Traffic & Quality Control
Keep your data clean: detect anomalies, suspicious traffic, broken UTMs, and tracking regressions before they corrupt decisions.
MCPs required: Sealmetrics MCP
SEAL-014 — Daily traffic anomaly digest
You are an analytics on-call engineer. Using Sealmetrics MCP for site {site_id}, run a daily anomaly check for {today} vs the trailing 30-day baseline.
Check anomalies on:
- Total entrances
- Conversions
- Revenue
- % bot/agent traffic
- Top 10 channels (entrances delta)
- Top 10 landing pages (entrances delta)
Flag any metric that deviates more than 2 standard deviations from its 30-day mean. For each flagged anomaly, provide: metric, baseline, today, % delta, plausible cause (paid spike, organic shift, tracking break, attack, holiday).
Output: a "Today's anomalies" markdown report ready to paste in Slack. Maximum 8 anomalies, ranked by business impact.
SEAL-015 — Top suspicious sessions of the week
Using Sealmetrics MCP for site {site_id}, return the top 50 most suspicious sessions in the last 7 days based on the bot suspicion score.
For each session: session id, suspicion score, detected flags, browser and device family, country, traffic source, entry landing, number of pageviews, micro and macro conversions if any. (Sealmetrics never stores visitor IPs or raw user-agent strings, so neither column exists.)
Group by source. End with a "candidates to block" list: the flags, browser/device families and sources that recur.
SEAL-016 — Bot share weekly trend
Using Sealmetrics MCP for site {site_id}, compute the weekly share of bot/agent_suspected traffic for the last 12 weeks.
Per week: total entrances, bot entrances, % bot, week-over-week delta.
Plot the trend in ASCII or describe it in a one-paragraph summary. Alert me if the latest week is more than 15% AND more than 3 percentage points above the trailing 12-week mean.
Add a section listing the top 5 sources contributing to the bot share this week vs last week.
SEAL-017 — Suspicious new referrers in top 20
For site {site_id}, query Sealmetrics MCP for the last 7 days.
List the top 20 referrers by entrances. For each referrer indicate whether it appeared in the top 20 in any of the last 8 weeks before this one (yes/no, last seen date if any).
Highlight referrers that are brand new in the top 20. For each new referrer flag: probable spam vs probable opportunity (use signal heuristics: domain TLD, bounce rate, session duration, % bot).
Output: 1 table + a "new this week" subsection.
SEAL-018 — Pageview-to-conversion ratio drift
Using Sealmetrics MCP for site {site_id}, compute the daily ratio of pageviews to macro conversions for the last 30 days.
Detect days where the ratio changed by more than 15% versus the trailing 7-day average. For each detected day, list:
- Pageviews, conversions, ratio, % delta
- Top 5 channels by entrances that day vs the prior day
- Whether a deploy or campaign launch is plausibly responsible (if I tell you {deploy_log} or campaign launches, factor that in)
End with a verdict per flagged day: tracking incident, real performance shift, or campaign-driven.
SEAL-019 — UTM hygiene audit
For site {site_id}, query Sealmetrics MCP for the last 30 days.
Compute the % of paid traffic (utm_medium in cpc, paid, ppc, paidsocial, paid_search, display) that has at least one of:
- Missing utm_source
- Missing utm_campaign
- Missing utm_medium
- Lowercased "google" / "facebook" / "instagram" written inconsistently
Group by destination platform (Google, Meta, Microsoft, etc.) and by campaign. List the top 10 problematic campaigns with sample URLs.
End with a recommended fix list (which campaigns/templates to update first).
SEAL-020 — Suspicious sessions by device and country
Using Sealmetrics MCP for site {site_id}, for the last 14 days:
1. Pull all sessions classified as suspicious (bot or agent_suspected).
2. Break them down by device type and by country.
3. Highlight country-device combinations that account for more than 5% of suspicious sessions and that historically have had less than 1% of legitimate traffic for this site.
Output: a heatmap-style table country × device. End with a 3-line proposal: which combinations to add to the blocklist, which to rate-limit, which to investigate further.
See also
Was this page helpful?