. **An attribution report** if your tier has it. Contact create attribution needs Marketing Hub Professional or Enterprise, deal create attribution needs Marketing Hub Enterprise ([HubSpot attribution reports](https://knowledge.hubspot.com/reports/create-attribution-reports)). Compare First Touch with Linear. If AI Referrals looks strong on first touch and weak on linear, AI is opening conversations that other channels close.

On Salesforce, Pipedrive or another CRM, recreate the same thing by hand. Add an original source field and a source detail field, fill them from the first-touch referrer and utm_source, and keep them read-only after creation so a rep's later edit can't overwrite the truth.

## Step 4: Add the question your analytics can't answer

Every setup above misses the buyer who read about you in an AI answer and typed your domain. The fix is old fashioned. Ask them.

Add a "How did you hear about us?" field to your signup form, demo form or checkout. Two rules make it work:

- **Make it free text, or put free text first.** A dropdown with "Search engine" as an option collects "Search engine". A text box collects "ChatGPT said you were the best option for agencies," which tells you the engine and the question.
- **Write the answer onto the CRM record.** Store it as a contact property next to Original Traffic Source, then report on both. When the analytics says Direct and the customer says ChatGPT, you've measured the Direct leak.

Tag mentions of ChatGPT, Gemini, Claude, Perplexity, Copilot, "AI" or "an assistant" once a week. In HubSpot a workflow with a contains filter can set a checkbox. The count will be imperfect, and that's fine. It's still the most direct evidence you'll get that AI answers are sending you buyers.

## Step 5: Read branded search against the weeks citations landed

This is where the action log pays off. In Search Console, filter Performance to queries containing your brand name and export weekly clicks and impressions. Put them in the same sheet as your action log and your direct sessions from GA4.

Then look for one specific shape. Branded impressions and direct sessions rise in the weeks after a cluster of pages or threads started getting cited, with no other campaign running. If that shape repeats across two or three clusters, you have a correlation you can defend in a board meeting. If branded search moved before anything was cited, something else drove it, and the log tells you so.

Keep measured and correlated apart. "AI referrals sent 41 sessions this month" is measured. "Branded search rose after citations landed" is correlated. Both belong in a report, but never merged into one flattering number, because the first time someone pulls the merged figure apart the whole report loses credibility.

## Step 6: Join AI visits to pipeline for B2B and ecommerce

The last record to connect is revenue, and the method depends on how you sell.

**B2B.** Most AI-referred visitors never fill in a form. A company identification tool on the site can name the business behind a visit. From there:

1. Gate the company against your own ICP rules (size, industry, region), not the vendor's intent score.
2. Route matching accounts to an owner with a short brief.
3. Write the AI source onto the company and contact records, so it stays attached to any deal created later.

Use the visit for timing only. Never mention it in outreach. "I saw you came from ChatGPT" ends conversations. Our article on [GEO for B2B](/blog/geo-for-b2b/) covers the routing side in more depth.

**Ecommerce and B2C.** Visits are many and cheap, so per-visitor identification isn't the point. Two things work better. Use promo codes in places you control, such as a code mentioned in a community profile or a directory listing, so redemptions tie back to that surface. And plot daily units sold against your action log. When a product page starts getting cited in AI shopping answers, look for a step in units after its live date, and check it against seasonality before you credit it.

## Step 7: Build a one-page monthly report

All of this should fit on one page a CEO can read in two minutes. Ours has these blocks:

- **Share of answers** against three named competitors, per engine, on the buyer questions you track.
- **What each published piece earned.** Citations, referral visits and any deals touched, row by row from the action log.
- **Branded search** month over month.
- **Pipeline.** Trials, demos or revenue against what was published, from Steps 3 and 6.
- **Commitments and targets kept apart.** What you control (pages shipped, threads posted) is reported separately from what you don't (what engines cite), every month, including the months that disappoint.

Two questions matter most to leadership. Does the brand show up on questions that don't name it? And by day 90, what does a citation cost compared with what a customer is worth? If you're building this for an enterprise team, our [enterprise GEO strategy](/blog/enterprise-geo-strategy/) guide covers who should own which number.

## When AI search analytics says AI does nothing

Sometimes the numbers really are flat. Here's the decision rule we use before calling it.

If AI referral traffic is flat, read the other five signals before touching budget. If citation share is rising, branded search is lifting after citations land and signup answers mention assistants, the channel works and the referral report is undercounting it. Keep going.

If citation share is flat after three months of focused work, the problem is upstream of analytics. The engines aren't naming you, and no dashboard fixes that. Go back to which questions you target and which sources the answers read, which our guide on [how to get ChatGPT to recommend your business](/blog/how-to-get-chatgpt-to-recommend-your-business/) covers step by step.

And if you don't yet have meaningful traffic at all, don't pay for attribution software. Set up the free parts first (the GA4 channel group, HubSpot's AI Referrals source, the action log and the signup question). They cost an afternoon, and they're the base any paid tool would sit on anyway.

## Frequently asked questions

### What is AI search analytics?

AI search analytics measures how AI assistants such as ChatGPT, Gemini, Claude, Perplexity and Google's AI features affect your business. It covers the visits they send, how often answers cite or name your brand, and the indirect effects that referral reports miss, like branded search, direct visits and what customers say when asked how they found you.

### How do I track AI traffic in GA4?

Check the AI Assistant channel Google added to the default channel group on 13 May 2026. Then build a custom channel group under Admin, Data display, Channel groups, with an AI channel that matches medium ai-assistant or a source regex for the assistants you care about. Move it above Referral, because GA4 assigns each session to the first channel it matches.

### Does HubSpot track AI referrals?

Yes. HubSpot has an AI Referrals traffic source covering chatgpt.com, claude.ai, perplexity.ai, copilot.microsoft.com, gemini.google.com, meta.ai, mistral.ai, poe.com and grok.com. It shows in traffic analytics and in the Original and Latest Traffic Source properties on contacts, with the AI platform domain as the first drill-down and the utm_campaign as the second.

### Why is my AI referral traffic so low?

Because most of what AI does for you never arrives as a referral. Clicks from Google's AI Overviews and AI Mode are counted as organic search. Links opened from apps or copied into a browser often arrive with no referrer and land in Direct. And many buyers read an answer, then search your name later. Low referrals alone don't mean AI search isn't working.

### How do you attribute pipeline to AI search?

Join three records instead of trusting one. Search Console shows branded search lift, product analytics or the CRM shows landing page to signup to deal, and a dated action log shows when each page or thread went live. Add a free text 'how did you hear about us' field at signup, and read every lift against the weeks your citations landed.

### Is the GA4 AI Assistant channel enough on its own?

Not yet. Google's documentation names ChatGPT, Gemini, Claude, DeepSeek, Copilot and Grok and excludes AI Overviews and AI Mode. The channel only fires when a referrer matches, so visits that arrive without one stay in Direct. A custom channel group lets you add the assistants Google leaves unnamed and can be applied to older data.