The reporting problem teams actually have
Most reporting time is not spent writing. It is spent logging into Google Ads, Meta Ads Manager, GA4, and Search Console, exporting the same views, pasting them into a spreadsheet, reformatting the numbers, and only then asking what they mean. That work produces no insight. It is data assembly, and it is why the analysis at the end is usually thin: by the time the numbers are in one place, the hour is gone.
Connecting the platforms to Claude through MCP Ads removes the assembly step. Claude pulls live data from every connected account in a single conversation, so the person doing the report starts at interpretation instead of at export. That is the entire gain. Nothing about the judgment changes; the time it takes to reach the point where judgment is needed does.
What a weekly report should do
A weekly report should answer three things quickly: what changed, why it changed, and what the team is doing next. Most reports handle the first part, often badly. They list metrics without context, and the reader is left to do the analysis themselves. A report that says spend was $4,200 and conversions were 61 has told an executive nothing. A report that says conversions fell 18% because the top search campaign lost impression share to a budget cap on Thursday, and the cap is being raised, has told them everything they need.
The audience decides the depth. Operators want the diagnostic: which campaign, which search terms, which creative. Clients want the outcome and the plan. Executives want the headline number, the direction, and the decision they are being asked to make. The mistake is building three reports. The right approach is one diagnostic and two derived summaries, so the numbers never disagree between versions.
The operating system: four steps, every week
The routine that holds up over months is the same four steps in the same order. Pull: one prompt gathers the last seven days and the prior seven from every connected platform. Diagnose: a second prompt asks what moved and why, with the evidence. Write: Claude drafts the operator report, then the client and executive versions from it. Review and send: a person checks the numbers against the platform, adds anything the data cannot see (a promotion, a site outage, a client decision), and sends.
The first three steps take a few minutes once the prompts are standardized. The fourth is the one to protect. Reporting is a trust product; a single wrong number sent to a client costs more than every hour the automation saved.
- Pull: last 7 days vs. prior 7 days, every connected platform, same metrics every week
- Diagnose: what changed, why, and what evidence supports the explanation
- Write: operator diagnostic first, then client summary and executive summary derived from it
- Review and send: verify numbers in the platform, add offline context, approve delivery
The structure that works
The most reliable format is simple: headline numbers, wins, concerns, anomalies, and next actions. It reads in under two minutes, it maps directly onto the three questions above, and it is easy to compare week to week because the sections never move. Claude can fill it from live data far faster than a person can assemble it, and because the structure is fixed, the reader learns where to look.
- Headline: spend, results, cost per result, and the week-over-week change for each
- Wins from the last 7 days worth keeping or scaling
- Concerns that need attention this week, each with a likely cause
- Anomalies: unexpected movements and what probably drove them
- Next actions: three specific things the team will do before the next report
The core prompt
The prompt that produces a good report gives Claude a date range, a scope, the metrics that matter for this account, and the format to return. A version that works well for a lead-generation account: "Summarize marketing performance for the last 7 days versus the prior 7 days across Google Ads, Meta Ads, GA4, and Search Console. Lead with total spend, leads, and cost per lead by platform. Then return: wins, concerns with likely causes, anomalies, and three specific actions for next week. Flag any campaign with spend and zero conversions, or with cost per lead up more than 30% week over week."
Run it the same way every week. The value is not in the wording; it is in comparability. When the structure and the thresholds are identical from one Monday to the next, "CPL is up" means the same thing it meant last week, and the trend across a quarter becomes visible without anyone building a chart. Save the prompt as a project instruction or a snippet so it does not drift.
The client version
Agency teams need a second version for the client: the same numbers, less diagnosis, more plan. Ask for it in the same thread so it inherits the data: "Now write a client-facing summary for [client name] from the report above. Open with a two-sentence overview of the week, then spend and results by platform, then what we changed and why, then what we will do next week. Keep it under 250 words and do not include internal notes."
Structure it for the client type. E-commerce clients care about revenue, ROAS, and cost per purchase. Lead-generation clients care about lead volume, cost per lead, and the share of leads that qualified. Brand clients care about reach, frequency, and engagement. Tell Claude the client type and the target metric once, in the project instructions, and every report will be framed against it. If you manage many accounts the saving compounds: the assembly step that used to cost most of an hour per client is now a prompt and a review.
