How to Track AI Search Referrals Through to Pipeline in GA4 and HubSpot
How to track AI search referral traffic in GA4 and HubSpot, from channel group and referrer list to CRM fields, then compare pipeline quality against organic search.
On this page
- What GA4 sorts on its own, and what a custom channel group adds
- Where HubSpot keeps the AI source on contacts and deals
- The metrics that carry a referral through to qualified pipeline
- How do AI-referred visitors convert compared with organic search?
- Which reports tie AI search traffic to pipeline?
- What counts as proof of qualified pipeline?
- What attribution cannot see, and how to plan for it
- Sources
To track AI search referral traffic in GA4 and HubSpot, three things have to agree: a GA4 channel group that isolates assistant referrals, the HubSpot source value that carries the same visit onto a contact and its deals, and a few extra fields for the visits neither tool can label. Many write-ups stop when the regex returns sessions, and that is the mistake this piece addresses. Sessions appear first, so they get reported first. The budget question concerns pipeline, and answering it takes one definition of "AI referral" shared by both systems and written down before anyone compares a conversion rate.
What GA4 sorts on its own, and what a custom channel group adds
GA4 has handled part of this natively since May 13, 2026. According to Google Analytics Help, the medium "ai-assistant" is assigned when a visit's referrer matches a recognized AI assistant, the visit is categorized under a new AI Assistant channel in the default channel group, and the campaign is named "(ai-assistant)".
The channel is narrower than its name suggests. The definition in Google Analytics Help reads: "AI Assistant is the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok. It excludes Google’s AI Overviews and AI Mode." The same page counts AI Overviews and AI Mode inside Organic Search. We found no published list of the matched referrers on that page or in its downloadable source list, which covers search, social, shopping and video sites.
With no published list that we could find, a custom channel group beside the default is worth keeping: you control its membership, and Google says it can be applied to reports retroactively. Google Analytics Help includes an "Example: AI assistants" walkthrough: in Admin, under Data display, open Channel groups, copy the default group, add a channel named AI assistants with a "matches regex" condition, and reorder it above Referral. Order decides the outcome, because "Traffic is included in the first channel whose definition it matches given the current order of channels in the group." Standard properties allow two custom groups, so check that one is free.
Google's sample pattern contains loose alternatives such as .*gpt.* and .*bard.*, which would also match a partner site with those letters in its domain. Our proposed starting pattern uses whole hostnames on the Source field:
^(.*\.)?(chatgpt\.com|claude\.ai|perplexity\.ai|copilot\.microsoft\.com|gemini\.google\.com|meta\.ai|mistral\.ai|poe\.com|grok\.com)$
These nine hosts are the platforms HubSpot lists for its own AI source, so both tools begin from one definition. Before saving, read the Session source values in your Traffic acquisition report and adjust the pattern to what appears there, then add other assistants (Deepseek, for one) as their hostnames show up. Google also notes that custom channel groups are unavailable in the BigQuery export schema, so a warehouse model has to rebuild the rule itself.
Where HubSpot keeps the AI source on contacts and deals
HubSpot has a dedicated value too. Per HubSpot, a contact's Original Traffic Source and Latest Traffic Source can read AI Referrals, with Drill-Down 1 holding the AI platform domain (chatgpt.com or claude.ai, for example) and Drill-Down 2 the campaign name when a utm_campaign parameter exists. Companies and deals associated with a contact mirror these properties, which lets a deal report group by AI source.
Classification follows a fixed order of fifteen rules documented by HubSpot. The AI rule is eighth, checked after the rules for tagged social, paid and email traffic and before Other campaigns. It fires when the utm_source, utm_medium or utm_campaign parameter contains a recognized AI platform domain, or when the referring domain is a recognized AI platform. The parameter half matters for ChatGPT, because OpenAI states: "ChatGPT automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs, enabling clear tracking and analysis of inbound traffic from ChatGPT search results." A session carrying that parameter meets HubSpot's rule even when the referrer is missing.
Three custom fields close the gaps. All three are our proposed setup, and none ships with HubSpot.
| Field | Object | How it is filled | Why it exists |
|---|---|---|---|
| AI platform (first touch) | Contact | Copied from Original Traffic Source Drill-Down 1 when the source is AI Referrals | HubSpot clears the drill-downs if someone changes the source value |
| Self-reported source | Contact | A required "How did you hear about us?" form question with an "AI assistant" option | Catches buyers who arrived with no referrer |
| AI influenced | Deal | Yes when an associated contact has AI Referrals as original or latest source, or an AI self-report | Counts deals where the assistant was a later touch |
The copy step can run through a workflow's Edit record action, which HubSpot describes as able to set, edit, copy and clear property values on the enrolled record or an associated record; workflows are listed for Professional and Enterprise subscriptions. Human approval belongs in two places. One named owner signs off on any change to the hostname list, in both tools on the same day. And because HubSpot lets users change Original Traffic Source on a contact record, corrections go into the custom fields while the system value stays intact for audit.
