Meta AI can now read your Meta ad account, though not from inside Ads Manager. The feature is Meta AI for small businesses, launched 19 August 2026, and it lives in the Meta AI apps: meta.ai, the mobile app, and a new Mac app (Meta for Business). You put your Meta AI profile in the same Accounts Center as the profile managing your business assets, ask a question about your ads, and grant permission when prompted. It then reads campaign data and answers in chat.
Fix two expectations first. It is read only: "Meta AI can review your ad performance. To create, edit, or publish ads, use Ads Manager or Meta Business Suite" (Meta help). And it is "free to get started with these features today", with a paid Meta One tier signposted for heavier use. Meta One's price, date and contents are unpublished as of 3 September 2026.
What this is, and what it is separate from
Three products get confused here, and the confusion costs an afternoon.
Meta AI for small businesses is the one described above. Chat interface, read access, no write actions.
The Meta Ads MCP server is a different thing entirely. It connects an ad account to external assistants like Claude, ChatGPT and Perplexity, and it can write. Its permission controls changed on 11 August 2026, covered below.
Ads Manager's own AI features (Advantage+ campaigns and creative, the recommendations panel) are unrelated to both.
Search Engine Land's headline on 20 August 2026 read "Meta AI can now analyze and optimize Meta ads campaigns". The word "optimize" does more work there than Meta's own help page allows. Meta AI tells you what to optimize. You do the optimizing.
Availability is uneven as of 3 September 2026: "The Meta AI and Vibes apps and websites are not yet available everywhere", with no country list published.
What it connects to
Per the launch post, "you can now choose to have Meta AI work directly with your Instagram and Facebook accounts, Meta ad campaigns, and Google Workspace (Gmail, Docs, Sheets, and Slides)." The help page adds Drive: "It works with your Instagram and Facebook business and ad information, as well as your Gmail and work files on Google Drive."
That combination is the interesting part. An assistant reading your ad results and your margin spreadsheet can compare the two. Ads Manager cannot, because it has never known what your product costs you.
Meta also claims a competitor view from public data: "Beyond your own accounts, you can get publicly available insights about comparable brands and how they are building their presence on Meta." That is Ad Library and public profile data restated conversationally. See how to see competitors' ads for free for the manual version, which shows the actual creative. Outputs go beyond chat: "Meta AI can turn that information into the decks, docs, and spreadsheets you need."
What it flags in an ad account
Meta's "Optimize your Meta ads" section lists four things it looks for, near verbatim: which audiences are delivering, patterns in your best performing content, which ad creative is no longer resonating and why, and where your budget could work harder. Those four are questions of very different difficulty.
Which audiences are delivering is a breakdown query. Meta AI reads numbers that already exist and are already correct. Trust level: high.
Patterns in your best performing content is pattern matching across a small, noisy sample. Useful as a prompt for your own thinking, weak as a verdict. A model seeing three winners with people in the thumbnail will tell you people in the thumbnail work.
Which creative is no longer resonating and why is fatigue detection, the claim to examine hardest. Fatigue is a two-signal diagnosis: an exposure signal such as frequency or first-time impression ratio, plus a response signal such as click-through rate or cost per acquisition. Our internal creative fatigue agent skill, verified 26 August 2026, holds the threshold bands and the differential separating real fatigue from auction pressure, seasonality and a broken landing page. A chat answer calling an ad fatigued without naming both signals has skipped that differential.
Where budget could work harder needs the most care, for the structural reason covered below.
Setting it up, step by step
From Meta's help page on connecting business assets, verbatim. Before you begin: "you must add your Meta AI profile to the same Accounts Center as the Facebook or Instagram profiles that manage your business assets."
For ad and social accounts:
- "Log into your Meta AI profile and make sure it's in the same Accounts Center as the Facebook or Instagram profile you use to manage your business assets."
- "Enter a prompt about your business assets. Meta AI will prompt you to grant permission before working with them."
- "Specify which business asset you want Meta AI to work with when prompting it."
- "To revoke permission, simply ask Meta AI to revoke permission to read data from those business assets."
For Google Workspace: open Settings, then Connected apps, then next to Google tap Connect, sign in, review the access request and tap Allow. "You can disconnect Gmail or Google Drive in Settings."
Step 3 is the one people skip and then complain about. Naming the asset avoids an answer about the wrong ad account.
The prompts Meta published
Meta's own examples, verbatim from the help page. Use them as written the first time: they show the shape of request the system was built to answer.
- "Review my ad performance over the last 90 days: what's working, what's not? What do my top campaigns have in common, and what should I change in the next 30 days?"
- "What types of content have worked best for my Instagram account over the last month? What can I do differently? Send me a report every Monday." One correction on the first prompt. Its 90 days is Meta's example window, not a documented lookback limit. No limit has been published. Ask for 12 months and see what comes back.
The "Send me a report every Monday" clause is the recurring task feature: "you can let Meta AI handle work you've been doing manually by setting up recurring tasks and reminders."