What you need to know
- Every report should identify the workspace, account IDs, period, currencies, and metric definitions before interpreting results.
- Platform-attributed conversions should not be added together and presented as unique customers.
- Ask for evidence tables and missing-data notes before requesting recommendations.
Define a reporting contract
A useful weekly report answers three questions: what happened, what explains it, and what needs a decision. The data must be consistent enough for another person to repeat the check. Before using the prompts below, replace the workspace placeholder and choose a reporting period supported by the connected tools. Ask the assistant to return the actual period used.
These prompts are designed for an authorized MCP connection such as PlugMind. They do not enable background monitoring or schedule recurring reports by themselves. Run them when you want a review, and use a separate scheduling system only if your setup explicitly supports one.
Build a brief containing the business goal, main conversion event, account currencies, time zones, relevant promotions, and known tracking changes. Keep the brief with the report so a future reader can distinguish a measurement change from a change in advertising performance.
1. Confirm accounts before pulling results
Use this whenever you switch clients or reconnect a provider. An assistant should resolve an ambiguous workspace before querying data. The intended output is a compact account map, not a report across every client the user happens to be able to access.
2. Produce a table before a summary
Check raw totals against the native platforms. Do not treat a missing value as zero. If the date boundaries differ, resolve them before comparison. A result should carry its metric name: “leads” is more useful than an unexplained “results” column when campaigns have different objectives.
3. Explain a change without inventing a cause
When the prior value is zero, a percentage change is undefined; explain the absolute change instead. A smaller conversion count after a budget increase is an observation. A claim that the budget increase caused the drop needs more evidence. Ask for campaign-level details before accepting an account-wide explanation.
4. Investigate Google Search intent
The search terms report has privacy-related omissions, so a list of returned queries is not a complete search history. Use it to find specific questions and exclusion candidates. Do not infer competitor budgets or customer identities from the report.
5. Investigate Meta delivery
Keep this investigation separate from the Search query analysis. Search intent and social delivery are different forms of evidence. A consistent report can contain both without forcing them into the same explanation or pretending they use identical attribution methods.
6. Turn the report into decisions
If an assistant cannot retrieve an input needed for the action, it should request that input or describe a manual step. A report recommendation is not an execution receipt. Preserve the actual tool result separately when you later approve a supported change.
Avoid three misleading cross-platform totals
| Tempting shortcut | Why it misleads | Better report |
|---|---|---|
| Add all platform conversions | The same customer can be attributed by more than one platform | Show platform-attributed results separately; use a deduplicated business source for unique customers. |
| Add spend across currencies | A dollar and a euro are not interchangeable | Separate currencies or document an exchange-rate source and date. |
| Average campaign CPA or ROAS | Small campaigns receive the same weight as large ones | Calculate ratios from compatible summed components. |
Illustrative example: if Google reports 12 purchases and Meta reports 9, that does not establish 21 unique purchases. Some may overlap and some may fall under different attribution windows. A CRM or commerce source with a defined deduplication method can answer the unique-order question; the two platform reports alone cannot.
Archive the exact prompts, report period, returned account IDs, and evidence table alongside the summary. This turns next week’s review into a comparison of defined measurements, rather than a fresh conversation with different assumptions.
Frequently asked questions
Do these prompts change my campaigns?
They explicitly request reporting and unexecuted proposals. Review the actual tool calls and permissions as well; prompt wording is not a substitute for server-side access controls.
Can I run a report if only one platform is connected?
Yes. Report the connected platform and label the other as unavailable. Do not invent missing results.
Are these scheduled reporting automations?
No. They are prompts to run during a review. Recurring execution requires a separately configured and supported scheduler.
Sources and verification
Platform references are listed below. PlugMind tool descriptions were checked against its implementation. Example prompts are suggested workflows, not claims of results from a live advertising account.