Six exports, a spreadsheet, and a narrative written at 11pm: that is too much manual work for a monthly PPC report. I use the five prompts below with groas connected to ChatGPT or Claude to turn live Google, Microsoft and Meta Ads data into a client-ready report without building the deck by hand.
Setup card: connect once, scope every report
When to use it: before the first reporting session for a client. Connect groas through MCP or an API key, then paste this card into the chat before running the five prompts.
REPORTING CONTEXT
Client: [client_name]
Google Ads account IDs: [google_account_ids]
Microsoft Ads account IDs: [microsoft_account_ids]
Meta Ads account IDs: [meta_account_ids]
Reporting month: [month]
Comparison month: [comparison_month]
Account time zone: [time_zone]
Reporting currency: [currency]
Monthly budget: [monthly_budget]
Budget by platform: [platform_budgets]
Primary business goal: [primary_goal]
Target metric and value: [target_metric_and_value]
Conversion definition: [conversion_definition]
Conversion source: [conversion_source]
Use only this client's named accounts. Do not substitute another account
if an ID is missing or inaccessible. Ask me to resolve it.
This is a reporting session. Read data only. Do not change campaigns,
bids, budgets, targeting, ads, account settings or website content.
Record the retrieval timestamp and reporting cutoff. Use the same date
range, currency and metric definitions throughout the report.
If required data is unavailable, identify the gap. Do not estimate it
unless I explicitly request an estimate, and label any estimate clearly.
The adjustment that matters: replace the account IDs, not just the client name. I want the reporting boundary written down, not inferred from whichever account happens to be open. Keep write access off in reporting sessions wherever the connection’s permissions allow it; a sentence in a prompt is an instruction, not an access-control system.
Connection card: what groas supplies
When to use it: when setting up the reporting workflow for your team. The prompts need connected account data; a plain chat only has the material you provide.
- Connection: connect groas to ChatGPT or Claude through MCP or an API key, then connect the client’s ad accounts.
- Reporting inputs: live Google Ads, Microsoft Ads and Meta Ads data, plus the log of actions groas took and their recorded reasons.
- Action boundary: autopilot stays off until you switch it on. Changes groas makes are logged with their reasons and can be undone in one step.
- Team cost: groas Unlimited is $19 a month for the whole team, including unlimited teammates, ad accounts and websites, every feature and all the data. Adding clients does not add a per-client fee.
The adjustment that matters: check change-log coverage before promising a complete account history. The log of groas actions is not automatically a record of every change somebody made elsewhere. A useful report names its evidence rather than dressing a partial history up as omniscience.
Separation card: keep one client’s report out of another’s account
When to use it: when creating a saved reporting chat or project. I keep the client boundary beside the reporting instructions, not buried in a general agency brief.
CLIENT SEPARATION CHECK
This report belongs only to [client_name].
Use only the account IDs in the reporting context.
Before retrieving data:
- Confirm the account names and IDs match that context.
- Exclude every other client accessible through the connection.
- Stop if an account cannot be matched confidently.
Before drafting:
- Confirm every table row belongs to this client.
- Exclude examples, targets and figures from other clients' reports.
- Do not use another client's data to fill a missing value.
The adjustment that matters: use a separate reporting context for each client. One client’s totals drifting into another client’s summary is not a formatting problem. It is a very short route to an uncomfortable call.
groas cutaway diagram placeholder
groas image placeholder for per-client separation
The report template: budget, results, decisions
When to use it: paste this after the setup cards and before prompt 1. It is the output structure for the five-prompt sequence, not another document to rebuild each month.
## [client_name] — PPC performance, [month]
_Spend through [cutoff_date]. Data retrieved [refresh_timestamp]._
_Recent conversions may still arrive after this cutoff._
### 1. Executive summary
Spend: [total_spend] of [monthly_budget] budget.
Outcome: [conversion_count] [conversion_definition] from [conversion_source].
CPA: [cpa] against [cpa_target], where comparable.