- E-commerce: revenue, ROAS, cost per purchase, top products, blended cross-channel ROAS from GA4
- Lead generation: leads, cost per lead, qualified rate if the CRM is connected, top campaigns by CPL
- Brand: reach, frequency, engagement rate, video completion, creative winners
- Every client: what changed, why, what we will do — the plan is the part they remember
The executive version
Executives do not care about CTR, Quality Score, or frequency. Those are operational metrics, and they belong in the operator report. An executive summary answers three questions: how much did we spend, what business result did it produce, and how did that compare to the last period — plus the one decision, if any, that needs their approval. Everything else is noise to that reader.
Translate before you send. Replace impressions with reach only when reach is the goal. Replace CTR with conversion rate and cost per lead. Replace average CPC with cost per acquisition or return on ad spend. For B2B accounts with a long sales cycle, add lead-to-opportunity rate and pipeline generated if the CRM is connected; raw lead volume is a weak proxy for business impact when lead quality varies. The prompt: "Prepare an executive summary of paid media for the last 7 days. Total spend, total results, cost per result, week-over-week change in each, one win, one concern, and one recommended decision. Under 150 words, no platform jargon."
Prepare the answers to the questions executives always ask. "Why did costs go up?" needs a one-sentence root cause — creative fatigue, a budget change, a competitor, seasonality — ready before the meeting. "How do we compare to last quarter?" needs the quarter-over-quarter figures framed as a trend, not a single number. Executives think in quarters, not weeks; a weekly report that also carries the running quarter-to-date line answers the question before it is asked.
What a finished weekly report looks like
The following is an illustrative example, written to show the shape and level of detail, not a report about any real account. Every number is invented. Use it as a template for what the operator version should read like once the prompt is tuned.
- Headline: spend $6,420 (+4% WoW) across Google Ads $4,100 and Meta $2,320; 84 leads (-9% WoW); blended cost per lead $76 (+14% WoW)
- Google Ads: spend $4,100, 57 leads, CPL $72 (+11%). Non-brand search lost 22% impression share on Thursday and Friday after the campaign hit its daily budget cap; brand search steady
- Meta Ads: spend $2,320, 27 leads, CPL $86 (+21%). The lead-form campaign top creative reached frequency 4.6 with CTR down 31% from its peak two weeks ago
- GA4: paid sessions -6%, landing-page conversion rate on /demo flat at 3.1%; no site-side change detected
- Search Console: clicks +3%, one page moved from position 11 to 8 for its main query
- Win: the new exact-match ad group launched last week produced 9 leads at $41 CPL — below the account average — and is worth more budget
- Concern: Meta creative fatigue is the main driver of the CPL rise; the two fresh variants approved last week are still in review
- Anomaly: 14 conversions on Tuesday came from one search term unrelated to the service; flagged as a negative-keyword candidate
- What changed this week: raised the non-brand search daily budget 20%; added six negative keywords; paused one Meta ad set with 11 days of spend and no leads
- What we will do next week: launch the two new Meta creatives and cap frequency at 3; move 15% of Meta budget to the exact-match Google ad group; review the Tuesday search term and add the negative if it repeats
Scaling it across accounts
The routine is the same for one account and for thirty; only the loop changes. Agencies typically run the core prompt per client from a project that holds that client instructions and targets, then a portfolio prompt across all of them: "Across every connected account, list any campaign with spend and zero conversions in the last 7 days, and any account whose cost per result rose more than 30% week over week." That single pass is the Monday triage; the per-client reports follow from it.
Keep the structure identical across clients even when the metrics differ. It lets a new account manager read any report without learning a new format, and it makes the portfolio view possible: if every report carries the same headline line, the numbers can be lifted straight into a management summary.
Guardrails and review
Three habits keep the automation honest. Verify the headline numbers against the platform before sending — not every number, the three or four that would embarrass you if wrong. Add the context the data cannot see: a promotion that ran, a landing page that was down for an hour, a client decision that changed the plan. And keep write actions separate from reporting: a report should describe what will change, and the change should be a second, approved step, whether that is a person in the platform or the same assistant with confirmation on.
Watch for the usual reconciliation traps. Google Ads and GA4 will not match on conversions because of attribution windows, conversion lag, and time zones; say which source the report uses and stick to it. Meta reported conversions overlap with Google on the same orders; use GA4 or the CRM for the blended number and the platform figures for the per-channel diagnosis.
Bottom line
The goal of weekly reporting is operational clarity, not a document. When Claude starts from the live data, the report becomes a short explanation with a plan attached, and the client and executive versions fall out of the same thread instead of being rebuilt by hand. Teams that standardize this routine spend their Monday on the recommendations rather than the spreadsheet. See Google Ads, GA4, and Search Console for the platform-specific setup behind each piece, and the death of the dashboard for why this shift is happening.
