AI change summaries

On the Scale plan, a monitor can add a short AI summary and a 0 to 100 importance score to every alert. Instead of reading a raw diff, you see one or two sentences saying what changed, and how much it matters against what you told Verid to care about.

What it does

When a check finds a change that matches the monitor's alert rule, Verid sends the changed fields, their old and new values, the monitor's name and URL, and your "what matters" note to Anthropic's Claude Haiku model (OpenAI's gpt-4o-mini is used as a fallback if Claude is unavailable). It returns:

• summary: one or two plain sentences, with the old and new values when they are short.
• importance: an integer from 0 (noise such as timestamps or rotating ads) to 100 (act now).
• reason: one short sentence explaining the score.

The summary sits at the top of the Slack, Discord and email alert, in the webhook payload, on the run page in the dashboard, and in the MCP server's list_changes and get_run results. Summaries are made only for changes that would send an alert, not for every check. Long values are cut before they are sent (about 1,500 characters per value), so on full-page monitors the summary describes the start of each changed value.

Turn it on

In the dashboard, open a monitor's settings and switch on Summarise each change with AI, then optionally say what matters and pick when to send the alert. Through the API, set the three fields on POST /v1/monitors or PATCH /v1/monitors/{id}:

curl -X PATCH https://api.verid.dev/v1/monitors/{id} \
  -H "Authorization: Bearer vrd_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "ai_summary_enabled": true,
    "ai_importance_prompt": "Only price drops and going out of stock matter. Ignore review counts.",
    "ai_min_importance": 50
  }'

Importance and thresholds

The score is judged against ai_importance_prompt. Write it the way you would brief a colleague: "Only price drops matter, ignore stock wording", or "Tell me about new plans or removed features, not copy edits".

With ai_min_importance set, a change that scores below it is still recorded (the run, the diff and the summary are all in the dashboard) but no alert is sent, and the run is not marked as having sent an alert. This is the way to keep a broad monitor, such as a whole page, from alerting on noise.

In the webhook payload

Webhook payloads gain an optional ai object. It is present only when a summary was made for that run, so receivers that don't know about it keep working unchanged.

{
  "id": "del_...",
  "version": "2026-05-01",
  "monitor_id": "6f1c2a9e-...",
  "run_id": "b41d0c77-...",
  "fired_at": "2026-10-17T06:00:04.000Z",
  "diff": {
    "fields_changed": ["price"],
    "before": { "price": "20.00" },
    "after": { "price": "18.00" }
  },
  "monitor": { "url": "https://shop.example/widget", "name": "Widget price" },
  "ai": {
    "summary": "The widget price dropped from 20.00 to 18.00.",
    "importance": 82,
    "reason": "A 10% price drop is exactly what you asked to hear about."
  }
}

Plan and limits

• Included on Scale and Scale + SERP at no extra charge.
• Each summary uses one AI call from the plan's monthly AI allowance (25,000 on Scale), the same allowance prompt-based extraction uses. A summary already made for the exact same change is reused and not counted again.
• If you move to a plan without summaries, monitors stop getting them automatically; you can still switch the setting off, but not on.

When a summary fails

A summary never blocks an alert. If the AI call fails, or the monthly AI allowance is used up, the alert is sent exactly as it would be without summaries, and only a missing summary on that run shows it happened. The importance threshold applies only when a score was actually produced.