- Win: [win_with_campaign_and_number]
- Concern: [concern_with_campaign_and_number]
- Next action: [action_linked_to_the_evidence]
### 2. Budget pacing
| Platform | Spend | Budget | Status |
|---|---:|---:|---|
| Google Ads | [spend] | [budget] | [status] |
| Microsoft Ads | [spend] | [budget] | [status] |
| Meta Ads | [spend] | [budget] | [status] |
[one_line_explanation_of_budget_variance]
### 3. Results by channel
Conversion source and definition: [source_and_definition]
| Channel / campaign | Spend | Conversions | CPA | CPA vs target | CPA vs previous month |
|---|---:|---:|---:|---:|---:|
| [channel_and_campaign] | [spend] | [count] | [cpa] | [variance] | [change] |
### 4. Wins and losses
**Win:** [campaign_or_change_and_supporting_number]
**Loss:** [campaign_or_change_and_supporting_number]
### 5. What changed and why
- [date] — [change] — [recorded_reason]
### 6. Next month's three priorities
1. [priority_problem_action_and_measure]
2. [priority_problem_action_and_measure]
3. [priority_problem_action_and_measure]
### Measurement notes
[attribution_limits_missing_data_and_comparison_caveats]
The adjustment that matters: define the conversion before filling the template. A platform form fill and a CRM-qualified opportunity are not interchangeable, as the ClientSignal reporting template makes clear. Keep keyword-level and Quality Score detail in the diagnostic work; campaign level is the client copy.
Measurement card: settle the conversion ledger first
When to use it: before asking for the summary. This prevents the first prompt from confidently adding together numbers that should stay separate.
MEASUREMENT RULES
Use [conversion_source] for the primary outcome defined as
[conversion_definition]. Compare it only with a target set against
that same outcome.
Do not add platform-reported conversions together and describe the sum
as unique customers, orders or leads. Platforms can credit overlapping
outcomes to their own touchpoints.
If an independent first-party total is available for this reporting
period, name its source before using it for cross-channel outcomes or
blended CPA. Confirm that its scope matches the spend being reported.
If it is unavailable, show platform-reported conversions separately.
Do not manufacture a deduplicated total or blended CPA.
Record the attribution settings available from each source. Flag
incompatible definitions or windows before making comparisons.
State the reporting cutoff and retrieval timestamp. Note that recent
conversions can arrive later; do not present an immature period as final.
The adjustment that matters: name one ledger for the headline outcome, not whichever platform offers the nicest number. If the CRM shows 82 qualified leads and Google Ads shows 110 form fills, show both with their labels. They answer different questions. Attribution differences are a reporting constraint, not permission to pick a winner; this platform-attribution example illustrates the overlap problem.
Prompt 1: write the five-line executive summary
When to use it: first in the drafting sequence, after the context, template and measurement rules. This is the section I assume the client will actually finish.
Through groas, write the executive summary for [client_name] for [month].
Follow the reporting context and measurement rules above.
Use five lines:
1. Total spend across Google Ads, Microsoft Ads and Meta Ads vs budget.
2. Primary outcome and CPA vs the agreed target, naming the source.
3. One win, naming the campaign and supporting number.
4. One concern, naming the campaign and supporting number.
5. One proposed action for next month, tied to the evidence.
If a unified outcome or comparable CPA is unavailable, say so on line 2
and keep platform outcomes separate elsewhere in the report.
Write plainly. Do not invent data, targets or explanations.
Mark missing required figures as [unverified].
The adjustment that matters: put the agreed target in the reporting context, including its conversion definition. Without that anchor, the model can write “performance improved” without answering whether the account did its job. I want the comparison to the client’s goal, not a congratulatory adjective.
Prompt 2: replace screenshots with campaign tables
When to use it: immediately after the summary. This supplies the budget, results, and wins-and-losses sections without making the client decode three platform interfaces.
Through groas, build the budget and results sections for [client_name]
for [month], compared with [comparison_month].
First, show spend vs the supplied budget for each platform and in total.
Explain any budget variance in one line. If a budget is missing, say so.
Then build a Google Ads and Microsoft Ads campaign table:
| Channel / campaign | Spend | Conversions | CPA | CPA vs target | CPA vs previous month |
Add a separate Meta Ads table with the same columns.
Name the conversion source and definition for each table.
Sort campaigns by spend, highest first. Include campaign rows with
spend above [minimum_campaign_spend]. Keep platform spend totals complete
and show excluded campaign spend separately so totals reconcile.
Round currency to the nearest whole unit and percentages to one decimal
place for display. Calculate from unrounded source values.
