Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- AI Search
AI referrals convert better than organic in many B2B GA4 accounts. Is that real intent quality, or a small-sample, late-funnel selection effect?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
Probably some of both, and the GA4 comparison on its own cannot tell you the split. A higher conversion rate for AI referrals is what you would expect even if AI visibility created no extra demand at all. The assistant answers the early research questions inside the chat, so the people who click through are more likely to be late-stage, to know your brand already, and to land on pricing, product or comparison pages. Add small counts and a channel that is only partly measured, and a 3x or 10x gap can appear even if the channel sends you no buyers you would not have reached anyway.
The distinction that matters for budget is between intent quality and incrementality. Assistants pre-qualifying buyers is a real effect, so in that sense the intent quality is real. What a conversion rate cannot show is whether those buyers would have reached you anyway, through branded search or a direct visit. That second question is the one a business case needs answered, and it needs evidence from beyond GA4.
Why the comparison is tilted before you look at the data
Comparing the two rates is like comparing the close rate of people who phone your sales line after a long conversation with a trusted adviser against the close rate of everyone who picked up a brochure at your trade-show stand. The callers convert better, mostly because of where they are in the decision, not because the phone is a better channel.
The mechanism: organic sessions include the research phase (glossary pages, how-to guides, “what is” queries), and research-phase sessions convert at low rates almost by definition. For people using an assistant, much of that research happens inside the chat and does not produce a session on your site. The clicks that do arrive are likely to be weighted towards checking a shortlist, verifying pricing and taking action. Both channels can serve the same buyer at different moments; the AI slice simply starts later in the journey.
One public example fits the pattern. In June 2025, Ahrefs reported that AI search sent 0.5% of its visitors over 30 days but 12.1% of its signups, a conversion rate about 23 times that of traditional organic search, and suggested those visitors could be further down their journey and ready to act. That is a genuine reading of one site’s data: a well-known brand, a short window, and a channel that made up 0.5% of traffic. It is consistent with both explanations in your question, which is why a conversion multiple on its own cannot separate them.
Four effects that can inflate the ratio
1. Small numbers produce wide intervals
Hypothetical example:
Over a quarter, AI Assistant sessions produce 9 demo requests from 180 sessions (5.0%), and organic search produces 350 from 24,000 (1.46%). The headline says AI converts 3.4 times better. But the 95% interval around 9 in 180 runs from about 2.7% to 9.2%, while the organic rate sits between 1.3% and 1.6%. Three demo requests more or fewer move the AI rate between 3.3% and 6.7%. At this size the true ratio could plausibly be anything from under 2x to over 6x, before any of the effects below.
B2B conversion counts carry extra noise: spam form fills, existing customers using the demo form, and several people from the same company. Clean those out before comparing, because at small counts two junk submissions move the ratio.
2. Landing-page mix
The same hypothetical account, split by the type of page each session landed on:
Landing page type Organic sessions Organic demos Organic rate AI sessions AI demos AI rate Blog and guides 20,000 120 0.6% 30 0 0% Product and feature pages 3,000 150 5.0% 90 4 4.4% Pricing and comparison pages 1,000 80 8.0% 60 5 8.3% Total 24,000 350 1.46% 180 9 5.0% Within each page type the rates are close, and the AI figures are too small to separate from organic anyway. The whole headline gap comes from where the sessions land. This is a composition effect, the everyday form of Simpson’s paradox, and it is the first explanation I would rule out.
3. Brand familiarity and navigation
Some AI sessions start from prompts such as “[your product] pricing”, “[your product] vs [competitor]” or “does [your product] integrate with Salesforce”. Those users already know you, so the fair comparison is branded organic search, not all organic. GA4 cannot split organic sessions by query, so use landing pages as a proxy, and Search Console’s branded queries filter to see how much of your organic demand is branded. Existing customers need excluding too: assistants can send them to help docs and login pages, and if those sessions trigger key events such as trial activations, the rate is inflated.
