The answer or result is wrong

An incorrect result is information about what needs checking. It does not mean you need to learn a new way of speaking before you can use your AI employee.

Keep the result in review while you identify the problem. Do not approve a draft you already know is wrong just to keep the work moving.

Name the kind of mistake

Choose the closest description:

• A fact is wrong: a name, date, amount, or claim does not match your source.
• The wrong material was used: the result refers to an old file or the wrong project.
• The work is incomplete: something you asked for is missing.
• The style is off: it is too formal, too long, or does not sound like your business.
• The job was misunderstood: the result solves a different problem.

You can have more than one issue. Start with the one that would make the result unusable.

Show the correction

Point to the exact sentence, section, or item. Provide the correct source or an example of what good looks like. Separate facts that must change from preferences you would like it to follow.

Make this request your own

Please revise this result. The problem is [specific issue]. Use [correct source or example]. Keep [part that is already useful]. Before changing it, tell me what you understand the correction to be. Return a revised draft and list what changed.

For a factual issue, ask where the original information came from. If no reliable source is available, leave the claim out or mark it for your review. A confident sentence is not evidence.

Check the revision against your source

Open the revised result and check the corrected item yourself. Also check that the change did not alter a nearby date, amount, name, or decision.

When the problem is style, give a short positive example: “Use this greeting and keep the message to three short paragraphs.” That is easier to apply than “make it better.”

If the preference should apply to future jobs, follow teach your preferences. Ask how the preference is recorded in your setup and test it on another small job. One corrected response does not establish that every future result will change.

Escalate a repeating or consequential problem

If the same error returns after a clear correction, ask for help. Include a safe example of the expected result and the actual result. If an incorrect action has already affected someone else, describe what happened and pause further related work through your documented process.

You are back on track when: the result matches the source and your stated requirements, or you have clearly stopped its use while the issue is investigated.