Every AI Search answer is built on sources, and the engine cites them. A practical question follows: are those sources still there? When ChatGPT, Gemini, or Google AI Mode points to a page as evidence, does that page still exist, or is the engine citing something that is already gone?
OtterlyAI studied 20M+ URLs cited by seven AI Search Engines over one full month, and checked the live status of each one. 19.3% do not return a live page. The source was deleted, moved, or can no longer be reached, yet the engine still cites it. We call this a dead citation: an AI answer built on a source that was live when it was first cited, then removed, archived, or left to go out of date, that still lives in the engine’s memory and keeps getting cited.
This study measures how common dead citations are, which engines produce them most, and which kinds of sources cause them. It is the companion to our live takedown experiment, where we watched dead citations form in real time by removing pages on known dates.
Key Findings
19.3% of cited URLs are not live. Across 20 million URLs cited in 1 month by seven AI Search Engines, nearly 1 in 5 returns a missing, moved, or unreachable page.
ChatGPT has the highest dead citation rate at 25.1%, but it also casts the widest net of the chat engines Google AI Overviews is cleanest at 12.6%.
1 in 7 cited Reddit URLs is no longer published. 15.1% of the analyzed Reddit URLs were deleted, removed by admins or moderators, purged, or restricted.
7.7% of cited Wikipedia URLs do not lead to a live article. Most point to pages that were deleted or never existed, Wikipedia pages that only exist as drafts.
77% of cited PDFs carry no date at all.The evergreen whitepapers, spec sheets, and reports that get uploaded once and never touched again, so nothing signals to an AI Search Engine whether the file is current.
Among dated PDFs 40% are more than five years old. Only 8.5% are less than a year old, so the engines are citing established/outdated documents rather than your latest ones.
What Are Dead Citations in AI Search And Why Do They Happen?
A dead citation is an AI Search answer built on a source that no longer exists or has content that is outdated. The page was live when the engine first cited it, then it was deleted, moved, banned, or left to go out of date. The engine keeps citing it anyway.
A naming note: the term ghost citations is also used in the GEO community for a different gap, an AI answer that links your content without naming your brand, as Kevin Indig describes. What we measure here is the source itself being gone, so we call these dead citations to keep the two problems apart.
They happen because AI Search Engines do not read the live web every time they answer. They answer from stored snapshots and indexes that refresh on their own schedule. When a page disappears, the citation only dies when the engine recrawls that exact URL, which can take weeks or months. Niche pages get recrawled least, so they linger longest. And a stale page that still loads sends no signal at all.
This month, 19.3% of 20 million cited URLs pointed to sources that are gone.
OtterlyAI tracks the exact URLs each AI Search Engine cites for your brand, so you can separate live citations from dead citations and watch removed pages fade. 👉 Track your AI Search citations with OtterlyAI
Scope of Study
This study is an original OtterlyAI analysis of the live status of URLs cited by AI Search Engines. For each cited URL, we recorded:
The AI Search Engine that cited it (ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, Microsoft Copilot, Claude).
The cited URL and its domain.
The HTTP status the URL returns when fetched: live, redirect, missing, server error, or unreachable.
For PDFs, whether the URL carries a year, and which one.
The corpus is 20 million cited URLs. The status check runs continuously, and the figures here reflect the URLs checked so far, roughly a third of the corpus. Two interpretive notes:
This is a descriptive study of live-status patterns, not a controlled experiment. It shows how many cited URLs are gone, not why any single one was chosen.
Status codes can change over time, and a single check is a snapshot. We report the direction, which is consistent across all seven engines.
Nearly 1 in 5 cited URLs is not live
Of 20,092,769 cited URLs checked, 80.7% return a live page and 19.3% do not, split across missing, unreachable, redirected, and server-error responses.
What the URL returns
Status code
Success (live page)
200
Redirect, page moved
3xx
Missing page
404 or 410
Server error, dead DNS
5xx
Page unreachable
000
The largest slice, 8.7%, is missing pages that return a 404 or 410. Another 6.4% are unreachable, 3.5% are redirects that move the reader off the page that was cited, and server errors make up the rest. Each one is a citation pointing at content the reader will not find where the engine said it was.
What this means for your GEO strategy: Treat AI citation counts as a gross number that needs a net. Some share of the pages an engine credits you with, or credits a competitor with, no longer exists. Separate live citations from dead ones before you report visibility.
Dead citations by AI Search Engine
Dead citations by AI Search Engine, at scale
The rate varies widely by engine, and so does the volume each engine cites.
