AI search attribution: how to measure pipeline from ChatGPT and AI answers (2026)

AI search attribution is how you connect ChatGPT, Perplexity, Gemini and Google's AI answers to the leads and pipeline they create. No single report can do it. GA4 sees only the clicks that arrive with a source, and AI-assisted buyers often arrive as Direct or as a branded Google search. So you read four signals together. They are GA4's AI Assistant channel, a "How did you hear about us?" field, branded search in Search Console and CRM deal fields. Then you report pipeline as a range.

I'm Brenden House, and I run Searchmaxxed, an SEO and AI SEO agency. This is the setup we check on client accounts, including the bot problem that turned one enterprise client's "AI traffic" chart into fiction.

In summary

Measure AI search with four signals, never one. Report the result as a floor and a likely figure.

• GA4 undercounts AI search, and in October 2026 a ChatGPT tagging bug pushed its clicks into Direct for about a week.
• It can overcount too, because bots pass as AI visitors, so check every spike before it goes anywhere near a board pack.
• The cheapest fix is one optional free-text field on your demo form, carried into the CRM.
• Read the signals together and call anything none of them catches "dark", never Direct.

What is AI search attribution?

AI search attribution is the practice of tying revenue and pipeline to the moments an AI assistant shaped a buyer's choice. It covers the clicks AI tools send you, and the buyers who heard your name in an answer and came back another way.

A few terms you'll meet, in plain words:

• AI referral traffic is a visit that arrives from an AI tool with its source intact, such as chatgpt.com or perplexity.ai.
• Self-reported attribution is what buyers tell you themselves.
• Brand search lift is more people searching your company name after AI answers start naming you.
• The dark funnel is the research your analytics never sees: AI chats, Slack threads, podcasts, word of mouth, a colleague's recommendation.
• AI-influenced pipeline is deal value with at least one AI signal.

Attribution and visibility are different jobs. Visibility asks whether AI answers name you. Our guide to tracking AI visibility covers that side: the buyer questions to run, how often, and the metrics that hold up. This article starts where it stops, at the lead and the deal.

Why doesn't GA4 show most AI search traffic?

GA4 credits a visit to whatever source arrives with it. AI tools often send no referrer. Google's AI answers arrive looking like ordinary Google, so much of the real AI influence lands in Direct or Organic Search.

Here's where each kind of AI touch usually ends up:

The tag is fragile, too. Search Engine Land reported on 20 October 2026 that ChatGPT links stopped carrying utm_source=chatgpt.com after the GPT-6 release. Across about 500 pages, Adobe's Cedric Huesler measured ChatGPT visits falling by about 31,000 a day while Direct rose by about 26,000. OpenAI called it a known bug and began rolling out a fix on 15 October. If your AI line dipped that week, annotate 8 to 15 October and leave your attribution model alone.

For an enterprise or B2B SaaS team, rows three and five matter most. A buyer asks ChatGPT for a shortlist, sees your name, and three days later Googles you from a work laptop between meetings. GA4 calls that branded organic search. It's correct about the click and wrong about the cause.

How do you track ChatGPT and AI referral traffic in GA4?

GA4 does most of it for you now. On 13 May 2026, Google Analytics added an AI Assistant channel to its default channel groups. It catches visits from assistants such as ChatGPT, Gemini, DeepSeek, Copilot and Grok, the sources Google's help names, when the referrer matches. Clicks from AI Overviews and AI Mode aren't in it; they count as Organic Search.

So start with the AI Assistant channel in Traffic acquisition and your key event reports. Then check what it misses. A newer assistant, a regional tool or an AI browser can still land in Referral if Google's list doesn't know it yet.

To catch those, build a custom channel group, using the recipe on Google's custom channel groups help page:

1. In GA4, go to Admin, then Data display, then Channel groups, and copy the default group.
2. Add a channel named "Other AI" for the sources the default list misses.
3. Set the condition to Session source "matches regex" and paste a pattern for those sources.
4. Drag "Other AI" above Referral, or GA4 files those visits as ordinary referrals before your rule runs.
5. Save. Then add the group as a dimension in Traffic acquisition and your key event reports.

