AI referral budgets need an absolute conversion share
In short: AI referrals can convert well and still contribute little, so budget them with absolute qualified conversions and attribution confidence, not rate alone.
Marie Haynes reported on 21 October that ChatGPT visitors converted at 13.8% across 55 sites with conversion tracking, against 3.02% for Google organic. On those rates alone, the budget decision looks easy.
Across her sample, however, Google organic still produced 96.5% of conversions. An AI-referral budget needs the channel's absolute share of qualified conversions beside its conversion rate. Otherwise a small numerator can wear a very flattering percentage.
Three studies agree on size, not one benchmark
Haynes analysed 12 months of GA4 data across 100 sites and more than 145 million sessions. Her study found that ChatGPT sessions were 0.62% of Google organic volume. All measured LLM referrals together reached 0.73%.
The SEO Works used a different sample and definition. Its 95 sites recorded 72,095 AI-referral sessions and 2,825,594 organic sessions from September 2025 to August 2026. Organic supplied 39 times more traffic. In the thresholded 63-site conversion subset, AI referrals converted at 7.18% versus 6.08% for organic, but lost to organic on 23 sites and across the B2B product group.
| Published sample | AI-referral volume | Conversion evidence | Boundary |
|---|---|---|---|
| Haynes, 100 sites | 0.73% of Google organic for all measured LLMs | 13.8% ChatGPT vs 3.02% organic on 55 sites | Google organic still supplied 96.5% of conversions |
| The SEO Works, 95 sites | 72,095 AI vs 2,825,594 organic sessions | 7.18% AI vs 6.08% organic on 63 sites | AI converted worse on 23 sites |
These are not rival estimates of one universal market rate. The properties, referral rules, conversion definitions and inclusion thresholds differ. The useful agreement is narrower: a higher rate does not erase a much smaller channel.
Direct traffic is uncertainty, not hidden AI evidence
A common reply is that AI influence disappears into Direct. That can happen, but the category cannot prove which source disappeared. Google Analytics defines direct traffic as traffic without a clear referral source; stripped parameters, offline documents, typed URLs and blockers can all contribute.
Kevin Indig tested the hidden-traffic idea across 12 GA4 properties. His analysis found total Direct sessions up 140.9% from January to September while engaged Direct stayed roughly flat. One property logged 409,000 Direct sessions, but only 198 were engaged. Identifiable AI traces averaged 0.003% of Direct traffic.
Indig also ran 632 comparisons across four brands, several AI-visibility measures and delays. Some correlations looked strong until wrong date alignments received the same chance to search for a peak. He could not establish a reliable relationship between AI visibility and later Direct or branded-search traffic in that sample.
That is an attribution limit, not proof that AI has no influence. Self-reported discovery and controlled customer research can add evidence. Quietly assigning unknown traffic to the newest channel cannot.
My budget rule uses a four-row receipt
This is editorial judgment, not a marketing experiment I ran at Trigent or for a client. For each channel, I would put four rows in the monthly review: referral volume, conversion rate, absolute qualified conversions and attribution confidence.
The confidence row should name the evidence. A tagged referral is observed. A buyer saying an assistant introduced the company is self-reported. A lift inferred from two rising trend lines is a hypothesis until a stronger design separates it from other causes.
This receipt leaves room for a small strategic experiment. A channel can teach you about emerging buyer behaviour before it wins on volume. The experiment should have a declared cost and outcome rather than inheriting budget from a dramatic relative rate.
What's in it for you
- Put absolute qualified conversions beside every channel conversion rate.
- Label unattributed traffic as unknown until another method supplies evidence.
- Fund small AI-visibility tests with explicit outcomes before moving core acquisition budget.
A high conversion rate earns attention; absolute qualified conversions earn budget.
