You Said Friday. AI Heard a Promise. Can You Spot What Changed?
Three AI-style edits sound polished. Each changes something the original writer said. Can you spot all three before reading the answers?
By Henry Chimanzi | FrontierAI.fi
You asked an AI assistant to tidy up a message. It gave you a cleaner sentence. You read it once and hit Send.
The deadline still says Friday. What did you just agree to send?
Below are three fictional edits. Read each pair and decide what a recipient might reasonably believe after reading the new version. Give yourself one point for each change you spot.
1. The Friday email
Before: I’ll send you an update on Friday, even if the report is still in review.
After: I’ll send you the reviewed report on Friday.
2. The customer complaint
Before: We can replace the item once the warehouse confirms it was damaged in transit.
After: We can replace the item while the warehouse confirms whether it was damaged in transit.
3. The project handover
Before: I checked the figures in the draft budget. Maya will approve the final version.
After: I approved the figures in the draft budget. Maya will review the final version.
The answers
1. An update became a deliverable. In the original, the writer could meet the Friday commitment by sending a progress email. In the rewrite, the recipient expects a reviewed report.
2. The order of events changed. The original makes confirmation a condition for replacement. The rewrite offers replacement while the check is still underway.
3. Responsibility moved. The writer originally checked figures, while Maya had final approval. The rewrite gives the writer an approval role and describes Maya’s role as review.
The tricky part is that all three edits sound plausible. You can read them for grammar and tone, find nothing wrong and still send a message that changes what someone expects.
Now give your AI the same quiz
Open a fresh chat and paste the three Before and After pairs with this instruction. Leave out the answer key.
“For each pair, explain what someone reading the revised message could reasonably expect that they could not expect from the original. Identify any change to a promise, condition or person’s responsibility. If the meaning is the same, say so.”
Give the assistant one point for each pair where it identifies the main change. For the third pair, look for both changes in responsibility. Compare its first response with the answer key.
There’s a second question worth trying. Give the assistant only the three original sentences and ask it to make them sound more professional. Then check whether its own rewrites would pass the same test.
An assistant could spot every problem in a comparison and still introduce one when it writes from scratch. Those are two different tasks. This quiz lets you try both with the same examples.
What would you check before pressing Send?
I’d read the revised message from the recipient’s side. What would they now put in their diary? What would they tell their manager? Who would they think signed off the decision?
If their answers would change, you have an editorial decision to make. Sometimes the stronger wording is exactly what you want. Make that choice before someone else starts planning around it.
If you share your result, include your score, the assistant’s score and the model name if it’s shown. These examples are fictional, and this article reports no model test results.
Sources & further reading
- FrontierAI.fi — The AI Rewrite Test (related reading)
- FrontierAI.fi — The One-Word Test (related reading)
Sources reviewed 8 Nov 2026. Analysis and reading guidance are editorial interpretation; provider claims are not independently benchmarked.