The short answer
"Entity ID" is a name media buyers use for the group Meta's system seems to put similar ads into. Meta has not documented it and there is no Entity ID field in Ads Manager. The useful idea behind it is concept count: among 97 brands with winners on our board, those whose biggest concept holds under a third of their ads reach the top tier at 15.4%, vs 9.8% when one concept holds half or more.
What "Entity ID" means, and what it doesn't
Entity ID is an industry term, not a Meta metric. You won't find it in Ads Manager, the Marketing API or Meta's help center. It is shorthand that spread among media buyers after the Andromeda retrieval system rolled out, to describe a behavior many saw: ads that are near-copies of each other seem to be treated as one option.
One agency write-up puts the working model this way: ads that share an Entity ID "compete against each other inside Andromeda, not against the rest of the market," and a new one is created when the creative changes across at least two of four things (format, persona, environment, benefit). That is a practitioner heuristic. It is useful, but nobody outside Meta has seen the grouping logic.
What is on the record from Meta:
- Its ranking foundation model uses a "creative representation" of each ad as an input (GEM engineering post).
- Andromeda was built with hierarchical indexing to cope with exponential growth in ad creatives (Andromeda engineering post).
Put together: the system has a representation of what each ad is, and organizes a huge pool of ads into an index. Grouping similar ads is a reasonable inference. The specific ID, threshold or rules are not public. If a tool shows you "your Entity IDs," it is showing its own estimate.
The useful part: count concepts, not ads
Whatever Meta does internally, the practical question is the same: how many genuinely different ideas are you running? We can measure that on winning ads.
On Ad Radar every ad is tagged with a narrative angle, a reader awareness level (Schwartz) and a format. We used those three together as a stand-in for "concept." Two ads with the same angle, same awareness level and same format count as one concept. It is a rough proxy (it ignores persona and setting), but it is consistent across brands.
We took every brand with five or more fully tagged winners in our October 9, 2026 snapshot: 97 brands, 961 ads.
5
median distinct concepts per brand
1.7
median ads per concept
15.4%
top-tier share when no concept holds over a third of a brand's ads
9.8%
top-tier share when one concept holds half or more
| Share of a brand's ads in its biggest concept | Brands | Ads | Share reaching top tier |
|---|---|---|---|
| Under 34% | 48 | 482 | 15.4% |
| 34% to 50% | 32 | 347 | 11.8% |
| Over 50% | 17 | 132 | 9.8% |
Top tier means the top 10% of the board by winner score (live 21+ days). The pattern is steady: the more a brand's winners pile into one concept, the smaller the share that reaches the top tier. The last group is only 17 brands, so treat the exact number with care.
Bigger brands add concepts and stack executions
How concept counts grow with the number of winners is also useful for planning:
| Fully tagged winners per brand | Brands | Median distinct concepts | Median ads per concept |
|---|---|---|---|
| 5 to 9 | 62 | 5 | 1.3 |
| 10 to 19 | 28 | 7 | 2.2 |
| 20 or more | 7 | 10 | 2.2 |
Smaller brands run nearly one ad per concept. As brands scale, they do two things at once: they add concepts, and they put two or more executions behind each one (a second cut, a second person, a static version). The 20+ group is only seven brands, so treat its numbers as a sketch.
Under the Entity ID model, this is a sensible way to grow. New concepts open new groups for delivery; extra executions inside a concept give the system options to rotate as frequency builds, without pretending to be new ideas.
One brand's concept map
Nobl, a luggage brand, has 26 winners live. Here are the concepts with tagged angle and awareness:
| Concept (angle / awareness / format) | Ads | In top tier |
|---|---|---|
| Root cause / unaware / video | 6 | 0 |
| Story / unaware / video | 4 | 3 |
| Systematic breakdown / unaware / video | 3 | 1 |
| Root cause / problem-aware / video | 2 | 2 |
| Root cause / solution-aware / video | 2 | 0 |
| Root cause / most-aware / video | 2 | 0 |
| Four other concepts | 1 each | 0 |
The largest group, six root-cause ads for unaware viewers, has no top-tier ad. The four story ads have three. Under the Entity ID model, that is what you would expect: six ads making the same pitch to the same reader split one slot, while a different idea opens a new one.
A TSA agent saw my carry-on and said she was getting one too.
- Est. spend
- $1.1M
- Days live
- 197
- Format
- Video, 93s
- Variants
- 13
A story concept, not a feature pitch. Thirteen variants on one idea suggests the brand found a winner and kept building around it.
Man, there's nothing worse than looking clueless at the airport.
- Est. spend
- $4M
- Days live
- 35
- Format
- Video, 44s
- Variants
- 3
Same product and same root-cause angle as the big cluster, but written for a different reader (problem-aware). Over $4M in estimated spend in 35 days.
The second ad is the interesting one. Its angle is the same as the six-ad cluster, but it speaks to someone who already feels the problem. One dimension changed, and on our board it lands in its own concept.
How to test whether your ads are distinct
Meta gives you no direct readout, so use these checks.
- The one-sentence test. Describe each ad in one sentence without naming the product: who is talking, to whom, about what. If two sentences match, it is one concept.
- The four-field grid. For each ad, fill in angle, awareness level, format and person on screen. Ads that match on three or four fields are probably one group. The practitioner rule of changing at least two of format, persona, environment and benefit is a good minimum.
- The spend-split check. In an ad set where Meta chooses delivery, look at near-twins after a few days. If one takes almost all the spend and the other almost none, and their age, gender and placement breakdowns look the same, the system may be treating them as one option. This is a hint, not proof: Meta also concentrates spend on the stronger of two different ads.
- The new-reach check. When you add a concept, watch reach and frequency for the ad set. A new idea should bring in people the old ads weren't reaching. If reach stays flat and frequency keeps climbing, the new ad probably isn't new to the system.
For the copy-level details of what makes ads look alike, see creative similarity. To plan a set of distinct concepts from scratch, use the creative diversity matrix, and for the difference between a concept and a variant, ad concept vs variation.
If you want to build a concept map like Nobl's for a competitor, follow the brand on Ad Radar and filter its ads by angle and awareness. The tags are already there for every ad on the board.
Figures marked as estimated spend come from Ad Radar's model of engagement on public Meta Ad Library ads. They are estimates, labeled as such, and are best used to rank ads against each other.