Update to AdsGPT

{
  "description": "Review how the user's ads perform on the platforms connected to AdsGPT and improve them, such as reports of a period, spend that does not convert, negative keywords, pausing what wastes money, moving budget to what works and pacing. Use it whenever the user asks how their ads are doing, what to improve or how to spend better.",
  "included_files": [],
  "name": "adsgpt-optimize",
  "skill_md_contents": "---\nname: adsgpt-optimize\ndescription: Review how the user's ads perform on the platforms connected to AdsGPT and improve them, such as reports of a period, spend that does not convert, negative keywords, pausing what wastes money, moving budget to what works and pacing. Use it whenever the user asks how their ads are doing, what to improve or how to spend better.\n---\n\n# Review and optimize\n\nEach platform has its own tools for what follows, which its skill names: its reports, how to pause,\nhow to change a budget and its limits. Read the skill of each platform before using its tools.\n\n## Review\n\n- **The numbers come from the reports.** Use each platform's reports, as its skill describes\n  them: what each reports, by what it breaks the figures down and in which currency. A platform\n  that reports no spend or no conversions has none to compare: say so instead of filling the gap.\n  When no report covers what the user asks, say so and where to see it on the platform.\n- **The period.** Default to the last 30 days unless the user names one, and say which period each\n  figure covers. Compare like with like: a period against the same number of days before it. Today\n  is partial, and a few days are not a trend.\n- **Measurement first.** Read the conversion actions before any figure about conversions: when the\n  account measures none, its cost per conversion means nothing. Say so before showing it, offer to\n  create the missing one where the platform's tools can, and check what a campaign bids for when\n  its conversions are not the business's.\n- **The platform's own suggestions.** Read the recommendations the platform gives, as its skill\n  says, and offer, after the numbers, the ones that fit: the improvement each platform expects is\n  its estimate, not a result. Apply the chosen ones and dismiss those the user rejects where the\n  platform's tools can.\n- **Lead with the answer.** The biggest problem and the biggest opportunity, each with its numbers,\n  and one recommendation. Do not average platforms with very different volumes without saying so.\n- **Offer, do not act.** A review changes nothing: propose the change and wait for the user's yes.\n\n## Optimize\n\nBefore cutting anything:\n\n- **Check the measurement.** A campaign that seems not to convert may not be measured: pausing it\n  would stop one that works.\n- **Check the volume.** A few clicks without conversions are not evidence: compare the spend with\n  what a conversion costs in the account before calling something a loser.\n\nThen fix it, in this order:\n\n1. **Exclude before pausing.** Negative keywords stop the searches that waste money, and excluded\n   placements the sites and apps that do, while the campaign keeps running. Add all the ones of a\n   campaign in one change, and put the terms that waste money in several campaigns in a shared\n   list where the platform has them.\n2. **Pause the narrowest thing** that solves the problem: an ad, then its group, then the campaign,\n   with the levels each platform has.\n3. **Move the budget** to what converts at the best cost, not to what has the most volume. Prefer\n   steps to doubling a budget at once: a large change makes automated bidding learn again. Keep\n   each budget daily or total as the campaign has it, within the limits the platform's skill gives.\n\nNever remove when pausing does: removing loses the history. Name what would be lost and ask.\n\n## Pacing\n\nCompare the spend so far with the days elapsed of the period. A campaign that spends less than its\nbudget may be limited by its bids, its targeting or its audience, not by its budget: check that\nbefore raising it. A platform may spend more than the daily budget on some days, as its skill says:\na single day above the budget is not a mistake.\n\n## After each change\n\nReport what the tool says it changed, and what to expect: automated bidding takes time to learn\nagain, and the next days of data say whether the change worked.\n"
}