In October 2026, three studies were widely discussed in search marketing, and all three reached the same finding: AI assistants send websites very little traffic. If your decision makers ask "what is AI search doing for us?" and you answer with a referral traffic chart, the chart will look like a failure. Referral traffic is the wrong single metric for this question, and you can set up a better measurement this month without buying new tools.
The traffic numbers
Marie Haynes analyzed 12 months of GA4 data across 100 websites with over 145 million sessions. In her post "GEO is a distraction" (21 Oct 2026), ChatGPT sessions came to 0.62% of the volume from Google organic. She also found that ChatGPT session growth ended around August 2025 and has declined since.
SEO Works reached a similar result in its AI referral traffic benchmark, which covers 95 websites from September 2025 to August 2026. Organic search delivered 39 times more traffic than AI assistants. For every 100 conversions from organic, AI referrals delivered 3.2. By their arithmetic, a 5% lift in organic would bring more additional conversions than doubling AI referral traffic.
The quality of that traffic is mixed. Of Haynes' 100 sites, 55 tracked conversions, and on those ChatGPT visitors converted at more than 4.5x the rate of Google organic visitors. SEO Works found that AI referral traffic converted at 1.18x the organic rate when pooled, but on 23 of 63 websites it converted worse. For B2B product businesses it converted at 0.62x organic.
Why clicks undercount
Mentions outnumber citations. A BrightEdge analysis of ChatGPT prompts from July 2025 found that ChatGPT mentioned brands 3.2 times more often than it cited them (2.37 mentions versus 0.73 citations per query). Most of that exposure never produces a link, so it can never produce a click.
Some of the clicks that do happen are hidden. SEO Works notes that visits from Google AI Overviews arrive as ordinary organic sessions in GA4 and cannot be separated, so the AI share it measured is a floor. Mike King of iPullRank makes the same point: clicks from AI Overviews and AI Mode usually show up as Google organic.
AI summaries also reduce clicking in general. Pew Research Center published browsing data from 900 US adults on 22 July 2025, collected in March 2025. Users clicked a traditional result in 8% of visits when an AI summary appeared and in 15% of visits without one. A link inside the summary was clicked in 1% of visits.
King draws his conclusion in "AI search is a branding channel" (22 Oct 2026): AI search is a brand channel that is being judged as a performance channel. In his words: "Nobody has ever attributed a billboard."
Where King and the data agree
King and the authors of the studies end up closer than the debate suggests. King proposes three layers of metrics. Input metrics ask whether your content can be retrieved at all. Channel metrics cover share of voice across the prompts buyers use, mention frequency, citation position and sentiment. Performance metrics cover branded search, direct traffic trends, assisted pipeline and self-reported attribution.
Kevin Indig, writing in Growth Memo on 19 Oct 2026, came to a similar structure from the data side. Across 12 GA4 properties, he found that much of the growth in Direct traffic was bots. One site logged 409,000 Direct sessions, of which 198 were engaged. In a separate analysis of four brands with 632 tests, he could not establish a reliable link between AI visibility and later branded search or Direct traffic. One site showed a 0.72 correlation between AI share of voice and Direct traffic at a three-week delay, but it did not hold up against shifted comparisons.
His recommendation combines three signals: where your brand appears across a fixed set of prompts, platforms and markets; what new leads say when you ask where they first heard of you, in their own words and including which AI tool; and what happens commercially, meaning qualified leads, purchases and revenue. He puts all three on the same monthly report.
What goes on the monthly report
- A fixed prompt set, checked every month. Write down the questions your buyers actually ask, run them through the AI tools those buyers use, and record whether your brand is mentioned, where it appears and in what tone. Keep the set fixed so the trend means something. This is King's channel layer and the first of Indig's signals, and a spreadsheet is enough to run it.
- A free-text "where did you first hear about us?" field. Put it on your contact form and into your sales call script. Use free text, not a dropdown, so people can write "ChatGPT recommended you" in their own words. Then connect those answers to qualified leads, purchases and revenue. Both King and Indig list self-reported attribution, and it is the one signal here that comes from the buyer directly.
- Conversion tracking that works, tied to pipeline. This sounds basic, but Haynes found that 45 of the 100 sites in her data tracked no key events at all. Her advice is to define what a conversion or key event means for your business and to track it. The other two measures need this layer to connect to revenue.
Before you read anything into Direct traffic, filter out bots. Indig's site with 409,000 Direct sessions and 198 engaged ones shows how far raw Direct numbers can drift from real visitors.
The report fits on one page: the prompt visibility trend, counts of self-reported AI discovery, and key events with pipeline. Reviewed monthly, it gives decision makers a more honest answer than a referral chart that shows 0.62%.
Start this week
The cheapest item is the free-text field. Add "Where did you first hear about us?" to your contact form, and ask your sales team to put the same question in every first call. After a month you will have the first answers that no analytics tool records for you.