Two details trip people up. GA4 regex is full match and case sensitive by default, according to Google. That's why perplexity alone misses perplexity.ai, and why each pattern needs .* on both sides. There's also a limit: a standard GA4 property gets two custom channel groups, and a 360 property gets five. Channel group edits apply to past data in your reports, but custom channel groups don't appear in the BigQuery export. If your pipeline reporting runs on BigQuery, rebuild the same rule there.

Either way, the AI channel is a floor. It counts the AI visits that kept their source, and nothing else.

How do you filter bots out of AI referral traffic?

Check every jump in AI traffic before you report it. Look for one-hour bursts, one screen size, all new users, no referrer, a few seconds per visit and only one or two landing pages. Real buyers don't arrive like that.

Here's what we found on an enterprise B2B client's account. A client's GA4 showed AI-assistant visits jump from 354 to 6,050 in a month. Almost none of it was real. About 5,700 came in three one-hour bursts: same 1920x1080 desktop Chrome, all new users, no referrer, about 20 seconds each, landing on two pages. Real AI traffic was 385 sessions against 354 the month before, roughly flat.

A monthly report built on that chart would have shown AI traffic up seventeenfold and sent budget toward a channel that hadn't moved. Each trait in that burst becomes a check you can run:

When a burst fails these checks, build a segment that excludes it, then report the clean number beside the raw one. Raw 6,050 beside clean 385 tells the truth.

The same account had a second fault, and it sat at the other end of the funnel, where the leads should have counted. A client's four enquiry types never counted as conversions. Their event names ran 41 to 50 characters; GA4 cuts at 40. The 40-character "career" event counted every time. Nobody had noticed, because the events still showed up. Open your key events in Admin and confirm every form you care about counts, before you credit any channel with leads.

How do you capture self-reported attribution?

Add one optional, free-text field, "How did you hear about us?", as the last question on your demo or contact form. Then store the answer on the contact and the deal in your CRM. It catches what referrers lose.

Why free text and not a dropdown? A dropdown of Google, Referral, Social and Other leaves a buyer who found you through ChatGPT nowhere honest to click, so they pick Google. Rock The Rankings showed the gap in a B2B SaaS audit of 8,307 HubSpot contacts. The CRM's referrer-based source showed 11 leads from AI referrals, while the free-text field showed 809 buyers naming ChatGPT, Claude, Perplexity, Gemini or "AI".

Set the field up like this:

Self-reports have their own bias. Buyers remember the last or most familiar touch, so "Google" on the form is often a branded search that an AI answer started days before. Many buyers never fill a form either. Arcalea found that 72% to 84% of one B2B distributor's ChatGPT-assisted customers checked out as guests each month, from April to August 2026. A self-report field never sees those buyers. That's why it's one witness among four.

How do you measure brand search lift from AI answers?

Track impressions and clicks for your branded queries in Google Search Console, and compare matched periods before and after AI answers start naming you. When AI answers create demand, it shows up as people searching your company name.

Here's the routine I'd run:

1. In Search Console's Performance report, filter queries by regex for your brand name, product names and common misspellings.
2. Export branded impressions and clicks weekly.
3. Line those weeks up against your AI visibility tracking: the dates your company started appearing in answers to your buyer questions.
4. Compare matched 28-day periods, and year on year where you have the data, so seasonality and product launches don't fool you.
5. Note anything else that drives brand demand that month: a funding round, a conference, paid campaigns, PR.

Brand lift is a correlation, not proof. It earns its place in a pipeline report when it moves alongside the other signals. Self-reports naming ChatGPT, tagged AI referrals and branded searches all rising in the same quarter tell a consistent story your CFO can follow. Branded search rising on its own, in a month you raised a Series B, tells you little.

Which CRM fields do you need to attribute leads to ChatGPT?

You need first- and last-touch source fields, the landing page, the referrer, the self-reported answer and an AI-influenced flag on every deal. Without them, GA4 can count sessions but nobody can count pipeline.

