AI bots crawl 11,000 pages for every visit. Read the trade.

Cloudflare's June data and a 973-site study agree: AI search eats your content, sends almost nothing back, converts worse than nearly everything.

The Editors · 7 min read · Jun 10, 2026

The deal publishers are being offered

Right now, Anthropic crawls 11,122 pages for every visitor it sends to a website. OpenAI's number is 857 to 1. Perplexity is 190 to 1, and it got worse between April and June 2026. Google's exchange is roughly 5 to 1.

If you publish, an AI assistant takes thousands of your pages and gives you, on average, one click in return. That click converts at a fraction of every other channel you already run.

This is the trade. The "make money with AI as a publisher" story rests on it. Before you reorganize a funnel around AI discovery, look at the math.

Short answer: AI traffic is real, it is rising, and as of mid-2026 it is the worst paid traffic source you have. The right move is to keep your current funnel intact, treat AI referrals as a free option, and build the agent yourself if you want AI revenue.

The crawl-to-referral numbers

The cleanest data on this comes from Cloudflare, which sits between most of the world's content and most of the world's bots. Their crawl-to-refer ratio measures pages indexed against visitors actually sent back.

For the May 25 to June 1, 2026 window:

Anthropic: 11,122 crawled pages per referral.
OpenAI: 857 to 1.
Perplexity: 190 to 1, worse than the 95 to 1 it ran in April.
Google: roughly 5 to 1.

Anthropic's ratio improved a lot. In January 2025 it was 286,930 to 1. By July 2025 it dropped to 38,066. By June 2026 it sits at 11,122. The direction is right. The level is still more than 2,000 times worse than the search-to-referral exchange Google built the open web on.

And these are the better assistants. Every AI chatbot combined sent 0.29% of measurable referral traffic in May 2026. Google sent 87.63%.

So before you ask "does AI traffic convert," answer the prior question: there is almost no AI traffic. The category is interesting on a chart and small at the till.

The conversion gap

When AI traffic does land on an e-commerce site, it converts worse than nearly every channel publishers already run.

A study by Maximilian Kaiser (University of Hamburg) and Christian Schulze (Frankfurt School of Finance & Management), published as SSRN working paper 5585812 in October 2025, looked at 973 e-commerce sites with combined annual revenue of $20 billion. They compared 50,000+ ChatGPT transactions against 164 million transactions from other channels across 49 countries and 24 vertical markets.

Findings:

Affiliate links convert 86% more often than ChatGPT referrals.
Organic search converts about 13% more often than ChatGPT.
Paid search and email also beat ChatGPT.
The only channel ChatGPT outperforms is paid social.

The authors wrote that the results "contradict widespread expectations of LLM superiority." Vendor case studies showing ChatGPT convert at 15.9% versus organic at 1.8% are, per the same paper, "cherry-picked outliers" with "methodological upward bias."

ChatGPT conversion is rising month over month. That is the honest counter-point. It is rising from a base where the channel underperforms every other tracked channel except one.

Instant Checkout, quietly retired

OpenAI shipped ChatGPT Instant Checkout in September 2025 on the Agentic Commerce Protocol. Buy products without leaving the chat. By March 2026, it was gone. Per Affiverse summarizing CNBC reporting, roughly 12 merchants ever went live out of millions of eligible Shopify sellers. OpenAI hadn't built the systems to collect US state sales taxes by February. The product was retired without a press release.

Discovery now stays inside ChatGPT. Checkout happens in the merchant's environment, which sounds like the old affiliate model and almost is, with one defect: the publisher who created the content ChatGPT cited often gets no attribution. The buyer searches in ChatGPT, lands on the merchant directly, and the publisher who informed the answer is invisible in GA4. The Affiverse analysis cites Cloudflare data showing 70.6% of AI referrals don't appear in GA4 at all, and reported AI contribution is undercounted by three to four times.

So even the small fraction of AI-influenced sales that publishers earn often doesn't get credited to them. The funnel works in reverse: content goes in, attribution doesn't come out.

What this means for a content-led funnel

If your business is a newsletter, an affiliate site, an SEO-built lead engine, or any content-to-customer pipeline, here's the read.

Treat AI referrals as a free option, not a target. A 0.29% share of referrals does not deserve a quarter of your planning time. Keep optimizing organic, email, and affiliate. Those still run the funnel.

Don't strip product comparisons or affiliate links from pages to win AI overviews. Some teams have started thinning content to look more like "answers," chasing AI citation. Affiliate links convert 86% better than the AI traffic you'd theoretically win. The math says keep your links.

Stop measuring AI success by referral volume alone. With most AI traffic invisible in GA4, the dashboard lies. Treat AI as influence that shows up later in branded search and direct traffic. Look at branded query growth and direct sessions with no UTM, not the AI bucket itself.

If you want AI revenue, run the agent. The operators making real money from AI agents in 2026 are not publishers feeding ChatGPT. They are the ones who built the agent. Salesforce hit $800M ARR on Agentforce by end of fiscal 2026, with Intercom's Fin agent crossing nine figures at $0.99 per resolved ticket, and individual creators on agent marketplaces keeping 70 to 85% of revenue. The money is in the agent layer. Not in being the content the agent reads.

What this doesn't mean

The data does not say AI is a fad or AI traffic will never matter. Three things could flip the picture.

ChatGPT conversion is rising every month. From the current base, a 3x improvement over 18 months is plausible and would leave it above paid social by a wider margin, still behind organic and affiliate.
If AI assistants start sending traffic in proportion to how much they crawl, the volume problem fixes itself. Anthropic's ratio dropped 96% in eighteen months. The direction is right.
Agentic commerce, when it works, removes the conversion gap by closing the purchase inside the assistant. Instant Checkout's first version didn't, and the redesign hasn't shipped at scale.

The honest position: AI search is real and will keep growing. The trade publishers get today is bad. The trade publishers get in 2027 could be different. Don't restructure for 2027 in 2026.

The move, in one line

Keep your current funnel intact. Run the agents yourself if you want AI money. Watch the crawl-to-refer ratios quarterly. When Anthropic crosses 1,000 to 1 and OpenAI crosses 100 to 1, the trade gets interesting. We aren't there.

Sources

This is not financial advice.