Meta AI Can Now Read Your Ad Account, but Not Run It

Meta AI can now connect directly to a business’s ad account — and to its Google Workspace. Meta announced on August 19, 2026 that its assistant, on web, mobile, and a brand-new Mac desktop app, can link to a business’s Facebook and Instagram accounts, its Meta ad campaigns, and Gmail, Docs, Sheets, and Slides, then analyse ad performance conversationally and turn what it finds into decks, docs, and spreadsheets.

The easy misreading — that Meta AI now “runs your ads” — is wrong, and this post exists partly to keep you from repeating it. Every capability Meta documents is analysis, benchmarking, document generation, and scheduled reporting. A human actions every recommendation. The system that does act on campaigns autonomously, Advantage+, is a separate, pre-existing product — and conflating the two is the fastest way to misbrief a client.

This guide covers exactly what Meta announced, the precise line between Meta AI and Advantage+, what the new Mac app does and does not do, and the question Meta’s own documentation leaves open: what happens to the Google Workspace content the assistant reads once you connect it. That last gap is a checked finding, not a guess — and for agencies it is the deciding factor.

The core of the Meta for Business announcement is a set of account connections. Once linked, Meta AI can answer questions about a business drawn from data Meta already holds — and, for the first time, from a connected Google Workspace. Three groups of assets connect:

Two behaviors turn those connections into a workflow. First, output: Meta’s own phrasing is that the assistant produces the “decks, docs, and spreadsheets you need to keep your business growing.” Second, recurrence: it can run on a schedule — ask for a report every Monday and it delivers one, via recurring tasks and reminders. Meta had already given the assistant general planning and follow-through features in an earlier update this summer; the August 19 announcement is the business-specific application of that trajectory, wired into ad accounts and Workspace.

The ad-performance data it reads is the same data whose meaning has been shifting all year — if your attribution setup changed recently, the assistant’s read of “performance” inherits that change. Our breakdown of what changed in Meta’s click-through attribution is the necessary companion context here.

Read the announcement closely and a pattern holds without exception: every verb Meta uses for the ad-account connection is a reading or writing-about verb. Analyse performance. Surface deliverability patterns. Flag fatigued creative. Identify budget-allocation opportunities. Benchmark against comparable brands. Draft the deck. Send the Monday report. Nowhere — not in the primary announcement, and not in any of the same-day coverage from Engadget, MacRumors, Unite.AI, or TestingCatalog — does Meta claim the assistant adjusts a budget, edits an audience, swaps a creative, or pauses a campaign on its own.

That makes Meta AI a different species from the automation advertisers already run. API-level automation — the Marketing API workflows we’ve covered before — executes changes programmatically under rules you wrote. Advantage+ executes changes inside Meta’s own optimization loop. Meta AI, as documented on August 19, executes nothing: it is a conversational layer that reads your data and hands a human a drafted next step. In the terms we used in our guide to Meta’s broader AI-in-ads push, this is the advisory tier of the stack, not the actuation tier.

Why does the distinction matter beyond pedantry? Because the two tiers fail differently. An actuation system that misreads your data spends money wrong. An advisory system that misreads your data persuades a human to spend money wrong — with a professional-looking deck attached. The discipline for adopting Meta AI is therefore editorial, not technical: treat its output the way you would treat a junior analyst’s first draft. Check the numbers against Ads Manager before any recommendation it drafts becomes a change you make.

Meta defines Advantage+ as “a suite of products that helps advertisers maximize performance by using AI to optimize campaigns in real-time and match ads to the people most likely to take action.” It responds to real-time data, delivers ads across the widest audience, and chooses the most effective combinations — automatically, inside Ads Manager. That is the decision-making layer of the campaign itself. Meta AI, by contrast, talks about the campaign. We mapped how Advantage+ already reads chat signals earlier this year; the table below puts the two systems side by side, sourced from Meta’s August 19 announcement for the Meta AI column and Meta’s Advantage+ Business Help Center page for the other. None of the same-day coverage we checked — Engadget, MacRumors, Unite.AI, TestingCatalog — draws this comparison, which is exactly the gap that makes the two easy to conflate.

