The short answer.
GA4 now has an AI Assistant channel and HubSpot now has an AI Referrals traffic source, so the clicks from ChatGPT, Gemini and similar tools can be counted without custom work. But those clicks are only part of AI’s influence: many buyers read an AI answer, then search your brand or type your URL later, and some app clicks arrive without a referrer at all. To report AI’s effect on pipeline you need two sources side by side: tracked AI referrals, and what buyers say when you ask how they heard of you.
This guide is about attribution: which source gets credit for a contact and a deal. If your problem is that GA4 and the CRM count different numbers of leads, start with our GA4 lead reconciliation audit. If you want to send CRM stages back to ad platforms, see offline conversion tracking for B2B SaaS. Both sit within our conversion tracking work.
- 01 · ANSWERBuyer reads an AI answer that names you
- 02 · VISITClicks through (GA4: AI Assistant) or arrives later as direct or branded search
- 03 · CONTACTFills a form (HubSpot: AI Referrals, or another source)
- 04 · ASKAnswers “how did you hear about us?”
- 05 · DEALBecomes an opportunity carrying both sources
Tracked source and self-reported source answer different questions. Keep both.
Why SaaS teams need this now.
Software buyers now use AI assistants during research as a matter of course, and they arrive at sales with a shortlist already formed. If your reporting cannot see that stage, it will undercount the channel that shapes the shortlist.
6sense, Breakthrough recap; TrustRadius 2026 B2B Buying Disconnect (1,862 technology buyers, January 2026); G2 AI search survey (1,000+ software buyers, August 2025, published 6 October 2025); G2 Buyer Behavior 2025 (1,169 decision-makers, April 2025). Different questions and samples; not directly comparable.
The surveys ask different questions, but they agree on direction. G2’s August 2025 survey found that half of software buyers now start the buying journey in an AI chatbot, a 71% jump in four months. TrustRadius found in 2026 that 63% of technology buyers used AI during their purchase, and 94% of those fact-check its answers at least some of the time, often against reviews, which 74% of buyers use.
Timing matters too. 6sense’s 2025 Buyer Experience Report, summarised by CustomerThink, found buyers made first contact 61% of the way through their journey, and bought from their top-ranked vendor 77% of the time. The shortlist forms before anyone fills in your form, which is exactly where AI answers sit. Getting a product into those answers is a separate job, which we cover in our guide to GEO for B2B SaaS.
What AI referral traffic looks like.
For most B2B SaaS sites, AI referrals are a small share of sessions and a much larger share of sign-ups or enquiries. That pattern is why session counts alone undersell the channel.
The clearest published example is Ahrefs’ analysis of its own site, published in June 2025: AI search traffic was 0.5% of its visits over 30 days but produced 12.1% of sign-ups, which Ahrefs described as a conversion rate 23 times higher than traditional organic search.
Ahrefs, analysis of its own website, published 16 June 2025. One company’s data; your ratio will differ.
One company is not a benchmark, and Ahrefs’ audience is unusually close to the subject. Treat it as evidence of a pattern to test in your own data, not a number to quote in a board deck. On the wider market, Similarweb reported in July 2026 that worldwide visits to generative AI platforms grew 70% year on year to 9.5 billion, while ChatGPT’s share of that traffic slid from roughly 76% to around 53%. Your AI referrals will increasingly come from more than one assistant.
What GA4 shows you.
GA4 now classifies AI assistant visits in its default channel group, so you can see them in standard reports without building anything. Check how your own traffic lands, then add a custom channel group as a backup.
Google’s “What’s new” log records that on 13 May 2026, “Google Analytics now provides a dedicated way to measure and analyze traffic originating from popular AI assistants.” The default channel definitions now include an AI Assistant channel, applied when the medium is “ai-assistant” or when “the referrer matches a list of AI Assistants”. Google’s examples include ChatGPT, Gemini, DeepSeek, Copilot and Grok. On 11 June 2026 GA4 added a Source Group dimension with “built-in grouping for emerging traffic sources like ChatGPT (OpenAI) and Perplexity”.
Two details catch teams out. First, Google’s AI Overviews and AI Mode are not in the AI Assistant channel; they count as Organic Search. Second, Referral and Direct remain separate channels, and AI visits with no referrer will land in Direct.
| Use | What it does | Watch for |
|---|---|---|
| AI Assistant default channel | Groups visits from known AI assistant referrers | Google AI Overviews and AI Mode are counted as Organic Search |
| Source Group dimension | Consolidates source values for platforms such as ChatGPT and Perplexity | New in June 2026; check it in your reports before relying on it |
| Custom channel group | Your own regex rule for AI sources, placed above Referral | Standard properties get two custom groups; they apply retroactively |
| First user source | Where a new user first came from | Use for acquisition; session source for each visit |
| Key events | Mark demo requests and sign-ups as key events | Count one key event per real enquiry |
Google’s own help page for custom channel groups includes an AI assistants example using a “matches regex” condition, placed above Referral so AI sources are caught first. Our short play on tracking AI referrals in GA4 walks through it.
