ads_get_errors: what Meta's connector reports, and what it skips

An ad set is not spending. You ask Claude whether anything is broken, it runs the tool that exists for exactly that, and reports nothing. Twenty minutes later you open Ads Manager out of habit and there is a red icon in the delivery column.

Both answers are true. "Is anything broken" has three separate homes inside this connector and the errors tool owns one of them. Knowing which one is the difference between a quiet answer you can trust and a quiet answer that means the question went elsewhere.

What the tool actually is

It takes a list of entity IDs — campaign, ad set, ad, or ad account — and accepts several at once, which is the genuinely useful bit for anyone holding more than one account.

The account-level behaviour is worth reading twice. Verbatim from the tool's own description:

If account ID is provided, it returns all errors of its child ad entities, but not the account itself.

So the sweep you would naturally reach for — hand it the account, ask what is wrong — cannot report that the account is the thing that is wrong. A billing failure, a restriction, a hold on the account itself: out of scope by construction. And if the account has stopped answering queries at all, you get neither the parent nor the children.

When there is nothing to report, this is what comes back:

{"errors": "[]"}

An empty list, delivered as a string rather than as an array — a non-event if you are reading the reply in a chat window, a real one if something downstream is parsing it. I ran it on 26 August 2026 against live ad accounts, singly and then every ID in one call, and got that same response each time.

Where each kind of "error" actually lives

Three of the rows below are the connector telling you in its own words that it will not answer. The rest is what the field-metadata tool reports as selectable and filterable at campaign, ad set and ad level.

That catalog row is a design choice worth noticing. The tool description instructs the assistant that when an error mentions a product set, a catalog, a feed or product eligibility, it should go to the catalog diagnostics and dynamic-ads health tools on the backing catalog for the root cause. The error text is a signpost, not a diagnosis — and unfollowed, it is a symptom and a dead end.

Most of the value here is knowing which question to ask before you trust the answer to it. See what the weekly routine checks →

The status field does the work the errors tool won't

Ask the connector's field-metadata tool about effective_status and it describes itself as the actual delivery status, which can differ from the status you set because it accounts for parent state, ad review, billing and budget conditions. It is filterable and sortable at campaign, ad set and ad level, and its values include WITH_ISSUES, DISAPPROVED, PENDING_REVIEW and PENDING_BILLING_INFO.

A second field, delivery, is coarser and easier to prompt with: an enum running draft, pending, active, inactive, error, deleted, completed and off. Filtering on error gets you the entities to hand to ads_get_errors. Between them you can find everything the errors tool is not allowed to mention.

One field that does not exist here is worth flagging. issues_info — the field the Marketing API uses to carry delivery diagnostics on an ad — comes back from the connector's field-metadata tool under unknown_fields. You cannot pull error detail as a column alongside spend. It is always a second call.

How to ask whether anything is broken

1. Ask for the status column before the errors START HERE

   "Show every ad set with its configured status and its effective status" gets the rejections, the review holds and the billing stops in one table — the categories the errors tool declines. Do it first and the errors call becomes a follow-up on a short list rather than a blind sweep. The trade: a table to read rather than a yes-or-no, padded with ordinary paused rows you have to look past.

2. Then run the errors tool on the account ID

   This is the cheap sweep and worth having — pass every account ID you manage in one call. Read the empty answer for what it is: nothing found among the child entities, in the one category covered. The trade: it can never report a problem with the account itself, so a suspended account and a healthy one produce the same empty list.

3. Chase catalog wording to the catalog tools

   If an error mentions a feed, a product set or product eligibility, that string is the start of the diagnosis, not the end. Catalog diagnostics, dynamic-ads health, data sources and feed uploads each have their own tool.

4. Confirm the fix in Ads Manager, not in the chat

   An error that has cleared and an error the tool was never going to mention look identical from here. One look at the delivery column closes the loop — the same habit that catches a metric reporting nothing as null.

An empty error list means nothing was found in the one place the tool is allowed to look.

• Treating a clean sweep as an all-clear. Three categories of stoppage are excluded by name, and the reply never mentions it.
• Running it on the account and expecting account news. The description is explicit that the account itself is excluded, so a suspended one answers exactly like a healthy one.
• Asking about "errors" when you mean underdelivery. Pacing and learning problems are not errors here, so you get a confident nothing.
• Stopping at the error string. For anything catalog-shaped it points at a different tool, and a symptom copied out of a chat window is not a diagnosis.

Try this tonight Ask for every ad in the account with its effective status, then run the errors tool on the same account ID, and put the two answers side by side. Anything sitting in DISAPPROVED or PENDING_REVIEW that the errors call did not mention is tomorrow morning's list — fixable today, and otherwise invisible.

This is one entry in a long manifest; the rest is in the full tools list. Not connected yet? Start with the setup.