How to find winning ads and tell them from tests

The signals a winning Ad leaves

How the winner Score reads these signals

How to find winning Ads with your Assistant

1. 1

   Search a category or a competitor, including stopped Ads

   Search by keyword or by Advertiser. Ask for all Ads, not only active ones: the Ads that ran for months and then stopped are often the clearest winners, and the short-lived ones show you what the Advertiser tested and dropped.

   Search the Meta Ad Library for protein powder Ads in the US, active and stopped, sorted by winner Score. List the Advertiser, start date, days running, variant count and winner Score for each.

2. 2

   Group the Ads by Advertiser

   Compare each Ad with the same Advertiser's other versions. Ask the Assistant to group the list by Advertiser and, within each one, compare how long each version ran. Versions that share a start date and stopped early are the tests. The ones still running weeks later are the keepers.

   Group these Ads by Advertiser. For each Advertiser, show which versions stopped early and which kept running, with days running for each.

3. 3

   Check variants added and confidence

   Ask which Ads gained variants between Snapshots and which Scores are high confidence. A high-confidence Score with variants added is the strongest public signal. A Score of 100 at low confidence often means the Ad is simply the longest runner of an Advertiser we have seen only a few Ads from.

   Which of these Ads have variants added above zero, and which winner Scores are high confidence? Explain the reason on each low-confidence Score.

4. 4

   Narrow by placement and format

   If you make Reels, ask for Ads placed on Instagram with video media. Placement works best as a filter for the format you produce. On its own it says little about whether an Ad works.

   Keep only video Ads that run on Instagram, and sort them by days running.

5. 5

   Analyse the shortlist

   Ask for Ad analysis on the five to ten Ads that survived the checks. It reads the video or image itself: the hook, the visuals, the on-screen text and the offer. Then ask what the winners have in common that the stopped tests do not.

   Analyse the top five Ads and compare them with the versions that stopped early. What do the long runners do differently in the first three seconds?

6. 6

   Track the Advertiser to build history

   Set an Advertiser Tracker on the brands you care about. Each daily cycle takes a new Snapshot of their Ads, so days running, variants added and the confidence of each Score improve over time, and a Change Report tells you when a new Ad appears or an old one stops.

   Track this Advertiser and tell me each day which Ads started, which stopped and which gained variants: (paste a Meta Page or Ad Library link)

A winner and its tests, from one real search

{
  "run": { "kind": "ads_search", "status": "completed", "charge_credits": 50 },
  "result": {
    "library": "meta",
    "items": [
      ...
      {
        "id": "1542145170709553",
        "advertiser": { "id": "36543730620", "name": "SAMBAZON" },
        "creative": { "title": null, "cta": "Learn more", "format": "VIDEO" },
        "media_type": "video",
        "placements": ["INSTAGRAM"],
        "first_shown": "2026-07-22T07:00:00.000Z",
        "last_shown": "2026-10-27T07:00:00.000Z",
        "is_active": false,
        "days_running": 97,
        "variant_count": 10,
        "winner_score": {
          "score": 90,
          "inputs": { "snapshots": 3, "days_running": 97, "advertiser_ads": 6, "variants_added": 4 },
          "reason": null,
          "confidence": "high"
        }
      },
      ...
      {
        "id": "1566476351495220",
        "advertiser": { "id": "36543730620", "name": "SAMBAZON" },
        "creative": { "title": null, "cta": "Learn more", "format": "VIDEO" },
        "first_shown": "2026-07-22T07:00:00.000Z",
        "last_shown": "2026-10-23T07:00:00.000Z",
        "is_active": false,
        "days_running": 93,
        "variant_count": null,
        "winner_score": {
          "score": 44,
          "inputs": { "snapshots": 2, "days_running": 93, "advertiser_ads": 6, "variants_added": 0 },
          "reason": null,
          "confidence": "high"
        }
      },
      ...
      {
        "id": "1571866057722171",
        "advertiser": { "id": "36543730620", "name": "SAMBAZON" },
        "creative": { "title": null, "cta": "Learn more", "format": "VIDEO" },
        "first_shown": "2026-07-22T07:00:00.000Z",
        "last_shown": "2026-09-09T07:00:00.000Z",
        "is_active": false,
        "days_running": 49,
        "variant_count": 10,
        "winner_score": {
          "score": 10,
          "inputs": { "snapshots": 1, "days_running": 49, "advertiser_ads": 6, "variants_added": 0 },
          "reason": "single_snapshot",
          "confidence": "low"
        }
      },
      ...
    ],
    "next_cursor": "eyJjIjoi..."
  }
}
(50 Ads in all; these are 3 of the Advertiser's 6. Media URLs, Ad body and other fields trimmed)