The practical takeaway from the table: the two systems are complementary, not competing. Advantage+ keeps optimizing whether or not you connect Meta AI, and Meta AI reads the ad-performance data Meta already holds and talks about it. What is genuinely new on August 19 is that the advisory layer now spans data Meta never held before — your Gmail, your Docs, your Sheets — which is where the next two sections go.

The Mac app is Meta AI’s first dedicated desktop client, giving the assistant a native desktop home for the first time, per Unite.AI’s same-day analysis. Two capabilities are desktop-only. First, screen sharing: point the assistant at any window on your desktop and it uses what it sees as context — to answer questions, suggest edits, or draft content. Second, system-wide dictation: use your voice to dictate in any app. Meta’s own product page files both under “Works where you do.”

The same discipline from Section 02 applies on the desktop: the assistant reads the window you attach; it does not click, type, or navigate on your behalf. On the build itself, the two same-day write-ups each add a different detail: a hands-on from TestingCatalog reports that the build it accessed was already labeled version 1.0 and that the app was live only in certain regions at launch, while MacRumors separately describes it as available in a beta capacity — together, a staged rollout, consistent with Meta giving no completion date anywhere in the announcement. Deeper technical specifics about the app’s build — installer size, UI framework, minimum macOS version — are the kind of detail unofficial hands-on inspection surfaces long before a vendor does. Meta publishes none of it in the announcement or on its Meta AI product page, so we will not state unconfirmed specs here.

Here is the checked finding this post leads its governance discussion with, because it is a documented absence rather than a suspicion: neither Meta’s Help Center page on connecting Meta AI to business assets nor the live Meta AI product page contains any statement — favorable or unfavorable — about how long connected Google Workspace content is retained, whether it is used to train Meta’s models, or how it is stored once read. Both pages were checked for that language; as published, neither carries it. Before an agency connects a client’s Workspace, that is the question it must be able to answer, and today Meta’s own documentation does not answer it.

Two adjacent facts must stay distinct from that absence, because merging them produces claims the sources don’t support. First, what is documented: the connection flow (Settings → Connected apps → Google → review what Meta AI requests → grant access) and revocation, two ways — ask Meta AI directly to revoke its permission to read data from those business assets, or disconnect Gmail and Google Drive from Settings. Users can also ask Meta AI to stop using their business data at any time. Second, Meta’s general privacy policy: MacRumors’ same-day coverage notes that “interactions with features that are part of AI at Meta” are used to train Meta’s models. Note precisely what that covers — the conversation with the assistant — not a specific statement about content read from a connected Gmail inbox or Google Doc. The training status of that connected content is exactly the undocumented part.

None of the launch coverage addresses the reader we care most about: the agency operating accounts on someone else’s behalf. That reader faces a decision the small-business owner does not — connecting a client’s Google Workspace to Meta AI is an access decision made for the client, on data the agency does not own. Our recommended ladder, in order of rising stakes:

The through-line: adopt the analysis where the data already lives with Meta, and hold the new data grant until Meta documents its terms. If you want help formalizing that posture — which AI tools get which access to which client systems, and what the disclosure language looks like — that governance work is part of our paid media management engagements, and the client-data side of it is what our CRM and automation practice handles daily.

Meta’s framing is consistent across the announcement and the same-day coverage: this is small-business and creator tooling — the businesses running their own marketing on Facebook and Instagram without an analyst on staff. It is not framed as an enterprise product, and the launch documentation matches that framing: no admin controls, no per-seat governance, no enterprise data-processing terms anywhere we checked.

Looking forward, the direction of travel seems clear even where the documentation is thin. Meta spent the summer giving its assistant planning and follow-through features, then in August wired those into ad accounts and Workspace for small businesses. The free-to-start pricing with a named subscription waiting suggests Meta expects the business analysis habit to form first and monetize later. If the habit forms, the pressure to close the gap between drafting a recommendation and executing it will be commercial, not technical — which is why the analyst-versus-operator line this post draws is worth dating precisely. As of August 19, 2026, the line holds: analysis is documented, execution is not.