A marketing mistake around Megalopolis has become a clear warning about generative AI: a sentence that sounds usable is not the same thing as a verified fact. In this case, fabricated review quotes made their way into a movie trailer, and the consultant behind them lost his contract.
According to Deadline, marketing consultant Eddie Egan was fired after using an AI tool such as ChatGPT to generate review quotes for the trailer. The issue was not simply that AI was involved. The problem was that the material presented as historical criticism of Francis Ford Coppola's earlier work did not match reality.
The trailer tried to make a case about misunderstood films
The apparent strategy behind the Megalopolis trailer was straightforward. Egan wanted to argue that the movie could follow a familiar pattern: first harsh criticism, then eventual recognition as a masterpiece.
To support that idea, the trailer used supposedly negative comments about Coppola's earlier movies. It cited renowned film critics including Pauline Kael of The New Yorker and Andrew Sarris of The Village Voice.
The trailer claimed that classics such as The Godfather and Apocalypse Now had been attacked in language that would look severe in any marketing campaign. Among the quoted lines were "sloppy, self-indulgent movie" and "epic piece of trash."
That framing would have served a useful promotional purpose if the quotes had been real. It would have suggested that early dismissal of Coppola's work had happened before, and that audience judgment can change over time.
But the foundation of that argument collapsed because the quoted criticism was not authentic.
The quotes were fabrications, not archival criticism
The scathing reviews used in the trailer never happened. The source article states that Vulture magazine reported the opposite: the critics had praised these films.
That distinction matters. A trailer can make a bold argument, but if it relies on named critics, named publications, and specific historical claims, those details have to be checked against the record. In this case, the supposed evidence was AI-generated fabrication.
Lionsgate responded by apologizing for the mistake, removing the trailer, and terminating Egan's contract. Those steps show how quickly an unchecked AI output can become a reputational and professional problem once it enters public-facing work.
The incident is especially sharp because the fabricated material was not vague. It involved named people, known publications, major films, and direct quotes. Those are the kinds of details that can sound authoritative when placed in a polished trailer, even when they are false.
ChatGPT generated words, not verified facts
The larger lesson is about how generative AI works. Large Language Models, or LLMs, generate words based on probabilities and are influenced by the user's prompt. They can produce sentences that read fluently and confidently without having a built-in fact-checking process.
That is a dangerous gap for research tasks. If a user asks for critical reviews, the system may generate critical-sounding material. The output may match the requested tone and format while still failing the basic test of factual accuracy.
This is why AI tools such as ChatGPT should not be treated as source databases. They can help draft, summarize, or explore a direction, but their outputs need independent verification before being used as evidence.
For marketing, journalism, law, entertainment, and any field that depends on trust, the issue is not only whether AI can produce useful language. The question is whether the user understands what the tool can and cannot confirm.
The same risk has appeared beyond marketing
The source article points to other examples of people being misled by reasoned-sounding chatbot output. Attorney Steven A. Schwartz initially used ChatGPT for research, unaware that the system could generate false content.
In another case, attorneys used ChatGPT to find and cite supposed reference cases that turned out to be AI inventions. Even OpenAI itself had a factual generation error in its first SearchGPT demo.
These examples share the same pattern:
- A user asks an AI system for information that appears factual.
- The system produces a polished answer that sounds credible.
- The answer is used or considered without enough independent checking.
- The false material creates consequences once it reaches a real audience or process.
The Megalopolis trailer case is therefore not just a film marketing mishap. It is a practical example of a broader misunderstanding: generative AI can imitate the form of research without actually performing reliable verification.
What the case says about AI in public work
The central problem was not that a marketing consultant used AI. The problem was using AI-generated claims as if they were confirmed historical facts.
For public-facing work, that difference is critical. A fabricated quote can travel farther than a private draft because it carries the authority of a name, a publication, and a finished campaign. Once it appears in a trailer, the audience has little reason to assume it came from an unchecked chatbot response.
The safer lesson is simple: AI-generated material should be treated as unverified until checked against trusted sources. That is especially true when the output includes names, direct quotes, past reviews, legal references, or claims about what someone said.
Egan's firing shows the professional cost of missing that distinction. ChatGPT can generate language that looks ready for use, but the responsibility for accuracy remains with the people who decide to publish it.