We run this on our own site. Every enquiry through the searchmaxxed.com contact form carries its first-touch and last-touch source, medium, campaign, landing page and referrer into its CRM record. HubSpot and Salesforce can hold the same fields as contact and opportunity properties, as long as the values copy onto the deal. Pipeline reports read deals.

How do you turn these signals into a pipeline number?

Classify every new deal by which signals agree, add up deal value by class, and report a floor and a likely figure. Keep the dark share on its own line, so nobody mistakes it for Direct.

A worked example: a B2B SaaS company's quarter

I made these numbers up to show the arithmetic for a venture-backed B2B SaaS company selling security software, so swap in your own.

• 120 new opportunities this quarter, worth A$4.8 million in pipeline.
• 6 deals carry a chatgpt.com or perplexity.ai first touch, and in 4 of them the buyer also named ChatGPT on the form. That's 4 confirmed and 2 referral-only.
• 14 deals came in as Direct or branded Google search, but the buyer typed ChatGPT, Perplexity or "AI" on the form. That's 14 recovered.
• 30 deals have a blank form and a Direct source. Those are dark.

The floor is the 18 confirmed and recovered deals: at the A$40,000 average deal size, that's A$720,000, or 15% of new pipeline. Add the 2 referral-only deals and the likely figure is 20 deals, about A$800,000. The 30 dark deals sit on their own line. If branded impressions rose 30% the quarter AI answers started naming the company, that's support for the trend, not extra deals.

Last-click would have shown 6 deals at most. Arcalea found the same gap in real data. Tracing one B2B distributor's orders from May to August 2026, it found 49% of ChatGPT-assisted revenue came from orders ChatGPT didn't close. That's about US$56,000 of US$116,000, and last-click would have credited all of it to other channels.

Should you use last-click or multi-touch attribution for AI search?

Use last-click to judge the channels that close. Judge AI search with a multi-touch or influence view. AI answers usually act early in a long B2B cycle, so last-click credits them least.

Whichever model you pick, keep the definitions fixed for a full quarter, because answers shift between runs and B2B buyers take months to sign. For the wider monthly report this feeds, see what an SEO report should include.

How does Searchmaxxed measure pipeline from AI search?

Searchmaxxed is an SEO and AI SEO agency for enterprise, B2B SaaS, venture-backed, manufacturing and public-sector companies. Every managed package wires and checks Search Console and GA4, and tracks what ChatGPT, Gemini, Perplexity and Google's AI answers say about you. Then we measure the pages we ship against enquiries.

In practice, that means the setup in this article, done for you:

• GA4's AI Assistant channel, extended for sources it misses and checked for bot bursts before any number reaches your report.
• Key events tested, so every enquiry type counts.
• AI answer tracking on your buyer questions, lined up against branded search in Search Console.
• A monthly decision memo that reports what changed against competitors, read across 7-day, matched 28-day and matched 90-day views.

Measurement is half the job. The other half is getting named. We build on a website you own, our Agentic Website. AI agents pull your Search Console and GA4 data, read the Google results and check what AI answers say about you and your competitors. That feeds the Managed Search Loop, which shows which page, fix or link should bring in the most enquiries next. We build it, check the live page and measure what changed. Off your site, the Off-Page Source Layer builds the links, digital PR and mentions AI answers draw on. Our AI search optimisation service sets out the work in detail.

You own the site, repository, content and data from day one. Prefer to start smaller? The AI SEO Audit takes five business days and shows where buyers search without finding you.

Searchmaxxed's work has generated more than A$10 million in revenue through search for its clients.

Is Searchmaxxed right for you?

We suit marketing and growth leaders who need AI search counted in pipeline and want one team building the pages that get them named. If you only need the tracking set up, a good analyst can do it from this article.

If you're in the right-hand column, start with GA4, Search Console and one form field, which cost nothing, and add a tool later. Our comparison of AI visibility tools covers the software side.

Frequently asked questions

What should you do next?

Add the "How did you hear about us?" field to your demo form today. Then open GA4's AI Assistant channel, add any sources it misses, check it for bots, and line both up against branded search in Search Console. Give it a quarter, and you'll have a pipeline number you can defend. If you want us to set up the measurement and build the pages that get you named, let's talk pricing.