Roshan (@roshanp), whose team publishes about 1,000 Meta ads a month for booked-call offers, shared 50 tactics built around an AI agent (Grok Bot). The agent work: build an ad scripting guide from past winners and losers; break the top 3-4 ads into angle, messaging, offer framing, who it calls out and proof; pull the exact words buyers use on sales calls each month; diagnose every ad that stops working (no spend = weak angle, low hook rate = weak first 3 seconds, high hook but low hold = weak body, high hold but low CTR = weak CTA or offer, clicks but no conversions = ad and page don't match); and every Monday pull CPM, unique outbound CTR, hook rate, show rate, qualified rate and close rate for every ad, then pick the one metric to improve that week. His numbers are self-reported for lead-gen accounts (for example a 5-10% average hit rate, with a winner defined as an ad that spends 5x target cost per result and stays inside KPI); for DTC, swap calls for purchases. The post links an article with the 12 prompts behind the system. The article's setup step: connect Meta Ads read-only, load a business-context document (offer, audience, funnel, targets, provable results only), and have the agent send any budget, status or targeting change to you as a draft with the current setting, the new one and why.
Prompt
Pull the transcripts of my [10] ads with the lowest cost per qualified call and the [10] with the highest, and write me a practical ad scripting guide.