LLM platform citation failures happen when an AI chat links a dead page, an outdated page or a third party for facts about your business, so the buyer who clicks lands somewhere you do not control. Each platform retrieves pages through its own crawlers and its search partners and selects its citations its own way, so a fix confirmed in one AI chat still needs retesting in the others. In March 2025, Columbia Journalism Review's tests of eight AI search tools found them citing the wrong article, publisher or link in more than 60% of 1,600 queries. Aemorph traces each bad citation to its cause, platform by platform, and gives each one a current page to cite.
LLM Platform Citation Failures Take Five Forms
LLM platform citation failures take five forms, a dead link, an outdated page, a third-party copy, a competitor's page or no source at all, and each sends your buyer to facts you did not approve. The failures appear across platforms: in Columbia Journalism Review's tests, Perplexity was the most accurate of the eight tools and still failed in 37% of cases.
Source: Columbia Journalism Review, March 2025
| Failure | What the buyer sees | Platform pattern | Likely cause to test | Fix |
|---|---|---|---|---|
| Dead or invented link | The cited link opens an error page | Reported for several AI search tools in Columbia Journalism Review's tests | An old URL was removed without a redirect, or the chat generated a link | Redirect retired URLs to the current page |
| Outdated page on your site | The link opens last year's price or service page | Appears on any platform whose index still holds the old page | The old page is still live and indexed | Update or redirect the old page and keep one current page per fact |
| Third-party copy | A directory, aggregator or reseller page is cited for your facts | Appears on category and comparison questions | The copy states your facts more plainly than your own page | State the facts in full on your own page |
| Competitor's page | A competitor's page is cited for a question about your type of service | Appears on "best" and "how much" questions | Your page does not answer that question | Answer the question on your own page |
| No source shown | A confident answer with nothing to click | Appears when the chat answers without searching the web | The chat answered from training data | Publish the fact where the platform's search crawler reads it |
The five LLM platform citation failures, with the platform pattern and the fix for each.
Case study · Free Hallucination Report, wine and spirits retail, Singapore
ChatGPT and Gemini sent The Liquor Shop's collection customers to two different, wrong warehouse addresses.
The website offers same-day collection at Kallang Distripark. Across 31 facts, 30 answers contradicted the website.
Source: Aemorph Hallucination Report, October 2026
The Hallucination Report, Aemorph's free hallucination audit tool, puts the facts on a business's own website to the AI chats and lists the answers that disagree. When the engines cite different facts as well as different pages, cross-platform AI brand inconsistency shows how each one settles the conflict its own way.
Why does an AI chat link to the wrong page?
An AI chat links to the wrong page when its index holds an older or clearer version of the fact than your current page, or when it generates a link from memory instead of retrieving one. The cause sits in the platform's evidence, so the fix changes that evidence.
Each AI Platform Reaches Your Site Through Its Own Crawler and Index
Each AI platform selects its citations from what its own crawlers and its search partners retrieve, so ChatGPT, Claude, Perplexity and Google's AI answers often cite different pages for the same question. Ahrefs found that, on about 3,000 short-tail queries, 65% of Perplexity's citations appeared in Google's top 10 against 10% of ChatGPT's. OpenAI states that ChatGPT search sometimes partners with other search providers to answer a question.
Source: Ahrefs, September 2025
| Platform | Search crawler or index | User-request fetcher | What to check first |
|---|---|---|---|
| ChatGPT search | OAI-SearchBot | ChatGPT-User | OAI-SearchBot allowed in robots.txt and at the firewall |
| Claude | Claude-SearchBot | Claude-User | Both Claude agents allowed; ClaudeBot (training) is a separate choice |
| Perplexity | PerplexityBot | Perplexity-User | PerplexityBot allowed; your Google rankings for the question |
| Google AI Overviews and AI Mode | Google Search index | Not applicable | Page indexed and eligible for a snippet in Google Search |
Crawler names from OpenAI, Anthropic, Perplexity and Google documentation.
Anthropic documents its agents in its crawler controls article, and Google states that a page is eligible as a supporting link in AI Overviews when it is indexed and eligible to be shown in Google Search with a snippet.
Sources: OpenAI, Anthropic, Perplexity and Google, October 2026
A crawler blocked in robots.txt or at the firewall leaves your site with no citations on that platform, the gap the no LLM citations fix closes.
Do ChatGPT and Perplexity cite the same sources?
ChatGPT and Perplexity often cite different sources for the same question, because each retrieves pages through its own crawler and search partners and selects citations its own way. A page Perplexity cites is often a page that already ranks in Google; ChatGPT's citations overlap far less with Google's results.
Which AI crawlers does a website need to allow to be cited?
Allowing each platform's search crawler lets it discover, fetch and include your pages in its answers: OAI-SearchBot, Claude-SearchBot and Claude-User, PerplexityBot, and Google's crawler for Google's AI answers. A page still gets linked when another source mentions it, and OpenAI states that sites opted out of OAI-SearchBot are left out of ChatGPT search answers, though they still appear as navigational links, so check each crawler by name.
Fixing Citation Failures Starts Platform by Platform
Fixing LLM platform citation failures means capturing what each platform cites, tracing each bad link to its cause and correcting the evidence that platform reads, then retesting on that same platform. This is GEO, which gets you cited inside the AI chats; it rests on SEO and works with AEO for Google AI Overviews. Aemorph runs the fix as a sequence:
- Captures the same set of buyer questions on each platform, with the date, the answer and each cited link.
- Traces each cited link to its status: live and current, live and outdated, redirected, missing or not yours.
