Ask AI for plain English and "will" can become "may".
What the Reel says
Ask AI for plain English and "will" can become "may".
A jargon-heavy reply warns: "account will be suspended."
So I asked GPT-5.5 to simplify it.
3
of 3 runs softened
Told "may," customers might dispute and lose access.
Add this line: "don't add or soften anything."
Then both runs kept it firm.
When AI simplifies, make it keep warnings firm.
Sources and assumptions
- Today's AI Tip test runs, GPT-5.5 (OpenAI API), Nov 2026 (record.tests): Vague prompt 'Rewrite this support reply in plain English for the customer.' changed 'will be suspended' to 'may be suspended' / 'may be paused' in 3 of 3 runs; the stronger prompt kept 'will be suspended' in 2 of 2 runs; the stronger prompt's term list showed 'API access' → 'API access' and 'prorated' → 'we calculate' in run 1 (checked 2026-11-07)
- CFPB: How do I dispute a charge on my credit card bill?: Disputing a card charge is a real, common customer action (a 'chargeback' in the sample is a card-charge dispute), so the warning in the sample email is about a real situation (checked 2026-11-07)
Assumptions:
- The support reply is fictional sample material written for this test (customer 'Dana', agent 'Sam', a made-up Pro/Basic plan); no real company's policy
- Tests ran through the OpenAI API as GPT-5.5, labeled 'GPT-5.5 (OpenAI API)', not the ChatGPT app; the app may behave differently
- Claude tests were attempted but the test tool failed (entries with output null in tests.json); nothing may be claimed about Claude
- '3 of 3' and '2 of 2' are small samples from our runs, not a general rate; say 'in our test', not 'always'
- The fix is shown working in our runs; it reduces but does not guarantee no softening, so the reader still compares warnings with the original


