Margaret Atwood on Claude: 'garbage in, garbage out' after one try

Margaret Atwood said she tried Anthropic’s Claude exactly once and left unimpressed, telling an audience at the Babell Literary and Cultural Festival in Porto, Portugal that the chatbot gave her a wrong answer about the British detective series Father Brown. She warned that language models reflect the limits and gaps of their training material and that users must check their outputs.

What happened when Atwood used Claude?

Margaret Atwood tried the chatbot precisely one time and received an incorrect plot answer, which she described bluntly: "Claude gave me the wrong answer, or it lied." She said she had been looking for information about the series Father Brown and that Claude had been misled because it "had skimmed and sampled a lot of television reviews, but they never give away the ending in online criticism, so it was misled by the things it had read about the show."

The account places the interaction at the Babell Literary and Cultural Festival in Porto, Portugal, where the author was interviewed. Atwood framed the error as a consequence of how large language models learn: from previously published material that may omit key facts, spoilers or endings, producing confidently stated but incorrect responses when the underlying sources are incomplete.

How did Atwood describe people who use AI?

Atwood said some users chase convenience, calling them "opportunists" who look for easy shortcuts that are hard to detect. She drew a line between human motivations and machine limitations, saying, "Human beings are not robots, but they are opportunists, so if there’s an easy way to cheat and it’s hard to detect, people will do it..."

She expanded that point into a critique of relying uncritically on AI: "But the thing about AI is that it’s garbage in, garbage out. Even people who use it for business reasons have to check it because it makes mistakes." Her remarks tied the bot's factual failure to both the sources those models sample and to how people might misuse the technology for convenience rather than verification.

Why it matters

Atwood’s anecdote illustrates a persistent problem with large language models: they can produce fluent but false statements when their training material lacks necessary specifics. The example about Father Brown shows how model outputs can be skewed by the nature of public writing—reviews and summaries often omit endings—and that can lead to confidently stated errors. That combination matters for anyone using LLMs for research, creative work, or business: users and consumers cannot assume factual reliability without independent checking.

What to watch

Watch for more public figures and authors recounting concrete AI failures, and for whether researchers or companies address how models handle spoiler-prone or summary-heavy sources. A clear next signal would be visible developer changes to training or retrieval methods that target the kinds of omission-based errors Atwood described, or more reporting of specific hallucinations tied to the same source-type problem.