AI ad compliance checks

The Adsidian dashboard's AI assistant drafts ads for Meta, Google, Reddit and LinkedIn — this article explains the automatic policy check that runs on that copy before you approve it.

What it is

Every ad the assistant drafts or edits — copy, headlines, descriptions, destination URLs — is checked the moment it's staged, before it reaches your approval screen. The check is independent of the assistant that wrote the ad: it never sees the conversation that produced the copy, only the finished text and links, the way a second reviewer would. That separation is deliberate — a check sharing the drafting assistant's own assumptions wouldn't catch what the assistant missed.

The four checks

Four layers run on every draft:

1. Platform rules by vertical. A set of pattern checks tuned to Google's and Meta's ad policies, plus extra rules for regulated verticals like healthcare. Examples: absolute cure claims ("cures your back pain for good"), phrasing that implies you know something about the reader's health or situation ("you're struggling with..."), before/after language, unsubstantiated superlatives ("#1", "guaranteed", "best in town"), trademark symbols and competitor names, and leftover placeholder text.
2. An independent AI policy review. A second, separate model reads the copy against Google's and Meta's actual policy areas (misleading claims, personal attributes, health & wellness, editorial requirements, destination requirements, and more) and flags anything the rule patterns wouldn't catch.
3. A landing-page check. For any ad with a destination URL, the gate loads the page and confirms it actually works, that it stays on your domain (not redirected somewhere unexpected), that a price mentioned in the ad copy actually appears on the page, and that a lead-collecting ad's page has a visible privacy-policy link.
4. Google's own pre-check, for Google search ads. Before anything is created, the gate submits the ad to Google in "validate only" mode — nothing is published — and reads back Google's own policy verdict, catching anything Google would reject at review time.

What you'll see on the approval screen

Every staged ad shows a status chip:

Chip	Meaning
Passed	Nothing flagged.
Needs attention	At least one finding, none serious enough to block.
Blocked	A serious finding is stopping this ad from being staged as-is.
Partially checked	One of the four layers couldn't finish (a timeout, a page that wouldn't load) — read the copy yourself before approving.

Below the chip, findings are grouped by the field they apply to (a specific headline, a description, the destination URL), each with the flagged text, a plain-English explanation of the policy it touches, and — where the gate can generate one — a suggested rewrite. Use suggestion drops that rewrite straight into the field. If there's anything to review, you'll also see a checkbox — I've reviewed these findings — that you need to tick before the Approve button becomes clickable.

Modes

Your workspace runs in one of three modes:

• Enforce (the default) — a serious finding stops the ad from being staged at all. The assistant rewrites the flagged copy and re-proposes it to you rather than asking you to approve around the block.
• Advisory — findings are shown and you tick the checkbox, but you can still approve the ad regardless of what was found.
• Strict — the same as Enforce, but minor findings block too, not just serious ones.

Advisory and Strict remain selectable in Agency Settings if Enforce isn't the right fit for your workspace.

A fourth mode, Off, exists for internal use only and isn't offered on customer workspaces.

Per-client settings

A handful of things are tuned per client rather than per workspace: terms that should always be blocked, competitor names to flag automatically, claims you can substantiate (so a real "#1" or "guaranteed" claim doesn't get flagged every time), extra rule packs for a client's industry, and which domains an ad is allowed to link to. There's no self-serve screen for these yet — contact support to set them up for a client.

After you publish

Passing this check isn't the same as passing the platform's own review. Once an ad is live, the hourly sync reads back Google's and Meta's own review status. If a platform later disapproves an ad or limits its delivery, you'll see a badge on that ad and get a notification — so the loop closes even when the issue only shows up after the ad has gone through the platform's own systems.

What it can't do

This check is a safety net, not a guarantee. It doesn't replace platform review — Google and Meta can still reject or limit an ad it passed, and their decision is final. It doesn't examine images beyond the flags Adsidian's creative analysis already produces when you upload an asset. It doesn't check trademarks against an official registry — only the competitor and blocked-term lists set for that client. And it never rewrites your ad on its own; it only suggests a rewrite for you or the assistant to apply.

A record of every check

Every check is saved — what was checked, what it found, and who approved the ad despite any findings — whether or not the ad ended up being blocked. That record exists so any AI-drafted ad can be traced back to exactly what was reviewed before it ran.