A 2026 study of 858,457 websites on the Duda platform offers a useful view of AI crawler requests and referrals, but its numbers are easy to misinterpret. The study describes activity on one website-building platform. It does not show that AI crawling causes more traffic, that every AI fetch produces a citation, or that Google has introduced a new ranking rule.
This review separates the study’s observable findings from what website owners can reasonably do with them. For a business measuring AI search, the practical question is whether bots can reach important pages, whether real visitors arrive from AI services, and whether either event changes a meaningful reader outcome.
The sample and the reported crawler activity
Duda’s 2026 report concerns small-business websites hosted on its platform. It examines 858,457 sites and, according to the published analysis, 68.9 million AI crawler visits during February 2026. Of the sites studied, 506,910, around 59%, received at least one such visit. Those figures describe observed requests in a defined sample, not a universal rate for all websites.
For context, Search Engine Journal’s April 20 reporting summarised the underlying crawler mix and the referral comparisons. The distinction between Duda’s original research landing page and third-party reporting matters: the complete primary report requires registration, so MOCOBIN has not independently reproduced its raw calculations or inspected the full underlying log dataset.
| Reported measure | 2026 result | What it tells you |
|---|---|---|
| Websites examined | 858,457 | Size of Duda’s platform sample |
| Sites with an AI crawler request | 506,910, about 59% | Observed crawler reach in February 2026 |
| AI crawler visits | 68.9 million | Request volume, not human sessions or citations |
| Reported AI referrals | 93,484 to 161,469, up 72.7% | Change across the comparison reported by Search Engine Journal, not a forecast |
The study also classified requests into user fetch, training and discovery activity. These categories describe bot behaviour inferred from request sources. They must not be merged into one measure of search visibility: an AI assistant fetching a URL for a person and a training crawler visiting it have different purposes.
The relationship between crawling and human visits is not causal proof
The reported comparison is striking: AI-crawled sites averaged 527.7 human sessions, while other sites averaged 164.9. Duda also reported higher average form completions and click-to-call actions in the crawled group. These are observational group differences, not the measured effect of switching on an AI crawler.
Established, active websites can attract more people and more bots for the same underlying reasons. The research does not randomly assign crawl access, control every difference between the groups, or demonstrate that a site would gain 3.2 times more sessions by changing a robots rule. It also does not justify describing every crawler as a customer or every fetch as a brand recommendation.
Some integrations and richer business information were associated with more crawler activity within the sample. Those patterns can identify useful questions for an audit, but they cannot establish that installing a particular integration or adding extra schema guarantees inclusion in AI answers.
What the referral figures do and do not show
Search Engine Journal reported that referrals from LLM services rose from 93,484 to 161,469 in the study’s comparison, a 72.7% increase. Its platform breakdown included ChatGPT at 81,652 to 136,095, Claude at 106 to 2,488, Copilot at 22 to 9,560 and Perplexity at 11,533 to 13,157. Small starting counts make percentage growth for some services look especially dramatic.
These are referrals observed by a particular research cohort. They do not establish the share of all search journeys that begin in AI tools, the percentage of Google queries showing AI Overviews, or a universal loss in organic traffic. The original version of this article combined such external estimates without a dependable shared denominator; those claims should not be treated as findings of the Duda study.
A referral occurs when a visit reaches a website and the referring source is recognised. A mention without a click may not appear in site analytics, and an AI platform may not pass a referrer consistently. Therefore, referrals, AI answer mentions and server-side crawler requests must be reported as distinct measures.
A measurement checklist for website owners
- Start with the website’s own task. Identify the pages that answer customer questions or support enquiries. A page with no relevant reader purpose is not made valuable merely because a bot can fetch it.
- Verify actual crawler access. Review robots rules, public HTTP status and, where available, verified server logs. Different user agents can represent training, search or user-initiated retrieval. A generic user-agent test is not proof of verified Googlebot behaviour.
- Measure human outcomes separately. In analytics, segment recognisable AI referrals and record the attribution limitations. Keep this separate from Google Search Console Web impressions, clicks and any generative-AI reports the property genuinely exposes.
- Check whether the content helps the reader. Make answers, source dates, product facts and limitations easy to understand. Clarify business details where they genuinely help customers, rather than adding markup solely to chase a citation.
- Review change with matched data. Compare similar date ranges after a documented change, ideally with relevant pages left unchanged. An increase in bot requests alone is not evidence of better qualified traffic or conversions.
Google’s guidance for its generative AI Search features continues to emphasise ordinary search accessibility, useful original content and accurate business information. It does not prescribe a special AEO file, schema type or guaranteed citation tactic. For interpretation of on-site performance, the SEO reporting guide covers the distinction between a measurement and a business decision.
What this 2026 study is useful for
The study is valuable as a dated platform-level benchmark: it shows that AI crawlers were contacting many small-business sites and that identifiable AI referrals existed in the sample. It is not proof that a publisher can buy, automate or force AI recognition.
A sensible next action is to document the pages and customer journeys that matter, measure what can actually be observed, and resist claims that a correlation reveals a secret Google or LLM ranking signal. For broader questions about search interfaces and strategy, see MOCOBIN’s Google AI Search strategy analysis, which addresses a different reader task.
Sources and scope
- Duda, The State of Local AEO for SMBs and Their Agencies (2026). Research landing page; full report registration is required.
- Search Engine Journal, April 20, 2026. Published analysis of the sample, request classifications and referral counts.
- Google Search Central, guidance for generative AI features. Platform-specific recommendations, not independent validation of Duda’s correlations.









