Ask AI which sentence your reader will misread.
What the Reel says
Ask AI which sentence your reader will misread.
You know what you meant. They don't.
One short email. Two ways to read it.
Tell AI it's for a weekend cashier.
It found this line right away.
Does "except" mean no approval, or no notice?
Its fix erased the weekend rule. Policy changed.
It flags the line. You write the fix.
Sources and assumptions
- Prompting best practices (Claude docs, Anthropic): General principle behind the move: "Think of Claude as a brilliant but new employee who lacks context on your norms and workflows. The more precisely you explain what you want, the better the result." and "Key test: Show your prompt to a colleague with minimal context on the task and ask them to follow it. If they'd be confused, Claude will be too." Also: "Providing context or motivation behind your instructions ... can give Claude more information about your goals and help it deliver more targeted responses." The page has no line specifically about naming an audience, so do not attribute the misread-sentence prompt to Anthropic. (checked 2026-11-08)
- GPT-5.5 (OpenAI API) test runs, engine/ai-test.js: The 'after' (named reader, one sentence) and 'before' ("Make this email clearer.") outputs quoted in tests.json, run on 2026-11-09 UTC (2026-11-08 Pacific) via the Chat Completions API, model gpt-5.5 (checked 2026-11-08)
Assumptions:
- The email, Sam, Priya, Jordan and the Rota app are all fictional. Sam's real meaning of "except weekend shifts" is left undefined on purpose: the point is that the AI can't know it either.
- Only GPT-5.5 via the OpenAI API ran successfully. The three Claude Sonnet 5.5 runs in tests.json failed to run (output null), so the post may name only "GPT-5.5 (OpenAI API), Nov 2026" as tested and must not say it was tested in Claude.
- Exact prompt taught (verbatim, from the 'after-gpt' test): "Don't rewrite this email. My reader is Jordan, a part-time cashier who only works weekends and reads staff emails on a phone during breaks. Quote the one sentence Jordan is most likely to misread, say how Jordan could take it the wrong way and why, and suggest one replacement sentence."
- What happened, from the real outputs: 'after-gpt' quoted the right sentence ("Swaps must be approved by a manager at least 48 hours before the shift, except weekend shifts.") and explained Jordan could read it as weekend shifts needing no approval or no 48 hours' notice. Its replacement, "All swaps, including weekend shifts, must be approved by a manager at least 48 hours before the shift.", removes the exception without saying so. That's the limitation: it found the right sentence, but its fix changed the rule.
- 'before-gpt' ("Make this email clearer.") rewrote the whole email, dropped "except weekend shifts" from the body, changed "any shift" to "shifts", and only flagged the exception in a note at the end. The changes have to be found by comparing the two versions.
- Try it: paste your next staff email, name one real type of reader (role, schedule, where they read it), ask for the one sentence they would misread, and then write the fix yourself so it says what you actually meant.
- Privacy: on real staff emails, take out names and personal details first and use your employer's approved AI tools.


