Attrifast's 200-Site Cohort Estimates 34% of GA4 Direct Traffic Was AI-Referred

In a cohort of 200 Stripe-connected sites, Attrifast estimated that a median 34% of GA4 Direct/(none) traffic was actually AI-referred. The pattern reverses on ecommerce, where Google organic converted at 2.1% against 1.6% for AI. This is one vendor's cohort of bootstrapped SMB sites that chose to join it, and the 34% is a modeled estimate rather than an observed count, so read it as evidence that the Direct bucket hides AI-referred demand, not as a figure for your site.

The Revenue Data Your Analytics May Be Hiding

Machine Relations rests on a thesis: earned authority in AI engines should produce higher-quality customers, and the standing objection has been that the channel cannot be measured. GA4 cannot distinguish a ChatGPT-referred visitor who arrives without a referrer from someone who typed your URL directly, so those sessions are bucketed as Direct/(none).

Attrifast's method estimates the AI share of that bucket with server-side fingerprinting joined to Stripe payments. Attrifast reports that its AI-attributed session share correlates at r=0.78 with SimilarWeb's same-window estimate on the 32 sites where it also had SimilarWeb data, "positive and meaningful but not perfect." Its current measurement guide is stricter: it says a Direct landing-page pattern is a diagnostic signal that must never be reported as a known AI visit, and that a modeled share of Direct should not be added to measured AI traffic and presented as observed. Treat every figure below as that kind of estimate.

Revenue per visitor and conversion by engine, for the 118 B2B SaaS sites in the cohort:

Engine	RPV	Conversion Rate
Claude	$1.94	4.7%
Perplexity	$1.81	4.1%
ChatGPT	$1.04	3.2%
Gemini	$0.49	1.6%
AI Overviews	$0.36	1.4%
Google organic (reference)	$0.71	1.7%

By this table Claude visitors converted at about 2.8x the Google organic reference and Perplexity visitors at about 2.4x. Attrifast's headline blended comparison for the same sites uses a 1.4% organic rate, which is where its 1.9x comes from, so the exact multiple depends on which Attrifast figure you hold fixed. Across all 200 sites the ranking differs: Perplexity led at $1.42 per visitor, then Claude $1.18, ChatGPT $0.87, Gemini $0.41 and AI Overviews $0.29.

What the Cohort Says About Customer Quality

On the SaaS sites, Attrifast reports average first-month subscription value of $57.40 for Claude-sourced customers, $51.20 for Perplexity and $44.10 for ChatGPT against $28.70 for Google organic (ChatGPT is a 54% premium), while Gemini ($23.40) and AI Overviews ($21.90) came in below organic. Blended across AI engines, 30-day SaaS churn was 9.2% against 14.4% for Google organic, which is 36% lower. On ecommerce sites, AI-sourced customers had a blended 30-day refund rate of 3.8% against 6.1% for organic. These are Attrifast's own cohort numbers, and they sit beside lower ecommerce conversion for AI traffic (1.6% against 2.1% for Google organic).

The mechanism is plausible. When someone asks ChatGPT "what is the best tool for X" and ChatGPT names your product, that person may arrive pre-informed. Attrifast describes that as an intent-quality hypothesis; the cohort data is consistent with it but does not isolate it.

Scrunch's prompt-to-purchase analysis looked at millions of search events from an opt-in panel between February and May 2026. When an AI platform recommended a brand to someone with no recent sign of using it, that person became about 182% more likely to search the brand on Google within a week (3.28% to 9.24%), 117% more likely to visit the brand's website (3.43% to 7.45%) and 185% more likely to view it on a retailer's product page (0.94% to 2.68%).

Position and framing inside the answer mattered in Scrunch's data. Brands named first were 389% more likely to be searched, brands named mid-answer 148% and brands named last 50%. Brands recommended as "best" or "top" were 425% more likely to be searched, against 128% for plain-language recommendations.

The Channel Your Dashboard Cannot See

In Attrifast's cohort, AI sessions land deep. Its landing-page table shows 64% of AI-engine sessions landing on a deep blog or content page, 12% on the homepage and 8% on top-nav inner pages, against 47% on deep pages for Google organic and 13% for Direct once the modeled AI share is removed. Blog posts, guides and comparison articles are the entry point, and most analytics dashboards treat them as organic long-tail visits.

AI search is not necessarily driving people to your front door. It may be sending them straight to the page where the answer lives. In the cohort, AI sessions on deep blog pages converted at 3.2% when the page had an inline call to action and 0.9% when it did not, a 3.5x gap on the same landing-page type.

Attrifast also reports that AI-attributed sessions across the cohort grew at a compounded 13.4% per month between December 2025 and May 2026, against 1.1% for Google organic on the same sites. That is a six-month window on self-selected sites, not a forecast.

What Earned Media Can Claim From This

Attrifast's B2B SaaS sites got 9.4% of total sessions and 13.7% of Stripe-attributed revenue from AI engines, so the revenue share ran about 46% above the traffic share. That is a correlation in one cohort. It does not show that any brand's earned media produced those sessions, and the data does not compare brands by how they earned their citations.

What it supports is narrower and still useful for Machine Relations: AI-referred visitors in this cohort were worth measuring on their own revenue line, and the visibility metrics that come before a click (mentions, citations, share of voice) need a revenue-side check. Earned authority that becomes citation architecture should show up there over time, and that is what to test on your own site.

Three Things to Do This Week

1. Measure the AI traffic you can verify, and label the rest. Build a GA4 channel group for recognized AI referrers and report it separately from Direct. Trakkr's measurement guide separates revenue from recognized AI-referred sessions, which is directly observable, from assisted conversions after a later return through another channel, which is directional, and treats Google AI Overviews and AI Mode as Search exposure rather than chatbot referrals.

2. Audit your deep-page conversion path. If AI visitors land on content pages on your site too, check the five blog posts and guides that get the most non-branded traffic. Do they have a CTA, a next step, a way to capture the visit? In Attrifast's cohort that one change separated 3.2% from 0.9%.

3. Check where your brand appears in AI answers. Open ChatGPT, Perplexity, Claude and Gemini and ask the buying question your best customers ask before they find you. If you are not named, or you are named last, Scrunch's panel data suggests the downstream behavior differs sharply. That is an earned media architecture problem, and it is the kind of problem Machine Relations exists to solve.

FAQ

How much of my direct traffic is actually from AI search?

No source gives a reliable figure for your site. In Attrifast's own cohort of 200 Stripe-connected SMB sites the median was an estimated 34% of GA4 Direct/(none) traffic, with an interquartile range of 21% to 47% across sites; by vertical the medians were 41% for B2B SaaS, 33% for services and agencies, 28% for creators and publishers and 22% for ecommerce. The estimate comes from server-side fingerprinting on sites that chose to join the cohort. Source: Attrifast, May 2026

Which AI engine sends the most valuable visitors?

In Attrifast's cohort, Claude led on B2B SaaS sites with $1.94 revenue per visitor and a 4.7% conversion rate, and Perplexity was second at $1.81 and 4.1%. All three beat the Google organic reference of $0.71 and 1.7% there. Across all 200 sites Perplexity led at $1.42 per visitor. Source: Attrifast, May 2026

Does AI search actually drive new customers or just recirculate existing ones?

The evidence here is about behavior after an AI recommendation, not about customer status. Scrunch's analysis measured people with no recent sign of using a brand and found that, after an AI platform recommended it, they became about 182% more likely to search the brand on Google within a week. That is top-of-funnel discovery behavior in an opt-in panel; it does not show how many resulting customers were new to the brand.