AI Search Engine
URLs cited in one month
Cited URLs not live
ChatGPT
5,779,229
25.1%
Google AI Mode
2,195,767
23.9%
Microsoft Copilot
2,418,957
20.0%
Gemini
739,850
18.7%
Perplexity
2,631,741
18.0%
Claude
67,191
16.7%
Google AI Overviews
6,260,034
12.6%
All engines
20,092,769
19.3%
ChatGPT has the highest dead citation rate at 25.1%, and context matters here: it also casts a far wider net than the other chat engines. ChatGPT cited 5.8 million distinct URLs in one month, nearly as many as Google AI Overviews and 86 times more than Claude. Casting that wide means citing more fringe, aging, and fragile pages, and more of them die. In absolute terms it adds up to about 1.4 million dead cited URLs from ChatGPT alone, the most of any engine.
Google AI Mode is close behind at 23.9%, but for a different reason: 17.7% of everything it cites is unreachable, by far the highest of any engine, suggesting it holds on to URLs whose hosts have gone dark. Google AI Overviews is the cleanest at 12.6%, roughly half ChatGPT’s rate, from the same company that runs AI Mode.
So which AI Search Engine has the most dead citations?
By rate and by raw volume, the answer is ChatGPT: 25.1% of its cited URLs are not live, about 1.4 million dead sources in a single month, more than any other engine. Its wide net explains part of the rate, but the exposure is real either way.
The picture shifts if you measure differently. In our companion takedown experiment, Claude was the most persistent, keeping about 78% of its citations to removed pages alive after deletion, and Gemini produced the most dead citations by volume in that test. Rate, persistence, and volume can point to different engines, so name the failure mode before you name the worst engine.
Dead Citations caused by outdated pages
Wikipedia: cited pages that are no longer articles
Wikipedia is the reference source AI Search leans on, so we classified all 33,490 cited Wikipedia URLs at the content level: does the URL lead to a live published article, or not.
AI Search Engine
Cited Wikipedia URLs
Not a live article
Google AI Overviews
8,259
10.1%
Google AI Mode
2,272
10.0%
Perplexity
6,574
9.4%
Claude
535
8.2%
ChatGPT
12,142
5.9%
Microsoft Copilot
1,951
4.8%
Gemini
1,757
2.1%
All engines
33,490
7.7%
7.7% of cited Wikipedia URLs do not lead to a live article. The largest group, 5.7% of all cited Wikipedia URLs, points to articles that were deleted or never existed at that address. The rest are non-article pages, mislabeled URLs, and the most interesting group: drafts.
Over 120 cited URLs point to topics that only exist in Wikipedia’s Draft space, pages that moderators pulled back from the live encyclopedia or that never passed review. ChatGPT alone cited 97 of them. An article does not need to survive Wikipedia’s review process to leave a trace in AI answers.
PDFs: still cited, years out of date
Out of 20M+ URLs cited by AI Search around 1.69% URLs (338,691) are PDF files. And most of them still load fine (89.34%). But they are not immune.
Dead PDFs
Of the cited PDFs, 10.66% of the total, are already dead: These PDFs no longer exist and the rest return client or server errors. ChatGPT (10.6%) and Perplexity (10.3%) cite dead PDFs most often, while Gemini is the cleanest at 1.1%. The bigger problem is age.
Outdated PDFs
77% of cited PDFs carry no date in the file name at all: the evergreen whitepapers, spec sheets, and reports that get uploaded once and never touched again.
23% PDFs cited by AI Search carry a date in the file name. Of these dated PDFs, old files dominate.
40% of dated PDF citations point to files older than five years, and 14% to files older than ten. Only 8.5% point to a PDF from the last year. Perplexity leans oldest, with nearly half of its dated PDF citations pointing to files older than five years. The engines are not citing your latest documentation, they are citing whatever established document has been sitting at a stable URL the longest.
What this means for your GEO strategy: A PDF you published years ago may still be doing your AI visibility work, quoting old pricing, discontinued products, or outdated specs. Audit which of your PDFs get cited, update or remove the stale ones, and date your documents so engines can tell fresh from fossil.
Reddit: dead citations after the crackdown
Reddit has announced a broad crackdown on bots, vote manipulation, and brand-mention spam from people trying to get cited by AI Search Engines. Accounts, threads, and entire communities have been removed, archived, or banned. That makes Reddit the perfect test of dead citations at platform scale: how much of what AI Search cites from Reddit is already gone?
We classified all 418,362 cited Reddit URLs at the content level: still published (live, archived, or locked but viewable) versus no longer published (deleted, removed by admins or moderators, purged, or restricted).
AI Search Engine
Cited Reddit URLs
No longer published
ChatGPT
245,370
14.4%
Google AI Overviews
100,613
14.5%
Perplexity
30,557
21.3%
Google AI Mode
25,446
14.3%
Gemini
14,588
17.0%
Microsoft Copilot
1,788
24.4%
Claude
0
no Reddit citations
All engines
418,362
15.1%
Share of each engine’s cited Reddit URLs that are no longer published. Claude cited no Reddit URLs at all.
