We clicked the receipts

When an AI answers a buyer’s comparison question, it shows its work: a list of citations under the answer, about a dozen per response. Those citations are the answer’s receipts. They’re why the recommendation feels trustworthy.

So we clicked all of them.

Across our benchmarks we’ve collected just over 115,000 citations from nearly 10,000 AI answers to buyer evaluation questions, spanning ChatGPT, Claude, Perplexity, and Gemini. We fetch every cited page, because our scoring works claim-by-claim and a claim’s supporting source is part of the measurement. Fetching them means we also learn something nobody asks about: whether the page is still there.

Often, it isn’t. One citation in eighteen points at a page that is permanently gone. Not slow or behind a cookie banner. Gone. HTTP 404, HTTP 410, or a domain that no longer resolves at all. And because citations cluster a dozen to an answer, the per-citation rate understates what a buyer actually experiences: nearly one answer in three leans on at least one citation that resolves to nothing.

This isn’t residue from old benchmarks, either. Restrict the corpus to citations captured this month and the rate barely moves. Fresh answers, generated in July 2026, citing pages that are already dead.

Not all engines rot equally

The rate varies by engine, and the spread is wide:

Perplexity cites dead pages at three to six times the rate of the other engines. Nearly half of its answers contain at least one. The gap has a mechanical explanation. Perplexity leans hard on software directories and aggregator listings, and directory URLs churn constantly: a product page moves, a category gets restructured, and the listing the index remembers returns 404. The engine that retrieves most aggressively also inherits the most link rot. The most-cited dead page in our corpus is a software directory’s head-to-head comparison listing. It returns 404 today but still appears as evidence in 125 different answers.

We’re strict about what counts as dead. A page that times out once, throws a transient error, or sits behind a bot wall gets classified separately and retried; none of that is in our count. “Permanently gone” means we probed it, retried it, and got a definitive no: this resource no longer exists.

What a dead citation looks like

The failure modes are more varied than “somebody deleted a blog post.”

The page moved and nobody redirected. More than a thousand distinct cited pages return plain HTTP 404. Docs restructures, blog re-platforms, marketing-site redesigns. The content usually still exists somewhere. The URL the AI knows doesn’t.

The domain itself is gone. The starkest case in our data is an open-source project’s retired domain, sktime.net. DNS doesn’t resolve. There is no server left to return an error. Three engines independently cited pages on it nearly 600 times and presented a nonexistent website as evidence.

The page was deliberately withdrawn. HTTP 410 means “gone, on purpose, don’t come back.” We found over a hundred citations to explicitly retracted pages in ChatGPT answers alone. Publishers pulled the content, but it still does work in live answers.

And it’s not just third parties. The pattern shows up on vendors’ own sites. We’ve watched documentation pages moved during a site restructure keep collecting citations at dead URLs because nobody added redirects. The AI is recommending those vendors from pages their own team deleted.

How do you cite a page that doesn’t exist?

Because the engine isn’t reading the web when it answers. It’s reading its memory of the web.

Every engine answers from some blend of model weights and retrieval indexes, both of which are snapshots. A page gets crawled, its content gets absorbed, and then the page dies. The snapshot remains. When a buyer’s question arrives, the engine synthesizes from what it ingested and attaches the URL it ingested it from. The citation is an artifact of when the engine last looked, not of what exists now.

This is the same gap we keep finding everywhere in this space: what AI says about you is built from a version of the record that can be months or years stale. Dead citations are the most literal, checkable version of it. You need an HTTP client, not a judge model, to verify this failure.

We’re not the only ones seeing this

We wanted to know whether our corpus was an outlier, so we went looking for independent evidence. It’s not an outlier. It’s a well-documented pattern arriving from two directions at once.

The first direction: the web itself rots. Pew Research Center’s When Online Content Disappears found that 38% of webpages that existed in 2013 were no longer accessible a decade later, and roughly a quarter of news pages contain at least one broken link. Any system trained on snapshots of a decaying web inherits that decay.

The second direction: AI engines handle that decay badly. The Tow Center for Digital Journalism tested eight AI search engines on news citations and found them collectively wrong on more than 60% of queries. Gemini and Grok 3 cited more fabricated or broken URLs than working ones, and Grok 3 pointed at error pages in 154 of 200 tests. At a much larger scale, Ahrefs studied 16 million URLs cited or recommended by AI assistants and found they send users to 404 pages nearly three times as often as Google Search does.

Our dead-citation rate runs higher than the rates Ahrefs measured across the open web because the corpora differ. Their corpus spans every kind of query. Ours contains only B2B buyer comparison questions, which engines answer by leaning on software directories, comparison listings, and vendor docs. Those page types churn quickly. The general web has a link-rot problem. The competitive-evaluation web has a worse one.

Dead ground is unclaimed ground

Here’s why we treat this as a competitive positioning finding and not a web-hygiene curiosity.

A citation is a claim on the answer. Whoever’s page the engine leans on controls what the buyer hears. A dead citation is a claim on the answer that nobody can contest. The buyer can’t click through and evaluate the source. You can’t read what it said about you. The page will keep shaping recommendations until the indexes catch up or something better replaces it.

That’s the opportunity. When the page under a competitive answer no longer exists, the ground that answer stands on is vacant. A live, well-structured page that answers the same buyer question is competing against a ghost. The fastest-moving content plays we see are precisely these: find where an engine is answering your category’s questions from dead or dying sources, and be the replacement the retrieval layer finds next.

The defensive version matters too, and it’s cheaper: if cited pages on your own domain are 404ing, add the redirects. You are letting citations you already earned rot.

What we built to fix it

None of this is visible if you treat citations as a list of strings. It only shows up if you resolve every cited URL and keep resolving them. Pages that are alive today die later, so the record has to move with the web. That’s the design principle behind KnitKnot’s Content & Source Intelligence, and it turns dead citations from a curiosity into a workflow:

Every cited page is a first-class record. Each URL an engine cites in your benchmarks gets fetched, classified by page type and owner, and tracked as its own row, with every answer that leaned on it attached. When a cited page is unreachable, you get the receipt: the specific failure and when we last confirmed it. Because dead pages sometimes come back, we re-probe them weekly so the record stays current in either direction.

Dead sources become ranked work, not trivia. When a dying page is shaping answers you’re losing, that’s a playbook: create the live page that claims the vacant ground, or fix the redirects rotting your own citations. Each play is tied to the dead source, the answers it drove, and the buyer question at stake, so your team ships the page that can move the answer.

The loop closes with proof. After the work ships, the next benchmark re-runs the same buyer questions and shows, answer by answer and citation by citation, whether the engines moved from the dead source to you.

The engines show buyers a dozen receipts per answer, and as of this month, one receipt in eighteen points at nothing. Somebody should be checking. A benchmark shows which answers and citations in your category stand on dead ground. The first one is free.