Every deck about AI search now carries the same slide: visitors from ChatGPT and Perplexity convert 23 times better than organic search. The number is real. It is also one SaaS company, measuring its own signups, over thirty days. Before you move budget on it, it is worth knowing what the AI search traffic conversion rate is actually measuring, and why the same metric comes back as 1.4x somewhere else.
The original figure comes from Ahrefs, who published their own numbers rather than a vendor study: AI search sent 0.5% of their traffic over a thirty-day window and that traffic produced 12.1% of their signups. Divide one by the other and you get the multiple everyone repeats. They were transparent that this was their site alone, and said a larger multi-site study was still in progress.
That transparency is the reason the number spread so fast, and also the reason it should not be load-bearing in your forecast. A single company publishing honest first-party data (as Williams-Sonoma did for its Olive shopping assistant) is more useful than most vendor research, but it is still one funnel, one price point, one audience of SEO practitioners who are unusually likely to be running the exact queries where Ahrefs gets cited. Every one of those conditions inflates the result relative to a business selling something less searchable. Conversion depends on the page as much as the channel, which is why landing page CRO for lower Google Ads CPA(x) is usually the faster fix.
What the 23x number actually measures
It measures conversions per visit, not conversion quality, and it measures them against a baseline that is doing a lot of quiet work. Organic search for an established brand like Ahrefs includes an enormous volume of branded and navigational queries — people typing the company name to log in, or searching a feature they already use. Those sessions convert to signup at close to zero, because the person is already a customer.
AI referrals carry almost none of that noise. When an engine names you in an answer, the person reading it is mid-decision on a category question, not looking for your login page. So part of the multiple is a genuine intent difference and part of it is an artifact of comparing a clean segment against a dirty one. Both are real effects. Only one of them survives if you clean up the baseline, which is the same reason attributing pipeline to organic and AI keeps producing arguments between marketing and finance.
Seven studies, seven different multiples
Once you line the published research up side by side, the picture stops looking like a law of nature and starts looking like a measurement problem. These are the figures currently in circulation, with the population each one actually studied.
| SOURCE | POPULATION STUDIED | REPORTED EFFECT |
|---|---|---|
| Ahrefs | 1 SaaS site, 30 days | 23x vs organic (0.5% traffic → 12.1% signups) |
| Opollo | 312 B2B firms, Jan 2025–Jan 2026 | 14.2% AI vs 2.8% Google organic (~5x) |
| Microsoft Clarity | 1,277 publisher domains, Q4 2025 | 17x vs direct |
| Digital Bloom | 2025–2026 | 1.66% vs 0.15% signup rate (~11x) |
| Seer Interactive | B2B client | 15.9% conversion on ChatGPT referrals |
| Visibility Labs | 94 brands, 2025 | 31% higher ecommerce conversion |
| Adobe Analytics | US retail, Q1 2026 | 42% lift |
The B2B SaaS studies cluster high, which matters if you are weighing SEO and GEO for a SaaS product. The retail and ecommerce studies cluster low — Adobe's 42% lift and Visibility Labs' 31% are real improvements, but they are a different order of magnitude from 23x. That is the single most useful thing in the table: the multiple is a function of your business model, not a property of AI search.
Sample sizes run the same direction. The two largest populations here — Opollo's 312 firms and Microsoft Clarity's 1,277 publisher domains — are also the two you would trust most for a general claim, and they land at roughly 5x and 17x against different baselines. The 23x headline sits at the extreme end of a distribution built from one site. When a range this wide gets collapsed into a single number in a board deck, the number that survives is always the most dramatic one, which is how a defensible finding turns into a bad forecast.
Why the AI search traffic conversion rate varies so much
Three variables move it more than anything else, and you control the measurement of all three.
Per-engine conversion rate on B2B referral traffic (Seer Interactive client data)
The per-engine spread in Seer's data is its own argument against treating AI traffic as one channel. ChatGPT referrals converted at 15.9%, Perplexity at 10.5%, Claude at 5.0% and Gemini at 3.0%. A five-fold difference between the top and bottom engine means a blended AI conversion number mostly tells you which engine happens to send you the most traffic. If you are tracking this properly you are splitting by engine, the same way AI engines disagree on sources forces you to split visibility reporting by engine.
The caveat Ahrefs published and nobody quotes
In the same post that produced the 23x figure, Ahrefs noted that users in AI search click links roughly 75% less often than they do in traditional organic search. They followed the arithmetic to its conclusion: if AI search became the dominant discovery surface, their total search traffic could fall to under a quarter of what it is now.
“A 23x conversion rate on a channel that sends 75% fewer clicks is not a growth strategy. It is a warning about the shape of the funnel you are about to inherit.”
That reframes the whole finding. The high conversion rate is not evidence that AI search is about to replace organic revenue. It is evidence that AI search delivers a small number of unusually qualified people, and that the large top-of-funnel volume organic used to provide is the thing at risk. Planning for both at once is the actual job, and it is why we treat generative engine optimization as an addition to search investment rather than a reallocation away from it.
How to measure your own AI search traffic conversion rate
You can produce a defensible version of this number for your own site in an afternoon. The point is not to reproduce 23x — it is to find out what your multiple is, so the budget conversation uses your funnel instead of a SaaS company's.
Expect the exercise to be unflattering at first. Most teams discover their AI referral volume is smaller than they assumed and their organic baseline is dirtier than they assumed, and the two errors partially cancel. What survives is a smaller multiple on a smaller base, which is a worse headline and a much better planning input. It also gives you a number that moves when you do the work, rather than one that stays fixed because it belongs to someone else.
The measurement work matters more than the headline because the headline is going to keep moving. AI referral volume is growing fast enough that a multiple calculated on last quarter's traffic is already stale, and the engines keep changing how often they surface links at all. Teams that built the segmentation early are the ones who can tell the difference between a real shift and a reporting artifact, which is the same discipline behind the AI search not-provided moment.
What to do on Monday
Stop quoting 23x. Build the two segments — non-branded organic and per-engine AI referral — and calculate your own multiple this week. If it comes back at 3x, that is a genuine finding and it should change how you value the channel. If it comes back at 20x, you now have a defensible number instead of a borrowed one. Either way you will have learned more from the segmentation than from the statistic, and you will be measuring the channel on the terms your own reporting and analytics already use.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.