AI Search drives signups that Google Analytics never sees. At OtterlyAI, ChatGPT and Claude together account for more than 21% of our self-reported signups, yet GA4 last-touch attribution credits Claude with just 0.1%. This article shares the exact numbers, the method behind them, and why last-touch analytics undercounts AI search as a channel.
I’m writing this for two reasons. First, Ethan Smith and the Graphite team published a sharp analysis of n8n’s AI search statistics, and it pushed me to open up our own data.
Second, I get asked about ROI constantly: what is the ROI of AI search, and how do you measure it? So I decided to share our numbers at OtterlyAI. I hope it nudges other teams to dig into their own reporting and judge AI search on evidence rather than vibes.
TL;DR: Key Takeaways
We compared GA4 last-touch conversion data against self-reported attribution from our post-signup survey. Three findings stood out:
GA4 last-touch credits Google Search with 35% of signups, ChatGPT with 7%, and Claude with 0.1%.
Self-reported attribution tells a different story: ChatGPT 11%, Claude 10.6%, Google Search 15.7%.
Visitors who did arrive from ChatGPT converted at a 66% higher rate than organic search visitors.
The gap between these two views is the story. AI Search behaves like a brand-awareness channel, so click-based analytics misses most of its influence.
The AI Search Attribution Gap: Why Last-Touch Undercounts ChatGPT and Claude
OtterlyAI is a fairly typical product-led SaaS business. Free trial signups are our primary conversion event. We measure them two ways. Google Analytics 4 tracks website traffic and conversions on a last-touch basis, and a post-conversion survey asks every new user how they first heard about us.
These two methods answer different questions. GA4 tells you where the last click before signup came from. The survey tells you what the customer remembers as the source of discovery. For most channels the two roughly agree.
For AI search, they diverge hard.
Google Analytics 4 Conversion Data
According to GA4, Google Search drove 35% of all signups. ChatGPT accounted for 7%. Claude was effectively invisible at 0.1%. No other AI engine held a meaningful share.
On the surface, that reads like AI search barely matters. Keep going.
Conversion rates tell a sharper story
The more revealing cut was conversion rate from visitor to signup. Visitors who arrived from ChatGPT converted at 25%, compared to 15% for organic search visitors. That’s a 66% higher conversion rate. Claude and Gemini visitor volumes were too low to be statistically meaningful, so treat those two rows as directional only.
Channel
Visitor-to-signup conversion rate
Google Search
15%
ChatGPT
25%
Claude
7% (low volume)
Gemini
7% (low volume)
All traffic sources
11%
The takeaway: AI search visitors are fewer, but they arrive with higher intent. Someone who clicks through from an AI answer has usually already been told what the product does and why it fits.
Self-reported attribution changes the picture
Here is where it gets interesting. In our “where did you hear about us” survey, Claude represents 10.6% of all signups, right behind ChatGPT at 11%. Google Search, meanwhile, shrank to 15.7%.
Channel
GA4 last-touch
Self-reported
Google Search
35%
15.7%
ChatGPT
7%
11%
Claude
0.1%
10.6%
Claude went from a rounding error to nearly one in nine signups, depending purely on which measurement you trust.
Why last-touch and web analytics miss AI search
When someone asks an AI engine a question and a product like OtterlyAI comes up, that mention works mostly as brand exposure. There may be a link back to our site, but most people don’t click it. They add OtterlyAI to a mental shortlist and return later through another channel, often a direct visit or a branded Google search.
Much of that visibility goes unattributed. AI search functions more as a brand-awareness channel than a click-driven performance channel, so any last-touch model will systematically undercount it.
The ChatGPT vs Claude gap explained
The most striking difference is between ChatGPT and Claude. ChatGPT showed up in GA4 with a real share of signups. Claude was almost absent. Why?
My best explanation is the interface.
ChatGPT started hyperlinking brand mentions in May 2026, which raised the click-through rate on those mentions.
Claude, by contrast, shows few or no visible outbound citations inside the answer text. Same influence on the buyer, very different amount of trackable click traffic. The screenshots below show ChatGPT hyperlinking a brand mention while Claude presents the same information with no clickable citation.
This is the mechanism behind the whole article. When an engine doesn’t link out, its influence lands entirely in the self-reported column and nowhere else.
The Business Impact of AI Search
If we trust self-reported attribution, a large and roughly equal share of our conversions is driven by Google, ChatGPT, and Claude together.
Claude is the standout. Nearly absent in web analytics, it jumped to 10.6% in self-reported attribution, more than a hundredfold difference from its 0.1% GA4 figure.
ChatGPT rose to 11% in the survey, roughly 55% higher than its GA4 share. In the same shift, fewer people named Google as the place they first heard about us, which fits a world where discovery starts on AI and finishes on Google.
Personal Reflection
Taken together, the data makes sense to me. I believe GA4 shows only a slice of the real buyer journey. G2 reports that more than 51% of B2B software buyers now start their research on AI, and our numbers line up with that behavior.
Our customers start researching on Claude and ChatGPT, and often still on Google, then continue the journey elsewhere before they convert. They tend to land on our site through Google organic, a direct visit, or another channel, rather than clicking straight from Claude or ChatGPT. The first touch is increasingly an AI engine; the last touch is usually something else.
What surprised me most was the size of the Claude gap between last-touch and self-reported, and the fact that Claude now pulls roughly even with ChatGPT. It tracks, though. Claude has seen fast global adoption, and B2B tech buyers in particular have moved toward it.
What this means for your measurement
If you rely on last-touch analytics alone, you are almost certainly undervaluing AI Search. Three practical steps:
Add a self-reported attribution question at signup. One field, asking how the user first heard about you, exposes channels that click tracking cannot see.
Compare the two views side by side. The gap between last-touch and self-reported is your AI-search blind spot, and it is measurable.
Treat AI Search as a brand-awareness channel. Optimize for being mentioned and recommended inside AI answers, not only for referral clicks.
Next, I’m going to look at revenue attribution. There are some real surprises in that data, including things I did not expect. I’ll share those numbers in a follow-up.
