What happens when an AI cites a page only agents can see?

The human who clicks lands on the closest real page on your site. Every agent-only page is published with a citation fallback, a page on your human website that answers the same question, and a human request for the agent-only URL resolves to that fallback instead of a 404.

Why an agent-only page gets cited at all

An agent-only page exists in the sitemap and robots.txt that TollBit serves to AI agents. It is not linked from the human site, and Googlebot is never routed to it. An AI search indexer sweeps the bot-only sitemap, fetches the page, and stores it. A retrieval bot fetches it live when a user asks the question in its title. When the assistant composes an answer, it may cite the URL it read.

Citations are a small fraction of reads. Across the network, on the order of 5 to 15% of fetched pages appear as citations, and advanced agents often cite nothing. On tollbit.com, Sep 15 to Oct 14 2026, our logs show about 44K AI crawl attempts and 47 AI referrals, roughly 1,000 reads per human click. The clicks are rare. That is exactly why each one has to land somewhere real.

The fallback, step by step

  1. You publish an agent-only page in the Agent Site CMS and name its fallback, an existing page on your site.
  2. An identified AI agent requests the URL. Your CDN or WAF rule routes the request to the TollBit origin on your domain, and the agent receives the agent-only page.
  3. The assistant cites the URL in an answer. A person clicks it.
  4. The person's browser is not an AI agent, so it is not routed to the agent origin. The URL resolves to the fallback page you named, with your normal human experience, analytics and navigation.

The agent sees a page written for agents. The human sees the page you would have shown them anyway. Same facts on both. Nobody sees an error.

Why 404s from AI answers are the number to watch

AI answers already send people to pages that do not exist. In the Bits and Bots analysis of referral errors, Perplexity had the lowest hallucinated-link rate at 1.16%. Anthropic went from a 55% 404 rate in Q2 2024 to 4.8% in Q2 2025. OpenAI moved the other way, from 0.3% in Q1 2025 to 3.7% in Q2 2025.

An agent-only page without a fallback would add a new category: links that were real for the agent and dead for the human. The citation fallback keeps your agent pages out of that count, and it means every click an AI answer sends you is a visit you can measure in your existing referrer data.

Choosing the fallback

The rule is "closest real page", judged by the question in the agent-only page's title.

Avoid falling back to the homepage. The person clicked because an assistant told them a specific thing. Put them where that thing is confirmed. If no real page carries the fact, that is a gap in the human site worth fixing, and the agent-only page is how you found it.

What stays constant

The same-facts rule applies to every agent-only page: the prices, specifications, dates and claims on the agent page match what a human can find on the fallback. Cleaner format, fewer words, no overlays. Same facts, bots only. The fallback is what makes that rule checkable, because anyone can click the citation and compare.

Frequently asked questions

Does the agent-only URL show up in Google search results? No. Googlebot and Bingbot are never routed to the agent origin, by design, and the agent-only URL is not in the sitemap Google reads. Search engines only ever see your human pages.

Can the fallback point to a different domain? Keep it on your own domain. The citation carries your domain, the person expects your site, and your analytics should record the visit as yours. A fallback on another domain breaks all three.

What happens if I unpublish an agent-only page later? The fallback keeps working. Assistants hold cached copies for a period that varies by platform, from hours to weeks, and may keep citing the URL after you remove the page. The URL continues to resolve to the fallback for humans, so no citation goes dead.

Related