13,828 Pages Crawled, 5 Visitors Sent Back

“AI crawlers fetched 13,828 pages from webdecoy.com and sent back 5 visitors.”

That is WebDecoy’s own result for the 30 days from 27 July 2026 to 26 August 2026: a Crawl-to-Referral Ratio of 2,766:1. Those are our own figures, not an illustration, and you can read the published report rather than take our word for them. It was generated by a feature that is live in the WebDecoy app now: the AI Traffic page can freeze the figures for a property into a shareable AI Traffic Exchange report.

The visitor number needs an immediate qualification. A referral is counted only when a visitor arrives with a Referer from a known AI platform (ChatGPT, Perplexity, Gemini, Claude, and others). Many of these platforms strip or omit that header, so this can stay empty even while AI tools cite your pages. Visitors sent is a floor, not a true count of AI-driven visits. The 2,766:1 figure is therefore an upper bound on the real crawl-to-referral ratio, not a precise measurement of everything AI products sent back.

An AI crawler fetching a page and an AI assistant sending a visitor are different events. They are often discussed together, but usually counted in different places. The AI Traffic Exchange report puts them in one snapshot, comparing pages fetched with visitors referred back.

The report does not decide whether that exchange is fair. It exposes the two sides with the attribution limits attached.

Generate an AI Traffic Exchange report

Open AI Traffic and select a property. The figures are per property, not per account. Choose a live window of 7, 14, or 30 days.

The live view states the exchange for that window: how many pages AI crawlers fetched, how many visitors came back, how many pages were crawled by operators that sent nothing at all, and how much crawling came from operators whose search and AI traffic cannot be separated. It breaks the same activity down by operator, so you can see which companies account for the crawling and which of them returned anything.

Once the property has crawl data, you can generate a report from those figures. Generating one freezes a snapshot at a fixed address, rather than pointing at a view that keeps moving.

A report window can be 7, 14, 30, or 90 days. The 90-day window is available for a report even though the live view stops at 30 days.

Visibility is yours to set. A private report can be read only by people signed in to the owning organization; a public one can be read by anyone holding the link, with no account. You can switch between the two at any time and the address does not change, so a link you have already sent keeps working when you make it public and stops resolving when you make it private. Deleting a report removes it permanently.

A report can also be published without naming the site, which states the figures without identifying whose they are. If a property has no site address recorded, the report cannot name it and publishes anonymously by default.

Worked example: webdecoy.com

The public report covers 27 July 2026 to 26 August 2026 (30 days) and was frozen on 2026-08-26.

All rows sum to 16,453 crawls. The Share column uses that total, including shared search-and-AI crawls. Subtracting the 2,625 shared crawls leaves 13,828 attributable AI crawls. The five visitors were sent by OpenAI (2), Anthropic (2), and Google (1), producing the headline ratio of 2,766:1.

The ten Nothing back operators account for 5,321 crawls, or 38% of attributable AI crawling. This is one domain in one 30-day window, not a benchmark or a trend.

* Google’s crawler is shared between search and AI. The displayed calculation does not make its 1,618 crawls AI-only; its one counted visitor was an AI-product referral, while conventional search referrals are outside the report by definition.

What the report measures

Each report covers a stated window of days and presents five views of the same activity:

• Pages crawled: pages fetched by an AI operator’s crawler.
• Visitors sent: arrivals referred by an AI assistant or AI search product.
• Crawl-to-Referral Ratio: pages crawled per visitor returned, written as N:1.
• Crawling that returned no traffic: the percentage of attributable AI crawling associated with operators that sent no visitors in the window.
• Operator breakdown: operator, pages crawled, visitors sent, ratio, and share of crawling.

The ratio is arithmetic, not a value judgment. A higher number means more pages were fetched for each referred visitor observed during that window. It does not measure what was done with a page after fetching it, whether a crawl produced an answer, or the value of any resulting visit.

Likewise, “nothing back” has a narrow meaning. In the webdecoy.com example, it applies to Meta, Amazon, ByteDance, Perplexity AI, You.com, Common Crawl, Exa AI, DeepSeek, Cohere, and Mistral AI. WebDecoy observed crawling from those attributable AI crawlers and no qualifying referral from those operators during the report window. It does not mean an operator provided no benefit of any kind, and it does not establish why no referral appeared.

What counts as a referral

“Visitors sent” is deliberately narrower than all traffic connected to the same company.

The report counts an arrival when its referrer is an AI assistant or AI search product, including chatgpt.com, gemini.google.com, claude.ai, and perplexity.ai.

A visit from a search engine results page is explicitly not counted. That boundary is necessary because the report is about the exchange between AI crawling and AI-product referrals, not total traffic from every product an operator owns.

It also means the visitor count depends on observable referral information. The report describes recorded referrals under that definition; it is not a complete reconstruction of every path by which a person may have learned about a page.

