Should I ask the same AI to grade its own work?

Watch the briefing

Watch the full video on YouTube

Full briefing transcript

The conference packing draft says everything is ready. Then its own review says the list is complete. But the source asks for a registration confirmation and a charger, and the list includes neither. A second confident answer from the same system is not independent evidence. Self-review can help you find candidate issues; use a separate checklist grounded in the request to decide what actually passes.

Our fictional attendee is preparing for a one-day conference on November ninth. The notes require a registration confirmation, photo identification accepted by the organizer, a laptop, its charger, a reusable bottle, and a notebook. Laptop use is planned for an afternoon workshop. We are drafting a packing list, not checking entry policy or claiming that a particular identification document will be accepted. The source requirements define this limited exercise.

The bad illustrative list has photo identification, a laptop, a bottle, and a notebook. Its equally illustrative self-review says all required items are included. We wrote both texts for the lesson; no named model produced them. The review repeats the conclusion without showing a comparison to every source item. That is exactly what we should inspect: what evidence supports the grade, and which requirements were actually checked?

Build the independent checklist before accepting the review. A person now compares the draft with the six source requirements. Confirmation is absent. Identification is present but should retain the organizer's acceptance qualifier. Laptop is present. Its charger is absent. Bottle and notebook are present. We have two omissions plus a wording qualification to restore. The word complete has not survived this explicit comparison.

The corrected illustrative list includes the registration confirmation, organizer-accepted photo identification, laptop and charger, reusable bottle, and notebook. The human reviewer traces all six back to the source and checks that no new requirement appeared. This independence comes from the checking process and source comparison, not merely opening another chat. A different AI could suggest another opinion, but that opinion still needs validation against the task.

You can still ask for self-review in a bounded way: identify possible omissions against this checklist and quote the source item for each concern. Treat the answer as a list of things to inspect, not an approval certificate. For important facts, policies, or safety requirements, check the relevant original material or qualified person. Our small packing checklist covers supplied requirements only; it is not a universal conference preparation guide.

Self-review can suggest issues; it is not independent verification. Keep the request, criteria, and original evidence available, and make the acceptance decision through an explicit check. In our example, two missing items were visible even though the self-grade sounded certain. Next, we will choose between announcement drafts using audience needs and keep the final editorial decision with the person responsible.