15.1% of cited Reddit URLs, more than 63,000, are no longer published. 1 in 7 Reddit threads that AI Search presents as a source has been deleted, removed, or restricted, and the share climbs to 1 in 4 on Microsoft Copilot. ChatGPT is by far the heaviest Reddit citer, responsible for 59% of all Reddit citations in the study, while Claude cited none at all, consistent with our Claude citation study.
We learned this one the hard way. During our Reddit engagement experiment, both of our experiment communities were banned for using bots to upvote and comment. We would not recommend it. Every URL those communities earned in AI answers is now a dead citation we cannot edit, and the ban is permanent.
The evidence: one of our two experiment communities, banned for violating Reddit’s rules after we used bots to upvote and comment. Both communities are gone for good.
What this means for your GEO strategy: Reddit visibility built on manipulation is a double loss: the platform bans you, and the citations that survive point at removed threads you can no longer edit or defend. Earn Reddit citations with real participation, because the crackdown is finding the rest.
The GEO Playbook for Dead Citations
Audit your own inaccurate dead content. Lost, outdated, or forgotten PDFs, reports, help docs, and comparison pages may still feed AI Search Engines the wrong information about your brand. Correcting dead content after it lives inside an AI answer is difficult, especially when the original source is hard to find or no longer maintained. Catch inaccurate citations early in OtterlyAI before they mislead your target audience, distort brand perception, or affect ROI.
Remove or update outdated PDFs first. PDFs are the biggest stale-source type in this study: 77% undated and 40% of dated ones older than five years. An old PDF quoting dead pricing or discontinued products can corrupt what AI Search says about your services today. Refresh the important ones, delete the rest, and put dates in your file names.
Do not use bots on Reddit. Reddit’s crackdown is real, 1 in 7 cited Reddit URLs is already unpublished, and we have firsthand evidence: both communities from our Reddit engagement experiment were banned for bot upvotes and comments. The citations that remain point at threads we can no longer touch. Earn Reddit visibility with genuine participation or not at all.
A Wikipedia attempt can outlive Wikipedia’s verdict. Our data shows engines citing over 120 draft-space topics, pages that moderators pulled or never approved. Even a page that gets redrafted can leave an entity trace that gives your brand a shot at being cited. Treat this as an observation, not an invitation: it works until it doesn’t, and a rejected page is a record of the rejection too.
Word of caution: gray hat tactics, it works until it doesn’t. Some AI Search experts may use dead citations deliberately, publishing content in places like Wikipedia, Reddit, or comparison pages with the expectation that it may later be deleted, moderated, or taken offline. Dead citations are unstable, engine-specific, and highly dependent on crawl behavior, retrieval systems, and model updates. Algorithms often change or get corrected.
The safer GEO playbook is to monitor, not manipulate. Track removed URLs, stale pages, dead citations, and competitor citation gaps in OtterlyAI. Separate live Link Citations from dead citations, because visibility pointing to deleted or outdated sources should be counted as risk, not a win.
Final Conclusion
Our check of 20M+ cited URLs points to a clear picture of how much of AI Search rests on content that is already gone:
19.3% of cited URLs are not live: deleted, moved, or unreachable. Nearly 1 in 5 sources AI Search Engines cite leads nowhere.
ChatGPT carries the most dead citations, at 25.1% of its citations and about 1.4 million dead URLs in a month, partly because it cites a far wider set of sources than any other chat engine. Google AI Overviews is cleanest at 12.6%.
The big reference sources are quietly rotting. 1 in 7 cited Reddit URLs is no longer published, 1 in 13 cited Wikipedia URLs is not a live article, and cited PDFs skew old and undated.
The real number is higher than 19.3%. A status check only sees dead links. Stale but live content, like the evergreen PDFs above, is dead content it cannot count.
Dead citations are about content, not links. AI Search Engines answer from what they remember, and they remember pages long after those pages are removed, archived, or left to age. Manage removals and updates with the same care as launches, because the page comes down in seconds and its citations come down in weeks. To see it happen in real time, read the companion live takedown experiment.
Got a GEO experiment idea? Let’s collaborate
At OtterlyAI, we don’t write about AI Search from the sidelines. We test it. With the community, we run and update a public GEO Experimentation Tracker showing what we are running and what moved AI Search. If you are running your own GEO experiment, email rick.tousseyn@otterly.ai and we may feature it with your name attributed.
👉 Want to measure your AI Visibility? Start with OtterlyAI and our GEO Guide.
