Twelve prompts for the work a PME hands to agents. Fill in the highlighted parts, paste them into the assistant or agent you use, and judge the answer against the checklist under each one.
Compare purchases recorded in the CRM with purchases each ad platform reports for {date range}. Show a table: platform, platform count, CRM count, gap in %. For each gap above 10%, list the most likely causes in order of likelihood and the check that would confirm each one. Do not change any settings.Review our conversion events for {platform}. For each event list: name, where it fires (browser, server or both), whether it carries an event_id for deduplication, value and currency, and consent handling. Flag anything that could double-count or drop conversions, and say how you would test it.Report the match rate or event match quality for our server-side conversions on {platform} for the last 4 weeks, by week. Explain which customer fields we send, which are missing, and what each missing field would likely add. Only suggest fields we are allowed to send under our consent setup.For each campaign in {account}, estimate the cost per purchase of the last 20% of spend over {date range}, not the average. Compare it with our target CPA of {target}. Recommend where spend could move, but do not move anything: I approve every change above {limit}.Build last week’s ledger: total marketing spend, revenue, MER, new customers and contribution margin, with the source of every number named. Then list what each platform claimed in conversions and how far that is from the CRM. Keep the platform claims in a separate column from the business numbers.
Propose a geo holdout to measure the incremental effect of {channel}. Give: the regions in each group and why they are comparable, the test length, the minimum effect we could detect at our current volume, the budget at risk, and the decision we will make for each possible result. Flag anything that would make the result unreliable.CPA on {campaign} changed by {change} on {date}. List every change made in the 7 days before, by us, by the platform or by the site: budgets, bids, creatives, audiences, tracking and landing pages. Rank the likely causes and say which data would separate them. Do not revert anything yet.Group last month’s ads by hook rather than by ad. For each hook give spend, impressions, CTR, conversions and CPA, and whether it reached at least {n} conversions. Do not call a winner for any hook below that threshold; mark it as inconclusive instead.Our CPA is about {cpa} and each variant needs about {n} conversions for a verdict. With a test budget of {budget}, how many variants can we test properly, and for how long? If the answer is fewer than we planned, suggest which variants to drop and why.Read the ad copy and the landing page at {url}. List every promise the ad makes and whether the page repeats it above the fold. Flag any claim the page does not support, and any step between the click and the conversion that the ad did not prepare the user for.Draft the agenda for this week’s performance review: the three numbers that moved most and why, every change the agents made with its result, every proposal waiting for approval, and one question we cannot answer yet. Keep it to one page.
Write the operating limits for a {type} agent on {account}: what it may change without asking, the maximum size of each change, what always needs my approval, how every change is logged, and how to roll it back. Use plain sentences I can sign off.The library is written by the people who use it. Send a prompt that has worked for you on a real account. We check it against the three rules above and publish it with your name and link.
Four short courses, each ending in a project you can show. Free and open to everyone.