Same creative, many ads: de-duplicating by creative_id

You ask which ads are working. Back comes a tidy table, sorted, five rows at the top. Two of those rows are the same picture running against different audiences, one is that picture again with a swapped headline, and you are now looking at a list of three things dressed as five.

The fix gets repeated everywhere, including on this site: group by the creative, not the ad. Nobody says how you actually get that grouping out of Meta's connector, and it is more awkward than the advice implies. An ad and a creative are two objects, they live behind two tools, and the field that joins them is inert.

Why nothing groups it for you

The connector ships a tool that returns metadata about its own reporting fields, which means you can interrogate the vocabulary instead of guessing at it. Asked about creative_id on 23 August 2026, it returns: "The unique identifier of the creative used in the ad", type integer, supported levels ["ad"] — a list with one entry — filterable: false, sortable: false, and an empty list of filter operators.

Read that against a normal metric. Amount spent is filterable and sortable at every level, with a full set of comparison operators. Spend is something the platform will reorganise your account around. The creative identifier it will hand you, and nothing more.

The consequence arrives late. There is no creative level to request — "creatives ranked by spend" is not a question the reporting surface can take, the way campaigns ranked by spend is. The collapse happens after the rows land, in a spreadsheet or in the assistant's working memory, which is exactly the arithmetic worth checking against Ads Manager.

Two surfaces, one join key

Performance and creative attributes come from different tools. The reporting tool knows what an ad spent and produced, plus the creative's identifier and nothing else about it. The creatives tool knows the body text, the headline, the image hash and the underlying post — and nothing about performance. creative_id is all they share.

There is a second trap, and the creatives tool warns about it itself: a plain listing returns only the identifier, name, account and status. Everything else is omitted, and the schema says in as many words "do NOT assume a field is empty just because it was missing from a previous listing response." Ask for the image hash or the underlying post by name, in a second call. An assistant that lists creatives once and then discusses their content is reading an empty table.

It matters because two ads can be the same picture and still carry different identifiers: duplicating an ad reuses the creative object, re-uploading the same file makes a new one. Motion, whose creative-reporting product exists to solve this, tells its users the same thing in its help documentation on grouping by creative — duplicate rather than re-upload, or the report splits one creative into several. Vendor doc, not Meta's. It matches what the field list implies: image_hash and object_story_id sit below the identifier, and they settle the question.

The order that survives contact with an account

1. Pull at ad level, with the field named START HERE

   Ask for a window, level ad, and name creative_id in the field list. It is not in the default set and it will not appear just because the question was about creatives. The trade: ad is the only level carrying it, so a busy account returns a lot of rows and the tool caps how many come back per call.

2. Collapse, then check the collapse

   Sum spend and results across rows sharing an identifier, then take the biggest resulting creative and add its rows up by hand. Nothing verified that arithmetic, and you are about to spend against it. The trade: a minute — the one that makes the rest of the table usable.

3. Ask whether two identifiers are one asset

   Where two creatives look identical, pull image_hash and object_story_id for both. Same hash, same picture. Same story identifier, same underlying post, which also means the likes and comments pool in one place. The trade: an extra call each, so run it on the top of the list, not the account.

4. Only now rank

   One row per creative, totals summed, and ranking is the easy part — by volume and by share of spend against share of results, not by the ratio everyone sorts on.

An ad is a placement decision. A creative is the thing you made. Reports count the first and you make more of the second.

Getting the unit right is the first thing the weekly routine does, before it looks at a single number. See how the week runs →

Asking for it in one go

Field names matter here, because the obvious guesses are wrong. There is no field called purchases; the catalog resolves it to omni_purchase, which displays as "purchases". Spend is amount_spent, with spend accepted as an alias. Both checked on 23 August 2026.

For ad account <ID>, window <YYYY-MM-DD> to <YYYY-MM-DD>:
level = ad. Fields: id, name, creative_id, amount_spent,
omni_purchase.
Echo the window back, then group the rows by creative_id and sum
amount_spent and omni_purchase within each group.
Show me the row count, the number of distinct creative_id values,
and the grouped table sorted by omni_purchase.
Do not sort or filter on creative_id — the API does not support it.

The two counts are the point. If the row count is much larger than the number of distinct identifiers, every per-creative figure you have read so far was split across duplicates.

• Grouping by ad name. Names are typed by humans under time pressure. The identifier is assigned by Meta and cannot be typed wrong.
• Assuming the assistant grouped anything. Ask it to state the row count and the distinct count. If it cannot, it summarised rather than grouped.
• Treating identical pictures as one creative without checking. Re-uploaded assets get their own identifier, and no amount of prompting will merge them for you.
• Doing this at campaign level. The field is not there. You will get an answer anyway, which is the problem.

Try this tonight Run the prompt above over your last thirty days and read only the two counts. If rows outnumber distinct creatives, take the identifier with the most rows and look at which ad sets they sit in. That is one creative competing with itself, and deciding whether that was deliberate is tonight's real decision.

This is question two of the four-question audit, expanded. If nothing is connected yet, start here.