EZY Research · AI Traffic Index #2 · 25 August 2026
Citation-driven traffic grew 3.3x faster than crawler traffic
AI assistants sent 65% more real visitors last month. AI platforms are not merely harvesting more content - they are increasingly sending real people to the sites they cite.
Published . Server-side logs from a fixed panel of 173 connected business websites, comparing 25 July-23 August 2026 against 25 June-24 July 2026. Every number below comes from a query run on production data. First published on X.
+65%
Citation-driven AI visits
Real people arriving from an AI answer
+20%
AI crawler visits
Training and indexing over the same period
67%
Sites that grew
Share of the panel whose AI visits rose
33%
Sites that fell
Share of sites whose AI visits fell
EZY Research · AI Traffic Index #2 · 173 sites · 30 vs 30 days
Citation-driven visits grew 3.3x faster
Month-on-month growth, 25 July-23 August vs 25 June-24 July 2026
3.3x
EZY Research · server-side logs · 173 connected sites · 25 June-23 August 2026
The finding
Across a constant panel of 173 business websites, total AI visits grew 36.9% month-on-month (69,916 → 95,691).
The aggregate hides the more useful result. When we split AI traffic by what the request actually was, the two halves move at very different speeds:
| Visit type | Prior 30 days | Last 30 days | Growth |
|---|---|---|---|
| Citation-driven (a person's query pulled the page) | 28,325 | 46,778 | +65.1% |
| Crawler (training and indexing) | 43,102 | 51,630 | +19.8% |
Citation-driven traffic grew 3.3x faster than crawler traffic. AI platforms are not merely harvesting more content - they are increasingly sending real people to the sites they cite.
That shifts the mix. Citation-driven visits were 39.6% of AI traffic in the prior window; last month they were 47.5%. On current trend they overtake crawler traffic inside two months.
This is the monthly follow-up to AI Traffic Index #1, which found AI crawlers already out-hitting Googlebot while sending back under 1% of visitors. Issue 2 asks whether the people side of that gap is now growing faster than the crawl. It is. For how that maps onto Answer Engine Optimization (AEO) measurement, see prompt and citation tracking.
Growth is real but unevenly distributed
The median site grew AI visits 33.3%. But the distribution is wide:
| Outcome | Share of panel |
|---|---|
| AI visits grew at all | 66.7% |
| Grew 25% or more | 54.8% |
| Grew 50% or more | 42.9% |
| Doubled or better | 28.6% |
| Declined | 13.3% |
One site in three lost AI traffic month-on-month. Any vendor claiming universal gains is either measuring a different thing or not measuring carefully.
By platform
| Platform | Prior 30 days | Last 30 days | Growth | Sites seen on |
|---|---|---|---|---|
| ChatGPT / GPT | 47,472 | 60,590 | +27.6% | 173 |
| Claude | 17,478 | 24,150 | +38.2% | 173 |
| Perplexity | 4,966 | 10,951 | +120.5% | 148 |
| DeepSeek | 260 | 2,317 | +791.2% | 78 |
| Cohere | 776 | 244 | -68.6% | 65 |
| Mistral | 489 | 194 | -60.3% | 53 |
ChatGPT remains the volume leader by a wide margin - 63% of all AI visits. Perplexity more than doubled. DeepSeek's 791% is real. Cohere and Mistral both contracted.
Why we are not publishing the bigger number
A naive query across every domain in our database returns +174% month-on-month growth. We are not publishing that figure, because it is mostly an artifact. Three corrections took it to 36.9%:
- Onboarding, not growth. 173 domains reported data. Counting new customers as "growth" measures our sales, not AI behaviour. Restricting to domains present in both windows: +67%.
- Partial reporting. Domains that connected mid-window have artificially low prior-period baselines. Requiring at least 25 reporting days in both windows: +50.8%.
- A single burst artifact. One domain logged 9,441 Perplexity visits concentrated in 7 days.
After these corrections, aggregate growth (+36.9%) and median site growth (+33.3%) converge closely, which is what a well-behaved distribution looks like. Before them, the mean sat at +149% while the median was +37% - a four-fold gap that signals outliers, not a trend.
Limitations
- Opt-in cohort. These are EZY.ai customers actively optimising for AI visibility. They are not representative of the web.
- Not a controlled study. We cannot separate rising AI usage generally from the effect of optimisation work. This measures what happened, not what caused it.
- Seasonality unaddressed. A single month-on-month comparison over a northern-hemisphere summer period cannot distinguish trend from seasonal variation.
- Classification is heuristic. "Citation-driven" relies on user-agent and request-pattern signals.
- AI remains a small share of total traffic - 2.67% of all visits, up from 2.48%. Growing fast, from a small base.
Methodology
Server-side request logs collected via the EZY WordPress plugin, with user-agent classification and signature verification. Daily rollups per domain.
Panel: domains with at least 25 reporting days in both windows and at least 30 AI visits in the prior window, excluding one domain flagged as a burst artifact. Comparison windows are 25 June-24 July 2026 and 25 July-23 August 2026. 24 August was excluded as a partial day.
"AI visits" counts requests from GPT, Claude, Perplexity, DeepSeek, Mistral, Cohere and xAI AI agents. It excludes Google, Bing, Meta, Apple, Amazon and ByteDance, which mix AI and conventional search or social crawling.
Aggregate CSVs available on request under CC BY 4.0.
Frequently asked questions
What is citation-driven AI traffic?
Citation-driven traffic is a live request made because a person's query pulled the page - a real visitor arriving from an AI answer, not a training or indexing crawl.
How much faster did citation-driven traffic grow than crawler traffic?
On a fixed panel of 173 business websites, citation-driven AI visits grew 65.1% month-on-month while crawler visits grew 19.8%. Citation-driven traffic grew 3.3x faster.
Why did EZY.ai not publish the 174% growth figure?
A naive query across every domain returns +174% month-on-month, but that figure is mostly an artifact of onboarding, partial reporting and one burst domain. After those corrections, aggregate growth is +36.9%.
Is this panel representative of the whole web?
No. These are EZY.ai customers actively optimising for AI visibility. It is an opt-in cohort, not a controlled study, and a single month-on-month comparison cannot separate trend from seasonality.
What period does AI Traffic Index issue 2 cover?
It compares 25 July-23 August 2026 against 25 June-24 July 2026. 24 August was excluded as a partial day. The panel is 173 connected business websites with at least 25 reporting days in both windows.
More from EZY Research
- We put llms.txt on 83 websites. In 12 weeks, OpenAI's crawler read it 7 times.
- For every training crawl OpenAI makes, real ChatGPT users trigger four live visits
- AI crawlers now hit small-business websites 1.8x as often as Googlebot
- Seven in ten business websites have never told AI crawlers anything at all
- EZY Research hub · FAQ · What is AEO?
Free AI Visibility Check
How visible is your website to AI?
Run a free AI Visibility Check and discover what ChatGPT, Claude and other AI systems can find about your business.