- Opens access for each platform's search crawler by name, in a specification your team or vendor applies and Aemorph verifies.
- Redirects retired URLs to the current page, so an old citation lands on today's facts.
- Consolidates each fact onto one current page, and publishes your approved answers as the LLM info page on your site.
- Lists the third-party pages whose copies carry outdated facts about you, and Aemorph recommends that you request each correction from that site's owner.
- Retests each platform on the same questions and records what each one now cites.
Aemorph's Hallucination Diagnostic is the first platform-by-platform check: it tests a sample of 10 to 20 showstopper questions across ChatGPT, Gemini, Claude and Perplexity and compares each answer with your website.
How do you fix a broken link that an AI chat keeps citing?
Fix a broken link an AI chat keeps citing by redirecting the old URL to the page that now holds the fact, so the citation works while the platform's index catches up. Then retest on that platform, because each one refreshes its index on its own schedule.
After a site move, migration redirect recovery maps each old URL to its matching page, so the links AI chats cite keep working.
Citation Failures Send Buyers to Pages You Do Not Control
A failed citation hands the buyer's next click to a page you do not control, and sourcing is where AI answers go wrong most often. A study of AI assistants run by the European Broadcasting Union and the BBC found 45% of answers about news had at least one significant issue, and 31% had serious sourcing problems. The sources AI chats pick also differ from Google's: across 15,000 prompts, Ahrefs found only 12% of the links cited by ChatGPT, Gemini and Copilot in Google's top 10 for the same prompt.
Sources: European Broadcasting Union and BBC, October 2025; Ahrefs, 2025
| Failure | Risk level | Cost of leaving it | When to recheck | Next step |
|---|---|---|---|---|
| Outdated page cited for price or service | High | Buyers arrive expecting a price or service you changed | The day the fact changes | Update or redirect the old page |
| Competitor's page cited | High | The buyer's next click goes to a competitor | After your answering page goes live | Answer the question on your own page |
| Third-party copy cited | Medium | The third party's wording becomes your story | After your page states the facts in full | Publish the complete facts and request the correction |
| Dead link cited | Medium | Buyers hit an error and lose trust in the answer and in you | After the redirect is live | Redirect the old URL |
| No source shown | Medium | Buyers have nothing to check the facts against | After your answers are published | Publish the facts where each platform's crawler reads them |
Risk level and recheck point for each citation failure, as illustrative scenarios.
Aemorph sets no fixed recheck interval. Your first working day with Aemorph is the first check date, and pages are then assigned for update in order of importance: bottom-of-funnel, middle-of-funnel, then top-of-funnel.
Which citation failure costs the most?
An outdated price or service page cited to a buyer who is ready to decide costs the most, because it misleads the buyer at the moment of choice. A competitor's page cited for your type of service comes second, because it moves the click to them.
Citation Fixes Follow Each Platform's Published Crawler Rules
Each AI platform decides which source it cites, so citation fixes work through the rules each platform publishes for its crawlers and show the same content to people and crawlers alike. Showing crawlers different content from what people see breaks Google's spam policies and puts the platforms' trust in your pages at risk.
The training crawlers, such as OpenAI's GPTBot and Anthropic's ClaudeBot, are a separate decision from the search crawlers that decide citations. Aemorph sets out both choices and applies neither without your sign-off. Aemorph takes the facts you supply as true and correct and publishes only what your nominated point of contact approves.
Who decides which source an AI chat cites?
The AI platform decides which source it cites for each answer, drawing on what its crawler and index hold on the day. You decide what your pages say and which crawlers reach them; questions about content rights or licensing are for a qualified legal adviser in your jurisdiction.
The Showstopper Sensor Gives Each Platform One Current Page to Cite
Citation failures close when each AI platform has one current, approved page to cite for each buyer question, so a citation sends your buyer to facts you stand behind. The Hallucination Diagnostic tests a sample of 10 to 20 of your showstopper questions, the buyer questions that stop a purchase when AI answers them badly, across ChatGPT, Gemini, Claude and Perplexity, depending on the size and complexity of your business, with the report delivered in 5 business days and presented on a call.
Bought on its own, the Showstopper Sensor (Sprint 1), a 30-day showstopper-answer sprint for ChatGPT, Claude, Gemini and Perplexity, includes the Diagnostic and publishes your signed-off answers as the LLM info page on your site. On the enterprise route, Sprint 1 runs inside Attain or Apex after the audit and the Influencer Blueprint. Sprint 1 does not correct public records; Entity Hardening (Sprint 2), Aemorph's technical entity SEO sprint, audits your Wikidata and Google Knowledge Graph entries. The Diagnostic bought on its own carries no guarantee.
Buy the Showstopper Sensor (Sprint 1) on its own and it carries the Ground Truth Guarantee, an AI accuracy guarantee. The Ground Truth Guarantee: within 30 days of publishing, ChatGPT, Claude, Gemini and Perplexity answer your showstopper questions with the facts in your Brand Codex. If they do not, Aemorph refunds 100% of your Sprint 1 fee and keeps working on the answers for a further 60 days, 90 days in total.
Terms: the Ground Truth Guarantee covers Sprint 1 bought on its own; Sprint 1 inside a monthly package carries no refund; it does not cover Sprint 2 (Entity Hardening), search rankings or Google AI Overviews. Full terms
Start with a free strategic conversation, which includes prompting the AI chats for your current brand answers. Any business, at any size, then buys the Showstopper Sensor (Sprint 1) on its own; terms are set out in your quote, and starting points are on the pricing page.
See the Showstopper Sensor (Sprint 1)