Name one win and one loss supported by these tables. Do not claim a
change caused a result merely because their dates coincide.
The adjustment that matters: set the minimum-spend filter to the account’s size. On a $5k monthly account, a $50 filter can keep the table useful; on a $50k account, $50 rows are noise. Filter the presentation, not the money that goes into the totals.
Prompt 3: show the changes worth discussing
When to use it: for the proof-of-work section. Results show what happened; the change log shows the decisions behind the account work.
Through groas, retrieve the available change history for [client_name]
for [month] across Google Ads, Microsoft Ads and Meta Ads.
State what the history covers. Distinguish groas actions from any other
available account changes. Do not imply the record is complete if it is not.
Select the 5–10 changes that matter most, sorted by date:
| Date | What changed | Recorded reason | Expected effect |
Consider bid adjustments, budget shifts, keyword and negative-keyword
changes, paused or launched campaigns, and ad copy updates.
Use plain English. If no reason was recorded, write “Reason not recorded.”
Label expected effects as expectations, not measured results.
Group repetitive routine adjustments when their recorded reasons support
that grouping. Do not invent a shared motive to shorten the table.
The adjustment that matters: cap the rows, not the honesty. If the log contains forty small bid changes with the same recorded purpose, summarize them together and save the remaining rows for decisions worth discussing. A wall of activity is not an explanation. The Pace Ads reporting guide puts that work in its proper place: alongside results and next steps.
Prompt 4: turn the evidence into three priorities
When to use it: after the tables and change log. This closes the report with proposed decisions rather than another paragraph about monitoring performance.
Through groas, use [client_name]'s [month] performance and the available
change log to draft next month's three priorities.
For each priority, provide:
1. The priority in one sentence.
2. The problem or opportunity, naming the campaign and supporting number.
3. The proposed change and the metric that will show whether it worked.
Order by expected impact on [primary_goal]. Explain that ordering without
inventing forecasts or promising a result.
If the previous three months' priorities are available, avoid repeating
them unless the evidence shows the problem remains open. If they are not
available, say that the historical check could not be completed.
Now assemble the full report using the supplied template and outputs
from prompts 1–4. Keep measurement caveats beside the affected numbers.
Do not fill gaps with guesses. These are proposals: execute no changes.
The adjustment that matters: specify the business outcome the priorities should serve. “More conversions” is a wish. “Reduce CPA below $85 on the brand campaigns” gives the model a defined problem, provided that is the client’s actual target. Keep the recommendation tied to the supplied goal, not a goal the model finds convenient.
For account work beyond reporting, the 25-prompt library covers prompts this sequence deliberately leaves out.
Prompt 5: verify the report before it leaves
When to use it: last, against the assembled report. Live access does not make a model infallible, and asking it to check its work does not remove my responsibility to review the result.
Before I send this report, verify it against the connected source data.
Do not change any account settings.
1. Retrieve spend totals again for the same client, accounts and dates.
Recalculate the combined spend and compare it with the report.
Flag any difference over 1%, and explain smaller differences caused
by display rounding where identifiable.
2. Re-pull [spot_check_campaign_name]'s spend and conversions from the
source, not from the draft. Confirm the campaign row, date range,
conversion definition and calculated CPA match.
3. Recalculate one percentage comparison in the summary from its source
values. Show the calculation. If the summary has none, use a table row.
4. Check that platform-reported conversions were not added together and
presented as unique outcomes. Confirm the source and scope of any
blended CPA and the definition behind its target.
5. Check that filtered campaign tables reconcile to complete platform
totals, including excluded spend.
6. Flag every report figure that cannot be traced to source data. Replace
an unsupported figure with [unverified], not a plausible estimate.
7. Confirm the report names the cutoff, retrieval timestamp, missing data
and any comparison limitations that affect its conclusions.
Return the verification findings first, then the corrected full report.
List unresolved items explicitly. Do not call unresolved figures verified.
The adjustment that matters: name the campaign you already suspect, not whichever row is easiest to confirm. I review the returned source values and resolve every [unverified] before sending. This is the artifact I would keep if I had to delete the rest: the drafting prompts save the manual assembly, but the verification pass is the one aimed at keeping a bad number out of the client’s inbox.