What HubSpot shows you.
HubSpot now records AI Referrals as its own traffic source on contacts, with the AI platform’s domain as the drill-down. That gives you a CRM-native way to follow AI-referred contacts into deals.
HubSpot’s knowledge base lists the Original and Latest Traffic Source values as Organic search, Paid search, Email marketing, Organic social, Referrals, AI Referrals, Other campaigns, Direct traffic, Paid social and Offline sources. Drill-down 1 for AI Referrals records the “AI platform domain (e.g., chatgpt.com, claude.ai)”. HubSpot’s own example: a contact who visits “by clicking a link from ChatGPT” gets Latest Traffic Source “AI Referrals” and Drill-Down 1 “chatgpt.com”.
The feature came from customers. A HubSpot Community idea posted in June 2025 asked HubSpot to “parse out leads from ChatGPT, Claude, Perplexity, etc.” into a new Original Source category; HubSpot later replied that it “went GA at INBOUND 2025”.
Attribution reports and their limits
HubSpot offers contact create, deal create and revenue attribution reports. Deal create and revenue attribution need Marketing Hub Enterprise. The models are First Touch, Last Touch, Linear, Time Decay and Empirical; HubSpot says “The Empirical model replaces the U-shaped, W-shaped, J-shaped, and Inverse J-shaped models.” Without Enterprise, copying the source onto the deal, as in chapter 8, gives most teams what they need first.
The dark-funnel gap.
The “dark funnel” is the research buyers do where your analytics cannot see it. AI assistants add to it in two ways: answers that never produce a click, and clicks that arrive without saying where they came from.
HubSpot’s own glossary says answer engines “don’t always pass referrer headers, which causes analytics tools to bucket those sessions under direct traffic rather than a named source.” One measurement vendor, Airefs, reports that on its own data ChatGPT accounted for about 2% of GA4 traffic while over 20% of new sign-ups cited ChatGPT or AI when asked. That is vendor data from a company selling a fix, so treat the size of the gap with caution; the direction matches what HubSpot describes.
Airefs, Why GA4 under-reports ChatGPT traffic, published 14 July 2025, updated 28 April 2026. Vendor’s own data; directional only.
Zero-click answers add to the gap. Pew Research’s 2025 study of 900 US adults’ browsing found that when Google showed an AI summary, users clicked a traditional result in 8% of visits, against 15% when there was no summary, and clicked a link inside the summary in just 1% of visits. Buyers who read an answer and move on still carry your name into their shortlist.
The gap is not a flaw you can fully fix. It is a reason to add a second, independent source of evidence.
Self-reported attribution.
Self-reported attribution means asking buyers how they heard of you, and storing the answer in the CRM next to the tracked source. It catches the influence that never produces a labelled click, at the cost of being imprecise.
How to set up the question
- One question on the demo or contact form: “How did you hear about us?”
- Open text, or a short list with “Other” and a text box. If you use a list, include “ChatGPT or another AI assistant” as its own option, separate from “Google search”.
- Required on high-intent forms only, such as demo requests, so free-trial sign-ups stay quick.
- Store it as a contact property and copy it onto the deal when the deal is created.
- Ask again on the first call, and log the answer; people remember better when asked by a person.
- Clean it monthly: group free-text answers into a small set of values without overwriting the original text. This clean-up is a fair candidate when deciding what to automate first with AI.
Read the two sources together, not against each other. A contact whose tracked source is Direct and whose answer is “ChatGPT recommended you” is evidence of AI influence the click data missed. A contact whose tracked source is AI Referrals and whose answer is “a colleague” tells you the AI answer was a check, not the start. Our play on asking leads how they heard about you has field wording you can copy.
Building the pipeline report.
The report you want shows, for each period, how many contacts, opportunities and won deals had AI involvement, by tracked source and by self-reported source, side by side. Build it in the CRM, because that is where opportunities live. Our guide to website lead tracking applies the same approach to every other source.