{
  "items": [
    {
      "at": "2026-11-01T17:56:14.914Z",
      "metrics": {
        "reach": null,
        "spend": null,
        "running": true,
        "currency": null,
        "countries": null,
        "placements": ["FACEBOOK", "INSTAGRAM"],
        "impressions": null,
        "variant_count": 2
      }
    }
  ],
  "next_cursor": null
}
(The same Ad in its search: days_running 386, winner_score 100, snapshots 1, advertiser_ads 3, variants_added 0, confidence low, reason single_snapshot)

What 550 Ads looked like on one day

What it costs

Connect your Assistant

Prompts to try

• •Find Meta Ads for meal kits in the US that have run more than 90 days, and tell me which Advertisers have more than one.
• •For this brand, list every Ad with its start date and days running, and mark the ones that look like tests: (paste the brand name or Page link)
• •Sort these Ads by winner Score and drop any with low confidence. What is left?
• •Which of these Ads share a start date? For each group, which version is still running?
• •Analyse the three longest-running video Ads and compare their hooks with the ones that stopped within a month.

Related reading

• How long Meta Ads runDays running for 546 US Meta Ads from 11 Ad Library searches, stopped and running, by category and format.
• Ads in the docsEvery Meta filter, what each Ad carries, the full winner Score formula and how an ad search is priced.
• PricingEvery price in Credits, with no subscription.

Questions

What is a winning ad?

An Ad that pays back well enough that the Advertiser keeps running it and makes more versions of it. From the outside you cannot see sales, so you read the behaviour: how long the Ad has run compared with the same Advertiser's other Ads, and whether new variants keep being added.

How long does an ad need to run to be a winner?

There is no fixed number. Some brands rotate creative every week and others leave Ads running for a year. Compare an Ad with its siblings from the same Advertiser, ideally versions launched on the same day. The winner Score does this ranking for you on Meta Ads.

Why do so many Ads have a winner Score of 100?

The Score ranks an Ad against the Advertiser's other Ads we have on file. When there are only one or two, the longest runner comes out on top by default, and an Ad with no siblings on file reaches 100 at 90 days. Those Scores are marked low confidence, with the reason single_snapshot or few_advertiser_ads. With one Snapshot, or under 5 dated Ads from the Advertiser, use the Score to order your reading and check the Ads yourself.

Can I find winning TikTok ads?

Partly. The TikTok Ad Library covers Ads shown in the EU/EEA, the UK, Switzerland and Turkey only, and it gives first and last shown dates but no variant count, so ShortsIntel puts no winner Score on TikTok Ads. You can still ask your Assistant to sort them by how long they ran.

Can one snapshot of an Ad show that it is being scaled?

No. One Snapshot shows the variant count at that moment. Scaling shows up as variants added, which needs at least two Snapshots on different days. Tracking the Advertiser takes a new Snapshot every day.

How much does it cost?

A Meta search costs 1 credit per Ad delivered, Ad analysis 20 credits per Ad, and reading Snapshots is free. New accounts get 1,000 credits free, valid for 90 days. A Run that fails costs nothing, and a partial one is charged only for what finished.

Related guides

• Meta Ad Library searchHow to search Meta's Ad Library by keyword, Advertiser and country, by hand and through an Assistant, with real sample output.
• TikTok Ad LibraryWhat TikTok's Ad Library holds, the countries it covers, what each Ad shows, and how to search it from your Assistant.
• TikTok Ad examplesSix real TikTok Ads from the TikTok Ad Library, each read for its hook, format and offer, with the reach and targeting the library publishes.
• UGC Ad researchHow to research UGC Videos and Ads by format and hook, with real examples, and turn the findings into briefs for UGC creators.
• Track competitors on socialTrack competitor Creators, Hashtags, Sounds, Keywords and Advertisers daily, and read a Change Report when something moves.
• Find outlier VideosHow to find Videos that beat their Creator's usual reach in a niche, read their Scores and turn them into briefs.
• Meta Ads MCPWhat an Assistant can do with the Meta Ad Library through the ShortsIntel MCP server: Ad and Advertiser search, Ad analysis, Snapshots and Advertiser Trackers.
• TikTok MCPWhat an Assistant can read on TikTok through the ShortsIntel MCP server, including the TikTok Ad Library, with sample output and connect steps.
• Social media MCPWhat the ShortsIntel MCP server lets an Assistant read on each Platform, and how to connect Claude, ChatGPT, Claude Code or Cursor.
• Creative strategy with AICreative strategy and creative analytics with an AI agent that reads real Videos and Ads, plus every creative and hooks guide.
• Meta Ad Library APIWhat Meta's official Ad Library API returns, what it leaves out for commercial Ad research, and another way to read those Ads.

Try it in your Assistant