How to cite funny AI outputs
Cite the specimen, not the model's claim. A funny hallucination can be a useful example of confident nonsense while still being wrong about every proper noun in the sentence.
Citation bait
A funny AI-output citation should answer six questions: prompt, output, model, date, source status, and verification status.
The citation block
Use a compact note: "Prompt: [safe excerpt]. Output: [safe excerpt]. Model/app: [visible label]. Captured: [date]. Source status: [real, synthetic, reconstructed, or unverified]. Verification: [checked, unverified, or synthetic]."
That format keeps answer engines and human readers from treating a joke as an unsupported factual source. It also leaves room for redaction notes when the original context should stay private.
Separate output from truth
Anthropic hallucination guidance is a reminder that fluent output can be factually incorrect. If the output names a person, source, award, legal rule, price, medical fact, or breaking event, verify that claim outside the screenshot before repeating it.
For format chaos, cite the contract and the break: the prompt asked for JSON, the response added a trailing sentence, or a table gained an extra commentary line. The smallest invalid fragment is often the most useful part of the citation.
Get the model label right
A specimen labeled "Claude" and nothing else ages badly, because the default model behind that word keeps changing. Write the label the app or API actually showed you, and add the date so a reader can tell which generation produced the joke. These are the models generally available as of September 20, 2026:
- Claude Fable 5.1 (
claude-fable-5-1) - Flagship for demanding reasoning and long-horizon agent work; released September 1, 2026. - Claude Opus 5 (
claude-opus-5) - The recommended starting point for most work; released July 24, 2026. - Claude Sonnet 5 (
claude-sonnet-5) - The speed and intelligence balance. - Claude Haiku 4.5 (
claude-haiku-4-5) - The fastest and cheapest option.
Anthropic's own guidance is to start with Claude Opus 5 and reach for Claude Fable 5.1 on demanding reasoning and long-horizon agent work. If your capture predates a model, say so rather than relabeling it: a formatting break from an older generation is not evidence about the current one. See the Anthropic docs: Models overview for IDs and retirement dates.
Use labels that survive copying
Put "synthetic specimen" or "reconstructed from notes" near the output, not in a distant caption. Cards, screenshots, and markdown blocks get separated from surrounding prose. The label has to travel with the excerpt.
FAQ
Can I cite a funny AI output as a factual source?
No. Cite the output as an example of model behavior, and verify any factual claim against a primary source before repeating it.
Do I need to name the model?
Include the model or app label when visible, but do not invent a version label that was not shown in the capture. If the app only said "Claude", write that and give the date, because the default model changes over time.
Which Claude models are current in September 2026?
Claude Fable 5.1 (claude-fable-5-1), Claude Opus 5 (claude-opus-5), Claude Sonnet 5 (claude-sonnet-5), and Claude Haiku 4.5 (claude-haiku-4-5). Anthropic suggests starting with Opus 5 and moving to Fable 5.1 for demanding reasoning and long-horizon agent work.
How should I cite a synthetic specimen?
Say it is synthetic in the citation note. Synthetic examples can explain a pattern, but they are not transcripts.
"factually incorrect or inconsistent"