Set up a Meta App campaign with AI

Where to run this

Your approved ChatGPT, Claude or Gemini workspace, files uploaded first.

Upload first

• ☐Approved campaign brief and primary KPI
• ☐Budget, markets, languages
• ☐App Store and Google Play URLs
• ☐SDK and MMP integration status
• ☐Conversion events and historical performance, if available
• ☐Creative assets

Steps

1. 01

   Pick your starting point before you build a blueprint

   Do this

   New app, new market, no suitable historical campaign: build fresh. An existing app campaign with credible performance history: audit it, don't redesign a winner without evidence. A proven campaign that needs to launch in a new market: clone and adapt it.

2. 1a

   New campaign: generate the blueprint

   You are designing a complete Meta App campaign. Using only the approved materials provided, generate: a campaign blueprint, event hierarchy (which post-install events exist and which is the optimisation target), an app-event optimisation framework, audience framework, retargeting framework, and creative framework. Use only information supplied. Do not invent audiences, offers, claims, promotions or budgets. Finish with READY TO BUILD or NOT READY.

   Do this

   As of 2026, Advantage+ App Ads is Meta's default structure for apps generating more than 50 installs a day, and the platform now optimises for the deepest post-install event you have sufficient volume for, not raw installs. Under 50 events a week, optimise for App Installs; once you cross that, upgrade to a deeper App Event (registration, tutorial completion, first purchase).

3. 1b

   Existing winning campaign: audit it, don't just redesign it

   Review the attached Meta App campaign and the supplied performance evidence. First identify what appears to be working, what should remain unchanged, and what is not supported by sufficient evidence, checking specifically whether reported install volume is actually being measured on a SKAdNetwork 4 / AdAttributionKit-aware setup, campaigns still running on pre-SKAN-4 measurement have been shown to systematically underreport performance. Then identify: event-optimisation upgrade opportunities, creative refresh needs, and scaling opportunities. Use only: KEEP, TEST, EXPAND, REMOVE, HUMAN REVIEW, NOT EVIDENCED. Do not redesign a successful structure without evidence. Do not invent performance results. Do not approve the campaign. Finish with STRUCTURE CAN BE REUSED or STRUCTURE REQUIRES REVISION.

4. 1c

   Cloning into a new market: adapt, don't rebuild

   Use the attached high-performing Meta App campaign as the source blueprint. Preserve: campaign structure, event hierarchy, audience framework, and creative framework. Review whether changes are required for: market, language, local app store listing status, local MMP/SDK regulatory constraints, and creative availability. Classify every recommendation as: KEEP, LOCALISE, REMOVE, ADD, HUMAN REVIEW, NOT EVIDENCED. Do not redesign the campaign unless the supplied local-market evidence requires it. Do not assume a winning structure will perform identically in another market. Do not approve the campaign. Finish with READY TO CREATE LOCAL BLUEPRINT or LOCAL EVIDENCE REQUIRED.

5. 02

   Build measurement and attribution before anything else

   Review my Meta App measurement setup against the attached blueprint. Validate: Meta SDK integration, Meta Events Manager configuration, MMP integration (AppsFlyer, Adjust, Branch or Singular), app events firing correctly, attribution configuration, deep links, and store URLs. Specifically check whether SDK and CAPI events share a common event_id for deduplication, and whether the MMP supports both SKAdNetwork 4 and AdAttributionKit if the campaign runs on a mixed iOS 16/17+ device base. Use only: MATCH, MISMATCH, MISSING, NOT EVIDENCED. Do not infer missing settings. Do not approve setup.

   Do this

   Without event_id-based deduplication between the SDK and CAPI, Meta can count the same conversion twice and over-optimise toward an inflated signal, this is a real, common failure mode, not a theoretical one. The MMP path with server-side event forwarding is the preferred primary source, with SDK events as a supplement, not the other way around.

6. 03

   Build campaign and audience structure

   Review this Meta App campaign and audience structure against the attached blueprint. Review: campaign and ad set structure, new-user-acquisition audiences, existing-user audiences, retargeting audiences, lookalikes, and audience exclusions. Use only: MATCH, MISMATCH, MISSING, CONFLICT. Do not infer missing information. Do not invent audience sizes. Do not approve setup.

7. 04

   Build creative assets

   Review these Meta App creative assets against the approved guidelines. Check: images, videos, Stories, Reels, primary text, headlines, descriptions and CTAs for relevance, duplication, unsupported claims, and missing formats. Use only: PASS, EDIT, REMOVE, HUMAN REVIEW, NOT EVIDENCED. Do not invent offers, claims or promotions. Do not assume policy approval. Do not approve the campaign.

   Do this

   Creative volume is a real, material lever on Meta, not a nice-to-have, most underperforming app campaigns fail from shipping too few creative variants, not from the channel itself. Aim for meaningfully more than one or two assets before judging the campaign's ceiling.

8. 05

   Reconcile the build against the blueprint

   Compare the approved blueprint against the completed campaign build. For every setting, provide: SETTING, APPROVED VALUE, ACTUAL VALUE, STATUS, EVIDENCE, REQUIRED ACTION. Cover measurement and attribution setup, campaign and audience structure, and creative assets. Use only: MATCH, MISMATCH, MISSING, NOT EVIDENCED, NOT APPLICABLE. Do not infer missing settings. Do not modify the campaign. Do not approve launch.

9. 06

   Final launch readiness review

   Review the final Meta App campaign evidence pack. Verify: SDK implementation, MMP implementation, event mapping, install tracking, in-app event tracking, audience implementation, retargeting implementation, creative coverage, app-store readiness, and policy status. Use only: PASS, BLOCKER, HIGH RISK, NOT EVIDENCED, CONDITION ACCEPTED. Missing evidence is not a pass. Do not approve launch. Do not recommend optimisation actions.

   Do this

   Once live, measure cost per retained user, not cost per install, a cheap install that never returns is a worse outcome than a more expensive one that does. Give the campaign 7-14 days to learn before judging it. Have a second person check the campaign before it goes live.