Performance insights

Compare two periods, find the campaigns and segments behind the change, and check the conversions against GA4 before calling anything a drop.

When to use

A question about a change in spend, clicks, CPC, conversions, CPL or ROAS. For a full account review use ads-audit; for the Monday routine use weekly-check.

Inputs

• account: profile alias.
• Periods: the user's, or last 7 days vs the 7 days before. Use dateFrom / dateTo (YYYY-MM-DD) for both periods so they are exactly comparable.
• The metric they care about (default: conversions and cost per conversion).

Steps

Call account_profile first (currency, linked GA4).

1. Totals. get_account_summary for each period. State the change in cost, clicks, CTR, CPC, conversions and cost per conversion as numbers and percentages.
2. Where. get_campaign_performance for each period. Rank campaigns by their contribution to the change (absolute difference in the metric, not percent).
3. Why, inside the top 1–3 campaigns. Use get_performance_by_date for the day the change started, then segment with run_gaql_query, for example:
   • device: SELECT segments.device, metrics.cost_micros, metrics.conversions FROM campaign WHERE campaign.id = <id> AND segments.date BETWEEN '<from>' AND '<to>'
   • network: same query with segments.ad_network_type
   • search terms: get_search_terms for the campaign and both periods (new expensive terms are a common cause)
   • auction pressure: metrics.search_impression_share, metrics.search_budget_lost_impression_share, metrics.search_rank_lost_impression_share on campaign
   • recent changes: changes_list (changes made through this package) and, for Demand Gen, guide("demand-gen") on reading the first weeks.
4. Reconcile. If GA4 is linked, ga4_conversions and ga4_traffic_sources for both periods. If GA4 lead events held steady while Ads conversions fell, the problem is measurement, not demand: say so and point to guide("measurement").
5. Read CPC/CTR/CPL movements with guide("benchmarks") in mind: they are ranges, not targets.

Output

# <alias>: <period A> vs <period B>

**What changed:** <one sentence with the main number>
| Metric | A | B | Δ |
|---|---|---|---|

**Why:** 2–4 bullets, each with evidence (campaign, segment, number).

**Do next:** 2–3 actions, each naming the tool that would make the change.

Keep it under one screen. Name the evidence; say "unclear" when the data doesn't explain the change.

Never

• Never apply a change from this skill. Suggest it; if the user says yes, preview it first (no confirm), then apply after a second yes.
• If the profile is read-only (no write tools), say that applying needs npx google-marketing-agent accounts set <alias> writes true and an MCP client restart. Never edit config files.
• Never compare periods of different length or a period that includes today against full days.
• Never report "0 conversions" as "no leads" without the GA4 check in step 4.