AI explained the result. Can the affected person actually challenge it?

Our view: an explanation and a working correction route solve different problems. Check where a person can present evidence and obtain a review.

The OECD AI Principles link understandable information with enabling people adversely affected by AI to challenge its output. That is a governance principle, not a guarantee that every service supplies the same process or a particular legal right.

An explanation can leave the result untouched

In an invented event-registration example, a person is put on a waitlist because the system uses an incorrect date. A clear statement of that reason helps locate the problem. Repeating the reason does not itself correct the date or reconsider the place.

Ask what happens after a correction request

Our proposed review asks: where can the affected person submit the correct information, who can reassess it, what response will they receive and how can they tell whether the result changed? These are editorial questions, not a claim that the OECD mandates this exact checklist.

Do not send everything to another chatbot

A second fluent answer does not establish that someone with authority reviewed the original outcome. Use the service’s published contact or correction channel and provide the relevant information it is permitted to receive. Do not invent a successful appeal or a deadline the service has not stated.

Describe the route honestly

For an organization using AI, our recommendation is to test the correction path as well as the explanation. Say which outcome can be reconsidered and which questions remain unresolved. Applicable procedures and rights depend on context; this column does not determine them for a real case.

Published October 14, 2026. Editorial analysis of a published principle with a fictional low-stakes registration example. No service, complaint, appeal, user study or legal case was tested or submitted.

Sources: OECD · AI principles.