- 01 · CAPTUREGA4 AI Assistant channel and HubSpot AI Referrals both on
- 02 · ASK“How did you hear about us?” stored as a contact property
- 03 · CARRYCopy original source, drill-down and self-reported answer onto the deal
- 04 · FLAGCreate an “AI involved” deal flag: tracked, self-reported, both or neither
- 05 · REPORTDeals and pipeline value by flag, monthly and quarterly
Step 3 is the one most teams skip; without it, the deal forgets where the buyer came from.
| AI involvement | New contacts | Qualified | Opportunities | Pipeline value | Won |
|---|---|---|---|---|---|
| Tracked only (AI Referrals source) | from CRM | from CRM | from CRM | from CRM | from CRM |
| Self-reported only (said AI, other source) | from CRM | from CRM | from CRM | from CRM | from CRM |
| Both | from CRM | from CRM | from CRM | from CRM | from CRM |
| Any AI involvement (total) | sum | sum | sum | sum | sum |
| All other contacts | from CRM | from CRM | from CRM | from CRM | from CRM |
Add a GA4 view alongside it: AI Assistant sessions and key events by source, so you can see which assistants send visits even before they become contacts. Our play on connecting AI-referred leads to closed revenue covers the HubSpot report settings.
Reading the report honestly.
Present AI’s contribution as a range, not a single number: tracked AI as the floor, and tracked plus self-reported as a fuller estimate. Say plainly what each number can and cannot prove. We apply the same reading to evidence from one B2B client in a separate article.
- Tracked AI referrals prove a click from an AI assistant, nothing more. They undercount.
- Self-reported AI captures influence without a click, but memory is imperfect and answers are often partial.
- Branded search and direct growth can reflect AI exposure, but also events, PR and sales activity. Use them as context, not proof.
- Small numbers move a lot quarter to quarter. Show counts as well as percentages.
When the board asks “how much pipeline came from AI?”, the honest answer is “at least this much tracked, and about this much when we include what buyers tell us”. Our guide on reporting AI visibility to the board shows how to show that uncertainty on one page, and turning AI visibility into qualified enquiries covers what to do once you can see the flow.
The gap calculator.
Enter one quarter of your own CRM numbers to see how much AI influence your tracked source misses, and what that might mean for opportunities. Nothing you enter is sent anywhere.
The default values are placeholders to show the arithmetic. The key assumption is that people who answered the question resemble those who did not; if only demo requests are asked, the self-reported share will skew towards high-intent buyers. Say so when you present it. To put a value on that influence, use our AI visibility ROI model.
Pitfalls that skew the numbers.
Most errors in AI referral reporting come from a few settings and habits. Check them before you trust a quarter’s figures. Checks like these belong in any conversion tracking setup.
A pre-report checklist
- GA4 data retention. It only affects explorations and funnel reports, but the options are 2 or 14 months on standard properties. Set 14 months if you compare quarters in explorations.
- AI Overviews hidden in Organic. Google’s AI features are counted as Organic Search, not AI Assistant. Do not report the AI Assistant channel as “all AI”.
- Your own UTM tags. GA4 assigns channels from source and medium, so links you tag yourself (in AI directories, for example) are reported under your tags. Keep a naming rule.
- Source lost at the deal. If the original source and self-reported answer are not copied to the deal, opportunity reports will not show them.
- Double-counted key events. If GA4 counts two key events per enquiry, AI conversion rates look inflated. Our GA4 lead audit covers the fix.
- Tiny samples. Five AI-referred deals in a quarter is a signal to watch, not a trend.
If you also want ad platforms to learn from these deals, the stages and click IDs in our offline conversion guide use the same CRM properties, so set them up once.
Questions teams ask.
How do I track ChatGPT traffic in GA4?
GA4’s default channel group now includes an AI Assistant channel for visits from known AI assistant referrers such as ChatGPT and Gemini. Check it in the traffic acquisition report, and add a custom channel group with a regex rule above Referral as a backup for sources it misses.
Why does AI traffic show up as direct?
Some AI tools and apps do not pass a referrer when a link is opened, so analytics records the visit as direct. HubSpot’s own glossary describes this. Buyers who read an answer and later type your URL also appear as direct.
Does HubSpot track AI referrals?
Yes. HubSpot’s Original and Latest Traffic Source properties include an AI Referrals value, with the AI platform’s domain, such as chatgpt.com, in Drill-Down 1. You can report contacts and deals by that source.
Are Google AI Overviews counted as AI referrals?
No. In GA4’s default channel group, visits from Google’s AI Overviews and AI Mode are counted as Organic Search, not AI Assistant. Report them separately, or describe the AI Assistant channel accurately as non-Google assistants.
Is self-reported attribution reliable enough to report?
It is imprecise but useful. It captures influence that never produces a tracked click. Report it next to tracked source, not instead of it, show counts as well as percentages, and wait for at least a few dozen answers before acting on the share.
Sources: GA4 default channel group; GA4 what’s new; GA4 custom channel groups; GA4 traffic-source scopes; GA4 data retention; GA4 key events; HubSpot traffic source properties; HubSpot Community idea; HubSpot attribution reports; HubSpot glossary; Airefs; Ahrefs; Pew Research; Similarweb; CustomerThink on 6sense 2025; 6sense recap; G2, May 2025; G2, October 2025; TrustRadius 2026. Accessed 10 November 2026.