What an AI Correction Campaign is
An AI Correction Campaign turns one representation problem into one accountable object. It stitches AITWIRE's existing steps — detect, correct, publish, re-measure, prove — into a single tracked lifecycle you can watch, approve, and defend. Each campaign carries the detected problem, the affected prompts and AI systems, the competitor benefiting from the gap (if any), the correction you approved, the baseline score, the scheduled re-measurements, the measured outcome, and a downloadable evidence brief.
The lifecycle
A campaign moves through five stages:
- Detected — a drift, displacement, or gap alert opened a campaign. Nothing has changed yet; it is waiting for your approval.
- Approved — you approved the correction. AITWIRE locks the baseline: the pre-correction score it will measure against.
- Publishing — the approved facts publish to the machine-readable surfaces AI systems read.
- Measuring — AITWIRE re-probes the affected prompts across the AI engines on its decaying re-measurement cadence.
- Verified — the improvement cleared the statistics. Until it does, the campaign stays in Measuring and reports its honest, un-cleared reading.
How "verified" is earned
A campaign is marked Verified only when the 95% confidence interval of the re-measured score clears the locked baseline — not on a single lucky reading. If the interval still overlaps the baseline, the campaign reports "measured, not yet significant"; if it sits below, "measured, no improvement." AITWIRE proves what changed with statistics. It never promises an AI system will recommend you, and it never claims a correction caused a change — where a dollar figure is shown, it is an estimated, attribution-based movement over the window: a correlation, not proof.
Where campaigns come from
Campaigns open automatically from your fired AI Monitor alerts (a wrong fact, a missing category association, a drop in accuracy, negative tone, a hallucination) and from statistically-confirmed competitor displacement — a competitor being cited where you are not, that survives the McNemar + BH-FDR test. You can also scan for issues on demand from the Corrections board, or open a campaign yourself.
The evidence brief
Every campaign has a downloadable evidence brief: the problem, the affected panel, the per-engine before-and-after with confidence intervals, the significance gate, the lifecycle timeline, and a hash-chained integrity anchor. It is print-friendly — open it and save as PDF for a stakeholder or an auditor.
Availability
AI Correction Campaigns are available on the Pro and Custom plans, under Prove → Corrections.