4. The channel is only partly measured
GA4 added an AI Assistant channel to its default channel group in May 2026. Google’s current channel definitions describe it as covering sources such as ChatGPT, Gemini, DeepSeek, Copilot and Grok, assigned when the referrer matches Google’s list of AI assistants, and state that it excludes Google’s AI Overviews and AI Mode. The help page gives examples rather than the full list.
Two consequences follow. Visits from assistants that arrive with no referrer and no campaign tags (from some apps, or from copied links) land in Direct. And clicks from AI Overviews and AI Mode sit inside Organic Search; practitioners report AI Mode clicks arriving as google / organic. So the AI Assistant segment is the subset of AI-influenced visits that kept a referrer, while Organic Search already contains some AI-assisted visits. The direction of the resulting bias on conversion rates is unknown, which is one more reason not to read the ratio literally.
Expect the segment to be small and dominated by one assistant. BrightEdge data released in September 2026 puts ChatGPT at about 95% of tracked AI referral traffic, with AI still a small share of referrals compared with search, although the methodology is not detailed.
How to find out which explanation fits your account
- Fix the definition. List every source/medium row containing an AI domain across AI Assistant, Referral, Unassigned and campaign-tagged traffic (links from ChatGPT can carry
utm_source=chatgpt.com). Then build a custom channel group with explicit rules, so the segment is defined by your rules rather than by whichever version of the default grouping was live at the time. - Pool data and put intervals on it. Use a quarter or more. As a rough guide, wait until the AI segment has a few dozen conversions: at 30 conversions from 600 sessions the 95% interval still runs from about 3.5% to 7.0%, a twofold range, and with fewer conversions it is wider.
- Match landing pages. Compare conversion rates within page types (pricing, product, comparison, blog, docs). If the gap disappears, it was mix.
- Separate new from returning users, and first touch from last touch. Comparing GA4’s first-user and session-level channel dimensions shows whether AI is starting journeys or finishing ones that began elsewhere. If the advantage sits entirely with returning users and brand pages, the assistant is acting as a navigation layer.
- Follow leads into the CRM. Compare sales-accepted rate, opportunity rate, win rate, deal size and sales cycle by source. A channel with a high demo rate and a low acceptance rate is not high intent, whatever GA4 says.
- Ask buyers. Add “How did you hear about us?” to demo forms, with an AI assistant option and a free-text field, and ask on sales calls. Self-reported attribution captures AI influence that arrives through Direct, branded search or a colleague’s recommendation, which GA4 cannot see. It has its own biases (people simplify and misremember), so use it alongside analytics, not instead of it.
- Accept the limits on incrementality. You cannot randomise whether assistants mention you. The realistic standard is triangulation: the AI segment, self-reported AI share, branded search demand and pipeline moving in the same direction over several quarters.
What would change the decision:
Three outcomes lead to three different plans.
- If the gap survives landing-page matching, holds for new users, and carries through to opportunities and wins with a few dozen conversions behind it, the intent quality is real. Value AI-referred visits at the higher rate in forecasts, and prioritise being cited accurately on comparison and evaluation prompts.
- If the gap disappears within page types, it is mix. Value AI referrals like organic visits to the same pages.
- If the gap sits with returning users and brand pages, AI is working as a navigation layer for people who already know you. The priority then is that assistants describe your pricing, features and integrations correctly, not a target for AI traffic volume.
What I would put in the business case
I would not put the conversion multiple in front of finance. I would report pipeline and won revenue from AI-attributed leads with the sample size stated, the self-reported AI share from forms and calls, and the landing-page-matched comparison. If those three agree, AI visibility has earned budget on its own evidence. If only the raw multiple looks good, the honest reading is that assistants are sending you people who were already on their way. That is worth protecting, by making sure assistants describe you accurately, but it is not new demand, and it should not be forecast as if it were.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
- Fix the definition. List every source/medium row containing an AI domain across AI Assistant, Referral, Unassigned and campaign-tagged traffic (links from ChatGPT can carry
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.