---
name: sitetooling-traffic-drop
description: Diagnose a drop (or spike) in website traffic with Sitetooling analytics. Use when the user asks why traffic, visitors or page views fell or rose, says "traffic dropped", "where did my visitors go", "what happened on Tuesday", or notices a sudden change. Needs the Sitetooling MCP server connected.
---

# Traffic drop diagnosis

Find which pages and sources lost the most, rank them, and give the likely cause.

## Steps

1. Work out the two periods. If the user named a day ("Tuesday"), compare it with the same weekday a week earlier. If they named a week or month, compare it with the one before. If they named nothing, use the last 7 days against the 7 days before.
2. Call `compare_periods` with those dates. It returns the movers by page, source and country, biggest first.
3. Call `get_traffic_summary` for the wider stretch (e.g. the last 30 days) to see the daily trend, so you can tell a one-day dip from a slide.
4. If the drop is still happening, call `get_realtime_overview` to check whether visits are arriving right now (a site that's down shows zero).
5. If the plan includes replays, call `list_session_replays` for the affected page and the period, to see whether visits there end quickly.

## What to write

- First line: the site, the two periods, and the overall change ("Visitors down 34%: 1,210 vs 1,830").
- **Where it fell**, ranked: the 3–5 pages and sources that lost the most visitors, with numbers and % change.
- **Likely causes**, matched to the evidence:
  - one source fell (e.g. Google): ranking or search change, or a campaign that ended;
  - every source fell at once: the site, the tracking script or a redirect may be broken;
  - one page fell: the page changed, moved or lost its links;
  - weekday or holiday pattern: normal seasonality.
- **What to check next**: one or two concrete actions.

## Rules

- Say a change is too small to judge when either period has fewer than 100 visitors.
- Separate what the data shows from what you're inferring.
- Never identify people from replays; describe behaviour only.