How to set up alerts for AI visibility drops.
Tell me the day a top-50 product falls out of the answer — not at month-end. Regressions discovered weeks late cost the entire recovery window: by the time the quarterly review notices, the slot has been a competitor's for two months.
The slow way: discover it in a meeting
Without alerting, regressions surface socially. Someone mentions soft traffic in the Wednesday trading call; someone else remembers the SKU used to show up in ChatGPT; an analyst gets assigned to "look into it" and comes back next week with a confirmation and no date. Now the team is doing forensics on a drop that happened five weeks ago, with no record of what the answer looked like before.
The manual mitigation is spot-checking on a schedule — rerun the hero prompts every Monday and eyeball the results. That is better than nothing and shares the weakness of every recurring manual task: it decays, it covers only the prompts someone remembers to run, and it cannot watch the second score at all. A robots.txt edit or a theme deploy that silently breaks JSON-LD shows no symptom a prompt check can see until the citations are already gone.
The eCommerce Insights way
- Define the segments worth interrupting for. Top-50 revenue SKUs, the hero collection, this quarter's launch cohort. Alerts on everything mean alerts on nothing — the long tail belongs in the weekly digest, not the interrupt channel.
- Set thresholds per segment. The defaults that work: citation count to zero on any engine, composite score drop over 15 points, any crawler flipping from allowed to blocked, Product JSON-LD disappearing from a PDP. Tighter on hero SKUs, looser elsewhere. Configuration details in the alerts docs.
- Pick the cadence. Alerts fire on the refresh that detects the breach — weekly on Starter, daily on Growth. Daily is what makes "tell me the day it broke" literal, and earns its cost in peak season and launch windows.
- Route to the people who act. Per-segment recipients: the merchandiser for hero-SKU citation drops, the SEO lead for schema and crawler breaks, the account owner per client for agencies. An alert nobody owns is a notification, not an alert.
- Tune monthly. Mute what nobody acted on; tighten where something was caught late. The target state is a quiet channel that is genuinely alarming when it speaks.
When an alert fires, the follow-on is the forensic job: recover a lost AI citation — date the drop, attribute it, ship the matching fix. The two workflows share the same per-SKU history.
What "good" looks like
The honest metric is the last row. Teams running tuned alerts plus the weekly digest stop being surprised in meetings — the surprise arrives by email, dated, with the diagnosis attached, while the recovery window is still open.
Ask AI about this job
Have your favorite AI engine apply this walkthrough to your catalog.
Frequently asked questions
What can I set alerts on?
How fast do alerts fire?
Won't I just get alert fatigue?
What's the difference between alerts and the weekly digest?
What should happen when an alert fires?
The day it breaks. Not the month-end.
Threshold alerts per segment, daily refresh on Growth. 14-day trial.