ChatGPT sent 0.62% of Google's organic sessions across 100 sites, but analytics misses answers nobody clicks

AI referral traffic is small, and it cannot tell you how much AI shapes who gets chosen. Ask the AI your buyers' questions in Russian and Uzbek without naming your brand, and record whom it names.

Key points

  • In Marie Haynes' sample (100 sites, 12 months of GA4, over 145 million sessions), ChatGPT sent 0.62% of Google organic's volume; all language models together sent 0.73%.
  • In The SEO Works benchmark (95 sites, September 2025 to August 2026), AI assistants sent 39 times fewer sessions than organic search: 72,095 against 2,825,594.
  • Per Pew (900 US adults, March 2025), on Google pages with an AI summary users clicked a regular result in 8% of visits, against 15% on pages without one.
  • None of these studies has figures for Uzbekistan. Whether AI names your company can be measured directly: buyers' questions without the brand, in Russian and Uzbek, dated.

Two Search Engine Journal articles, published on 5 and 6 November 2026, argue over one question: is AI visibility worth the effort if AI sends almost no visitors? We opened the studies they rely on and checked what holds up for a business in Uzbekistan.

What the studies found

  • Marie Haynes analysed 12 months of GA4 data from 100 sites, more than 145 million sessions. ChatGPT sessions were 0.62% of the volume driven by Google organic; all other language models added together brought it to 0.73%. ChatGPT visitors converted at 13.8% against 3.02% for Google organic (55 sites that track conversions). The article links no raw data, only charts.
  • The SEO Works benchmark covers 95 sites from September 2025 to August 2026: organic search delivered 2,825,594 sessions and AI assistants 72,095, 39 times fewer. AI converted at 7.18% against 6.08%, yet on 23 of 63 sites it converted worse than organic.
  • Kevin Indig, in Growth Memo of 19 October 2026, looked at direct traffic on 12 sites and tested the link between AI visibility and later direct traffic or branded search on 4 brands (two separate analyses). His conclusion: "I couldn't establish a reliable link." He adds that an effect could still exist.

In Indig’s sample, 10 of the 12 sites target the US; the Pew study below surveyed US adults. None of the studies has figures for Uzbekistan, and neither do we.

What analytics cannot see

SEJ argues that AI shapes the choice before any click: a person gets an answer in the chat, later searches the brand, and search gets the credit. That is reasoning, not measurement; Indig could not confirm it. There is indirect evidence for fewer clicks, though. Pew Research Center tracked 900 US adults in March 2025: on Google pages with an AI summary, people clicked a traditional search result in 8% of visits, against 15% without one; only 1% of visits included a click on a link inside the summary itself. That the answer itself shaped the choice without a click is our inference; Pew did not measure it. Where the answer settles the question there is no click, and the report stays silent.

What it means for a business in Uzbekistan

A referral report tells you how many people arrived from ChatGPT. It does not tell you whether ChatGPT named your company when a buyer asked. A referral count alone cannot reveal the second, and the second can be measured directly.

That is how our GEO audit works: we put your buyers' questions to AI in Russian, Uzbek and English without naming the brand, and record whom it names instead of you. In the audit of JoyTop, our own product, on 27 August 2026 we asked 26 questions (22 about restaurants in Tashkent without naming JoyTop, and 4 about JoyTop by name) and collected 78 answers from two systems, Claude with web search and Gemini without search. The 22 restaurant questions got 66 answers (22 from Claude, 44 from Gemini), and JoyTop was named in none of them. Claude named 2GIS most often (19 of its 22 answers) and Yandex Maps (18). That run came before 10 September, when we made every run blind, so Claude could search the brand name; that could only have helped JoyTop, and it did not. A visit counter would never show this.

What to measure

  1. Write 10–20 questions your buyer asks, without your company name.
  2. Put them to ChatGPT, Gemini, Claude and Yandex Alice, in Russian and in Uzbek.
  3. Record whom each names, which sources it cites, and the date. Repeat monthly.
  4. Keep AI referral traffic as its own line in your report: it is small, and its conversion may be higher or lower.

One run proves nothing: AI answers change from one ask to the next, so you need repeats and dates.

Questions and answers

How much traffic does ChatGPT send compared with Google?
In Marie Haynes' 100-site sample, ChatGPT sessions came to 0.62% of Google organic's volume over 12 months. In The SEO Works benchmark of 95 sites, AI assistants sent 39 times fewer sessions than organic search. None of these studies reports figures for Uzbekistan.
Why can't referral traffic show AI's influence?
A referral report shows how many people arrived from ChatGPT, not whether ChatGPT named your company in its answer. If the answer settles the question, the reader may never click, and the report does not show that.
How do I check whether AI names my business?
Write down 10–20 questions your buyers ask, without your company's name. Put them to ChatGPT, Gemini, Claude and Yandex Alice in Russian and Uzbek, and record whom they name, which sources they cite and the date. Repeat monthly: AI answers change from run to run.

Sources

Prepared by the Nusrat Tech team with AI assistance, from the sources below.

  1. 4 Datasets On AI Visibility: ChatGPT Sends 0.62% Of Google's Organic Traffic — Search Engine Journal (secondary), 2026-11-06
  2. GEO Isn't A Distraction – AI Influence Doesn't Show Up In Referral Data — Search Engine Journal (secondary), 2026-11-05
  3. GEO Is A Distraction: Here's Where The Real Opportunity Is (Marie Haynes' study of 100 sites) — Search Engine Journal, 2026-10-23
  4. AI Referral Traffic Benchmark (95 GA4 properties, Sept 2025 - Aug 2026) — The SEO Works
  5. Does AI drive more direct traffic? — Growth Memo (Kevin Indig), 2026-10-19
  6. Google users are less likely to click on links when an AI summary appears in the results — Pew Research Center, 2025-07-22
  7. AI visibility (GEO) in Uzbekistan: JoyTop audit — Nusrat Tech, 2026-08-27

More news

  1. Search Console API now splits branded queries and filters several countries at once: what it means for sites in Uzbekistan

    On 16 November 2026 Google added three Performance report features to the Search Analytics API: a branded vs non-branded split, web multimodal search data, and filtering several countries or devices in one request. It matters to anyone who pulls Search Console data into their own reports.

  2. How to tell in analytics that a visit came from Google: what Google wrote and what it means for a site in Uzbekistan

    The referrer shows that Google sent a visit, but not the query, and not always the source: Search, Discover and Images send the same values. Queries are in Search Console.

  3. Google's November 2026 core update: what a site in Uzbekistan should do and when to read the traffic

    Google released its November core update on 10 November 2026; it applies worldwide and in all languages, so it covers Russian- and Uzbek-language sites too. The rollout can take up to two weeks, and Google advises waiting at least a week after it ends before analysing Search Console data.

All news

Want this working for your business?

Tell us what you sell and to whom. We answer in Telegram the same day.

Reach usDiscuss your project on Telegram

  • We reply the same working day, Tashkent time.
  • You write to Nodirkhon; he brings in whichever of us your problem needs.