Specifically, detection depends on the HTTP Referer header, and many AI platforms strip or omit it. A referral is counted only when that header arrives and identifies a known AI platform. A window can therefore read zero while those tools are actively citing your pages, and a zero does not establish that AI products produced no citations or drove no visits.

For the same reason, Visitors sent is a floor. Any ratio calculated from that observed denominator is an upper bound on the real crawl-to-referral ratio. The report can state what WebDecoy observed at the site boundary; it cannot recover referrals whose identifying header never arrived.

The mixed-crawler problem

Some operators run one crawler for both a search engine and an AI product. In the webdecoy.com report, Google, Microsoft, and Apple are marked search + AI. A request from those crawlers does not say which downstream system caused the fetch. Assigning every such request to AI would manufacture precision the data does not contain.

Their crawling remains visible in the operator table. Microsoft and Apple read Not separable. Google’s 1,618:1* is explicitly qualified because its crawler total is shared; it is not an ordinary AI-only ratio.

Those crawls are also excluded from the headline percentage of AI crawling that returned no traffic. This is more than a display choice. Including them would let inseparable search crawling support a claim specifically about AI crawling.

The limitation carries through to interpretation: an operator with a shared crawler can show crawled pages and no AI-product visitors in this report while its conventional search engine is still sending real traffic. The search visits are outside the report’s referral definition, and the shared crawls cannot be assigned cleanly to search or AI.

So the operator table and the headline percentage answer related but not identical questions:

• The table shows the observed operator activity, including mixed infrastructure.
• The headline percentage uses only crawling that can be treated as AI crawling without that ambiguity.

This exclusion makes the headline less comprehensive, but more defensible.

Reading the ratio without overreading it

If a report shows an N:1 ratio, it means the operator fetched N pages for every qualifying visitor referred during that snapshot, after the report’s attribution rules were applied.

The figure is useful for comparison within the same report because the window and definitions stay fixed. It can show which attributable operators accounted for the crawling and which sent recorded visitors back.

It does not explain causation. A referred visit need not have resulted from a specific page fetch in the same window. Crawlers refresh indexes, products may rely on previously collected material, and a report does not attempt page-level matching between a crawl and a later visit.

The ratio also should not be read as a conversion rate. Its denominator is referred visitors, not impressions, citations, answers, or clicks on a known set of appearances. It measures an observable exchange at the site boundary: fetches in, referrals out.

Because AI platforms frequently omit the Referer header, the denominator can undercount AI-driven visits. Read N:1 as no more than N observed crawls per identified referral: an upper bound on the real ratio, not a complete accounting of visits produced by AI tools.

When no qualifying visitor was observed for an attributable operator, the table says Nothing back instead of inventing a finite ratio. Division by zero is not insight.

A snapshot, not a dashboard

An AI Traffic Exchange report is frozen when it is created. It states the measurement window and the snapshot date; reopening it later does not extend the window or replace its figures with current traffic.

That makes a published report reproducible. Two people opening the same link are discussing the same observation period rather than a dashboard that changed between views.

It also limits what the report can support. The snapshot describes that domain during that window. It is not a forecast, an industry benchmark, or evidence that the same ratio persists. A different window can produce a different result.

To assess change, generate another report for the later period and compare the stated windows rather than treating the earlier snapshot as live.

What the report can answer

The report is designed for a bounded set of questions:

• How much attributable AI crawling did this domain receive during the window?
• How many visitors arrived from the named AI products under the report’s referral definition?
• What was the crawl-to-referral ratio for each separable operator?
• Which operators accounted for the largest share of crawling?
• How much attributable AI crawling came from operators that sent no observed visitors?
• Which rows cannot be separated because search and AI share a crawler?

It cannot tell you whether a crawler used a page for training, retrieval, indexing, or another downstream purpose. It cannot assign a shared crawler’s requests to search or AI. It does not count conventional search referrals as AI referrals, even when the search engine and AI product have the same operator.

Those are not footnotes to remove later. They define what the numbers mean.

Why make the exchange visible

Crawler logs answer how often automated clients fetch a site. Referral analytics answer how visitors arrive. Looking at only one side encourages two opposite mistakes: treating every crawl as evidence of distribution, or treating every AI referral as if it arrived without an acquisition cost borne by the publisher’s infrastructure and content.

The AI Traffic Exchange report joins those observations without claiming they are a transaction that can be reconciled page by page. Its useful output is narrower: a frozen, operator-level account of what was fetched and what qualifying referral traffic returned during the same window.

Generate an AI Traffic Exchange report for your own property from AI Traffic. You can also start with WebDecoy. For request-level investigation, the WebDecoy tools hub includes crawler verification and supporting references. Verification answers whether a crawler identity is credible; the exchange report answers what the observed crawling and referrals looked like over time. They are separate questions, and the distinction matters.