Meta quality ranking: not a field the connector returns

The three relevance rankings are the columns an operator adds in Ads Manager when an ad starts costing more and the creative gives no reason why. They are comparative, and they point at whether the problem is the creative, the audience or the page after the click. Asking the connector for them looks like a one-line request.

The request works. That is the whole difficulty. Nothing throws, nothing warns in the reply body, and the answer that comes back looks like an account with no ranking data rather than a connector without the field.

What the auction tool says it does

Its own description is the only technical documentation of this tool that exists. Verbatim, it “highlights which ads have generated stronger results in the auction and identifies factors (bid, ad quality) that can be optimized for improved performance”, and it “recommends consolidating similar ad sets and reducing audience fragmentation in cases of high auction overlap, as high overlap leads to under-delivery and budget fragmentation”.

The parameters are thinner than the sentence suggests: an ad account, an optional list of entity IDs that must all be the same type, and a date range. No metric argument, no threshold, no ranking to request.

What the call actually returned

Run on 30 August 2026 against every ad account on this login — account-wide and scoped to named ad sets, over the last month and over lifetime — the reply never changed:

No auction ranking benchmarks data available for the given criteria.

That included an account with months of continuous delivery behind it, and ads whose impression counts sit comfortably above the volume Meta's help centre requires before relevance diagnostics appear at all.

The message is not the connector failing to understand the request. Handed an entity ID that does not exist, the same tool returns a proper validation error, names the ID, and states it is not a valid campaign, ad set or ad. Valid inputs produce the sentence; invalid ones produce an error. The sentence is therefore a statement about coverage, and it is the identical failure mode as the industry benchmark tool, which answers the same way in the same family.

The failure is silent, and that is the risk

Passing all three to the entity tool anyway does not produce an error. The ads come back with the fields the connector does hold, the ranking columns are missing without comment in the data, and the explanation is parked in a separate note beside the rows:

Unsupported fields: quality_ranking, engagement_rate_ranking,
conversion_rate_ranking. Supported fields at this level are: adset_id,
amount_spent, ... cpc, cpm, creative_id, ctr, frequency, impressions, ...

A model that asked for those three columns and received rows has, from its own point of view, succeeded. Unless it reads that note back, the natural summary is that the ads carry no relevance rankings — which a founder reads as a fact about their account. It is a fact about the connector.

Knowing which questions the connector cannot answer is half of running an account through it. See how the weekly review is structured →

How to get the rankings anyway

1. Read them where Meta publishes them START HERE

   Ads Manager, ad level, with the three ranking columns added to the view. They are comparative labels rather than numbers, they only appear once an ad has enough delivery behind it, and they are the only place these three are exposed to an ordinary advertiser today. The trade: it is manual, and because the values never travel back through the connector they cannot land in an automated weekly note.

2. Ask the connector what it holds before trusting any report

   One call to the field tool, with no arguments, returns the entire catalogue and the levels each field is available at. Keep the answer where the model can read it, and a request for a column that does not exist stops being a silent gap in a report. The trade: it is a snapshot, and Meta does not announce catalogue changes.

3. Rebuild the same diagnosis from fields the connector does carry

   Click-through rate, frequency, cost per thousand impressions and cost per result all come back at ad level, and read against the same ad's own history they cover most of what the rankings were being consulted for — the same argument as picking winners on the wrong column. The trade: these are outcomes, not the auction's opinion of the ad, so they move for reasons that have nothing to do with relevance.

A tool that returns nothing and a field that does not exist look the same in a report.

• Reading the empty sentence as a competitive verdict. No ad was ranked against anything. The sentence describes what the tool covers, not how the account is doing in the auction.
• Expecting an unsupported field to fail loudly. It does not. The field is dropped, the call succeeds, and the only trace is a note beside the data.
• Accepting a paraphrase. A weekly summary reporting that the auction rankings look fine is describing an absence, and it will be filed as reassurance.
• Hunting for the argument that was wrong. There is no ranking parameter to correct and no date range that changes the answer.

Try this tonight Open Ads Manager at ad level, add quality ranking, engagement rate ranking and conversion rate ranking as columns, and sort by amount spent. Look only at ads that are spending. Where conversion rate ranking sits below average while engagement holds up, the ad is doing its job and the page after the click is not — so tonight's job is the landing page, not another creative. Where quality ranking is the low one, it is the ad. Write down which of the two it is before you touch anything.

The pattern repeats across this connector: make the call, read the exact words back, then work out what was never in scope. The counted manifest lists every tool it exposes, the anomaly signal tool does name auction overlap when it has something to say, and the connection walkthrough covers setup.