The metrics that carry a referral through to qualified pipeline
A B2B metrics framework for AI search referral tracking follows each referral through qualified leads and into pipeline, pairing a count with a rate at every stage. GA4 supplies the visit stage: sessions and key event rate by channel group. HubSpot supplies the rest: new contacts by Original Traffic Source, the share of that cohort sales accepts as qualified, deals created and pipeline value, then win rate, average deal size and days from deal creation to close. Run identical rows for Organic search, and keep the counts beside the rates, since a rate on a small cohort can swing on a single lead.
How do AI-referred visitors convert compared with organic search?
To track and measure AI-referred traffic conversion vs organic search in the CRM, treat AI-referred visitors and organic visitors as two contact cohorts, defined by Original Traffic Source and created in the same window, then walk both through the same stages. The table below is a hypothetical quarter for a B2B software company. Every figure in it is an illustrative assumption, including the $15,000 deal value.
| Cohort measure (hypothetical) | AI Referrals | Organic search |
|---|---|---|
| Sessions | 900 | 30,000 |
| New contacts | 36 | 600 |
| Contacts per 100 sessions | 4.0 | 2.0 |
| Sales-qualified contacts | 9 (25%) | 108 (18%) |
| Deals created | 4 | 40 |
| Pipeline at $15,000 per deal | $60,000 | $600,000 |
| Pipeline per contact | $1,667 | $1,000 |
Every rate favors the AI cohort, while the counts advise patience. One lead is worth 2.8 points of a 36-contact cohort's qualification rate, and had two of the nine failed qualification, the rate would read 19.4%, close to the organic cohort's 18%. Compare cohorts of the same age so slower deals have time to appear, and hold any "converts better" claim until a couple of leads can no longer reverse it.
Which reports tie AI search traffic to pipeline?
To measure AI search traffic's impact on pipeline, conversion and attribution inside HubSpot, you have two levels of reporting, and your subscription decides the second. The first is a deal report grouped by Original Traffic Source, built on properties HubSpot lists as available on all products and plans. The second is attribution reporting: HubSpot lists Contact Create Attribution for Marketing Hub and Content Hub Professional and Enterprise, and limits Deal Create Attribution and Deal Revenue Attribution to Marketing Hub Enterprise accounts. Original Traffic Source credits the assistant only when it was the first known web source, which is why the AI influenced flag exists.
What counts as proof of qualified pipeline?
To prove AI search drives qualified pipeline rather than just traffic, report numbers that sales has already accepted, under three conditions. The qualified stage is set by a sales decision, such as an accepted lead or a held meeting, and applied identically to both cohorts. Each figure is labeled sourced (AI Referrals was the original source) or influenced (the deal flag is set), and the two are never added together. A person reviews a monthly sample of AI-tagged contacts before the report reaches finance. Source reports show association and cannot show what the buyer would have done without the assistant, so word the claim to match.
What attribution cannot see, and how to plan for it
An AI answer engine optimization strategy has to plan for attribution uncertainty, because several kinds of AI-driven demand arrive with no AI label or never arrive as a click at all.
- Google's own AI answers. GA4 counts AI Overviews and AI Mode in Organic Search, and HubSpot keeps regular Google and Bing search in Organic search. Search Console's generative AI performance reports, announced by Google Search Central on June 3, 2026 and rolled out to all websites as of August 31, 2026, show impressions by page, country, device and date. We found no click metric in that announcement. The organic cohort in your comparison already contains AI-answer visits, and we found no documented way to count them.
- Visits with no referrer and no parameter. HubSpot's final rule sends a session with no referring domain or tracking URL to Direct traffic, and GA4's AI Assistant medium depends on a referrer match. OpenAI documents its parameter for referral URLs from ChatGPT search results; we found no equivalent statement for other assistants in the documentation we read.
- Answers that end without a click. A buyer who reads a recommendation and later types your URL or searches your brand is recorded as Direct or Organic search.
- Blocked tracking. HubSpot notes that an ad blocker may block its tracking cookies and affect traffic source values.
Report the result in three tiers and keep them apart: tracked AI Referrals pipeline as the floor, self-reported AI pipeline with no tracked AI source as a second line, and directional indicators that carry no dollar value. Those indicators are generative AI impressions in Search Console, branded search and direct trends, and a sampled AI share of voice, whose limits we cover in where prompt-based tracking breaks. None of the documentation above supports a multiplier from the tracked floor to a true total, so state the gap as a gap. The published citation benchmarks give context on how often assistants link to sources at all.
Our AI Visibility Checker is a further resource for the visibility side, and our AI search optimization and data and analytics teams build this kind of reporting. Whatever stack you run, put the hostname list, the field definitions and the tier labels on one page that the GA4 owner and the HubSpot owner both sign. The pipeline comparison is only as trustworthy as that page.
Sources
- Google Analytics Help: What's new in Google Analytics (entry of May 13, 2026)
- Google Analytics Help: Default channel group
- Google Analytics Help: Custom channel groups
- HubSpot: Understand Original and Latest traffic source properties (updated August 4, 2026)
- HubSpot: Understand HubSpot's traffic sources in the web traffic analysis tool (updated August 1, 2026)
- HubSpot: Choose your workflow actions
- HubSpot: Create attribution reports (updated August 18, 2026)
- OpenAI: Publishers and Developers FAQ, OpenAI Help Center
- Google Search Central: Introducing Search Generative AI performance reports in Search Console (June 3, 2026)