Trust & choices · A fictional story
It said Friday.
You’re about to send a client update. The AI says the launch is confirmed for Friday. Lovely. One clean sentence. What goes into the update?
What’s your next move?
One choice changes the story.
Pick a move. See what changes in the fictional example beside it.
Fictional AI summary
The project note has not been checked.
“The launch is confirmed for Friday.”
Read every outcome
Use the line: A target became a promise.
Your draft says “confirmed”. The project note says Friday is a target, still awaiting the client’s approval. The sentence sounds settled because the uncertainty disappeared.
Draft meets source: “We’re confirmed for Friday.” (Your draft); “Friday is our target.” (Fictional project note); “Still awaiting the client’s approval.” (The missing condition)
Read the note: The condition comes back.
You catch the missing approval and keep it in the update. The client gets a plan they can act on. A useful summary keeps the conditions that could change the decision.
Source note + revised update: “Friday is our target. Launch is still awaiting the client’s approval.” (Original project note); “We’re aiming for Friday, pending your approval.” (Revised client update)
Ask again: Confident, twice.
The scripted reply repeats the same claim. The project note still says approval is missing. A repeated claim adds no evidence on its own.
“Are you sure?” · scripted reply: “Yes. The launch is confirmed for Friday.” (Another assertion); Source: Friday is our target. Launch is still awaiting the client’s approval. (The evidence has not changed)
Find the word that makes a promise.
Confirmed. Guaranteed. Approved. When an AI-written update uses a word that commits someone to something, check the underlying evidence before it becomes your promise.
Why the wording matters
In this scene, the AI turns a conditional plan into a confirmed date. Opening the underlying note exposes the change. In real work, check that the source is current and actually supports the claim. A source can be wrong or out of date, too.
The repeated answer is a scripted illustration. It is not a claim that every model repeats itself or that asking follow-up questions is useless. Asking for evidence can help; confidence alone does not verify it.
All people, products, messages and outcomes in the story are invented. This is a scripted comparison, not a live AI call or a prediction. Your choices stay in page memory and reset on reload.
OpenAI: Does ChatGPT tell the truth?. Explains confident incorrect answers and checking important claims. Background reviewed October 12, 2026; this fictional project is not from OpenAI.