AI output as object of study
Original research draft · October 2026
Working pattern
[System/model as identified], response to [prompt description] ([date]), [preserved record].
Capture from the source
Prompt; output; available version; settings; date; preserved artifact
Check before you use it
Use this form when the output itself is evidence or the research object, and verify underlying legal claims separately.
Common error: Generated text is not an independent legal authority merely because it contains citations.
Completed example
System, model label, session, letter, and generated response are fictional. The example reports no actual model behavior and makes no legal-authority claim for AI output.
TRAIN-LanguageModel (displayed label TRAIN-LM-3), response to prompt requesting a summary of letter TRAIN-OL-80 (Aug. 5, 2025), preserved session TRAIN-W80, output ¶ 3.
Read the supplied source facts
Fictional packet TRAIN-W80 preserves a research session with the invented system TRAIN-LanguageModel, whose interface displays model label TRAIN-LM-3 but no finer build number. On August 5, 2025, the researcher asks it to summarize a supplied fictional agency letter. The response overstates the letter as binding every tribunal, although the supplied letter limits itself to a specific arrangement. The packet retains the exact prompt, supplied text, output, visible model label, timestamp, and settings available to the researcher. The assignment studies the system's overstatement rather than using its response as legal authority. The letter identifier is TRAIN-OL-80, and the cited output passage is paragraph 3.
Why this works
The citation identifies generated output as the object being examined. It should not imply that the output independently establishes the agency letter's effect. Compare the response with the supplied letter and attribute the overstatement to the system's answer. Record only the model information actually visible; inventing an exact backend build would create false precision. Preserve prompt context because different supplied material or instructions could produce a different answer. If the output contains source citations, inspect those sources separately before relying on their legal propositions, and distinguish an output's citation string from a verified supporting authority.
When the facts change
If a later run gives a different answer, preserve both sessions and identify the changed prompt, settings, or available version information. Do not overwrite the earlier output used in the study. If AI merely assisted drafting and its output is not the research object, document that assistance according to the relevant disclosure requirements rather than citing generated text as though it were primary law.
Apply the receiving court or publication’s requirements. These working forms do not establish citation permission, precedential weight, or support for a legal proposition.