AI search referrals: what the studies actually measure

AI search referral studies don't measure one thing. Semrush, Ahrefs, Seer Interactive, Similarweb and Previsible counted clicks on Google results, sessions from chatbots, conversions on one site, or how often AI answers appear. Those answer different questions, so the figures can't be merged into one range. Measure your own referrals instead.

What does each study measure?

Nine figures from five publishers, each checked against the original page on 22 October 2026. All are vendor-reported: each publisher sells software or services in search marketing or web analytics (Semrush, Ahrefs and Similarweb sell software; Seer and Previsible sell agency or consulting services).

Two rows aren't about visitors sent to a site. Semrush's AI Overviews row measures how often Google shows an AI answer. Seer's compares click-through on queries where a brand is cited with queries where it isn't, using Search Console and Google Ads data for clicks, not visits from an AI assistant.

Why do the figures disagree?

They disagree because each study sits on a different population, counts a different thing and divides by a different total.

• Populations. Ahrefs measured its own site, and its author doubts the higher conversion rate would hold if AI search became the default. Semrush didn't disclose its sample. Similarweb estimated across ecommerce sites, and Previsible read analytics from 166 properties. Even Similarweb's two 2025 releases describe different groups: 11.4% for ChatGPT-referred ecommerce visits, and about 7% for "Gen AI referrals to transactional sites".
• Metrics. Three publishers compare conversion, and none defines it precisely. Semrush reported the average AI visitor as 4.4 times as valuable as an organic one "based on conversion rate". Ahrefs reported a 23x higher conversion rate on its own site. Similarweb estimated 11.4% against 5.3% for organic search. Each is a different conversion, on a different population, over a different or undisclosed period. Averaging them, or quoting a range that spans them, would describe no real site.
• Denominators. Ahrefs' 96% is a proxy: a platform's share of tracked traffic divided by its daily query volume. Semrush's 92 to 94% counts AI Mode sessions with no visit to an outside domain. Seer's 35% compares click-through on queries where an AI Overview appeared and the brand was cited with those where it wasn't. Queries, sessions and impressions are three bases that can't be set against each other.
• Dates. Figures move with sample and period, even inside one publisher. Previsible's 2025 edition (opens in a new tab)^[12] put ChatGPT at 84.2% of 1,963,544 LLM sessions (November 2024 to November 2025). Its 2026 edition reports 92.4% of trackable standalone referral traffic over 19 months, and warns: "Absolute figures should not be compared directly across Previsible reports without accounting for differences in sample composition and size." Semrush's AI Overview trigger rate, measured across a set of 10M+ keywords, was 6.49% of queries in January 2025, 24.61% in July and 15.69% in November. Seer's April 2026 update (opens in a new tab)^[10] used a larger cohort (53 brands, 5.47M tracked queries) and reports that organic click-through on AI Overview queries rose from 1.3% in December 2025 to 2.4% in February 2026, after falling through 2025. Its figures can't be set against the September 2025 ones.

Seer is candid about its own limit: "We cannot definitively prove that citation causes higher CTRs." Stronger brands may be both cited more and clicked more. The same Seer page also gives 1.2% and 11.05% for cited queries in a later section, so only its Q3 2025 averages are used here.

What do the studies say about scale?

Three publishers describe AI referrals as small, without agreeing on a number. Ahrefs found 0.5% of its own traffic in the 30 days before its June 2025 post. Previsible writes that "LLM traffic represents a fraction of total sessions". Similarweb's 2025 ecommerce release calls it "a small but fast growing and important share of traffic from AI chatbots like ChatGPT".

When counts are small, rates are unstable: a handful of extra enquiries moves a conversion rate a long way. That is one reason a single-site multiple such as Ahrefs' 23x shouldn't be carried over to another business.

What can you measure on your own site?

Your analytics already records where each visit came from. Four steps turn that into a usable view:

1. Group AI referrers into one channel. Previsible's 2026 study tracked ChatGPT (chatgpt.com), Perplexity (perplexity.ai), Gemini (gemini.google.com), Copilot (copilot.microsoft.com), Claude (claude.ai), Grok (grok.com), You.com and Meta AI (meta.ai), plus subdomain and app variants captured in GA4. Keep your own list and review it as new assistants appear.
2. Report landing pages by source. Similarweb (opens in a new tab)^[9] said in July 2026 that, in ChatGPT referral data to late May, "roughly six in ten referred visits land on a homepage", a pattern it ties to a May 2026 change in how ChatGPT shows brand links. That is one platform in one period, so check which pages receive AI visits on your own site rather than assuming the cited article gets the click.
3. Count conversions by source, with the conversion fixed in advance. Choose one: an enquiry, a booking or a purchase. Show counts beside rates.
4. Compare like with like. Use the same period and device mix against Google organic, and keep the same definition each month.

How quickly a new enquiry gets a reply is a separate lever; speed to lead covers it.

Why does referrer data under-count?

Some AI-driven visits carry no referrer, and some happen after the answer rather than from it. Semrush's traffic model (opens in a new tab)^[1] allows for people who click AI links without referrer data (shown as direct or unknown), people who see an AI answer and then navigate manually (also direct), and buyers who convert later through channels such as branded search.

Google's AI answers are a second gap. Previsible excludes AI Overviews because they "don't generate trackable referral sessions the same way standalone platforms do". It adds that AI discovery inside Google's results almost certainly represents more AI-driven traffic than all standalone assistants combined, though the page describes no measurement of it. Google's documentation (opens in a new tab)^[13], last updated 10 December 2025, says clicks from these features are included in overall Search Console traffic, reported under the "Web" search type. Previsible also cautions that its Gemini numbers "may undercount Gemini's real influence on how users find and evaluate brands."

Treat your AI-referrer channel as a floor, not a total. Analytics shows visits, not citations: a person can read your name in an AI answer and never click. Whether assistants cite you accurately depends on upstream work, covered in generative engine optimisation for expert businesses, answer-first writing for people and AI and entity SEO for experts.

How should you quote these numbers?

Put the publisher, year, population, metric and denominator in the same sentence as the figure, and say it is vendor-reported. One study per claim, never an average or a range across studies.

Compare two versions of the same claim. The sourced version: Ahrefs reported a 23x higher conversion rate for AI search visitors than for organic visitors on its own site in the 30 days before June 2025 (vendor-reported). The shorthand: AI traffic converts 23x better. The first can be checked and the second can't. For a decision about your business, use your own numbers, and publish them with their period and denominator.