# AI Result Evaluation (/docs/products/ai-output-evaluation) A confidence score doesn't tell you whether an AI-generated explanation actually makes sense to a human. AI Result Evaluation asks a crowd panel to rate exactly that — for any completed Multimedia Verification or Language Technology run, or for the output of your own AI system — and logs every rating as a durable feedback signal. Request an evaluation against a completed run and Crowdee automatically generates the crowd task, showing the run's pipeline, verdict, and explanation — no task template required. Reviewers rate clarity, evidence sufficiency, actionability, and bias risk, and can flag a result for retraining. Individual ratings roll up into a single transparency score. Have your own AI's output and no Crowdee run to point at? A dedicated endpoint accepts the verdict and explanation directly — see [Evaluating Your Own AI's Output](/docs/concepts/ai-output-evaluation.mdx#evaluating-your-own-ais-output). ## What You Get [#what-you-get] * **A transparency score (0–100)** per evaluated run, aggregated from at least three independent crowd ratings. * **Per-response feedback**, including free-text notes and retraining flags. * **An append-only feedback log**, keyed by the specific AI component evaluated — the durable record behind improving it over time. ## Cost [#cost] 45 credits by default (15 credits per crowd response × 3 responses), blocked upfront when you request the evaluation. Requesting more responses for a higher-confidence score costs proportionally more. ## Where to Go Next [#where-to-go-next] See [AI Output Evaluation](/docs/concepts/ai-output-evaluation.mdx) for the full technical reference: the request API, rating dimensions, batch statuses, and the feedback log. For an always-on version of this same evaluation, see [Continuous Monitoring](/docs/products/continuous-monitoring.mdx).