· 18 min read
Facebook Ad Library Search: How to Query It Like an Analyst
Most people browse the Meta Ad Library. Searching is a different skill: three query types, what each filter really scopes, and why the URL is what you save.
Almost everyone who opens the Meta Ad Library does the same thing: types a brand name, scrolls, screenshots three ads, closes the tab. That is browsing, and it answers exactly one question — what is this advertiser running right now.
Searching is a different operation with a different index. It answers questions that start with “who”: who in this market is making this claim, who is still paying to say this six months later, who appeared in this category last month that was not there before. Those are the questions that produce a creative brief, and they cannot be answered by typing a brand name.
The gap between the two is not effort. It is knowing how the query grammar works, what each filter actually scopes, and which parts of the interface are lying to you by omission. This guide covers the search layer specifically: the three query types, the filter-by-filter semantics, the reason a saved URL beats a saved screenshot, and a repeatable protocol you can run in under half an hour.
If you have never used the library at all, start with the free walkthrough of competitor ad research and come back. This piece assumes you know what an ad card looks like and want to get better at finding the right cards.
Browsing and Searching Are Two Different Indexes
An advertiser search is anchored on a Page ID. It returns every ad that Page is running in the country you selected, newest first, and it is complete within that scope. A keyword search is anchored on text. It returns ads from many Pages that contain your words somewhere in the ad’s copy, and it is inherently incomplete — the index only knows what was written down.
That difference decides which one you use:
- Use a Page search when the question is about one advertiser. “What is this brand testing this week” is a Page question. It has a clean answer and no noise.
- Use a keyword search when the question is about a market. “Who is selling a two-week detox in the UK right now” is not answerable per advertiser, because you do not know the advertiser list yet. The search builds that list for you.
- Use both in sequence when the question is about a pattern. Search a claim, collect the advertisers running it, then open each one’s Page view to see whether the claim is their main angle or a single test.
The common failure is using a keyword search to study a brand you already know. You will get a partial, order-dependent sample of their ads and mistake it for their strategy. If you know the advertiser, go through the Page.
Three Query Types, and What Each One Is For
The library exposes three search types, and the distinction between the two keyword modes matters more than it looks.
| Query type | What it matches | When to use it |
|---|---|---|
| Advertiser page | Every ad from one Page ID | Studying one known competitor |
| Keyword, unordered | Ads containing all your words, in any order | Mapping a category or claim |
| Keyword, exact phrase | Ads containing your words in that exact order | Sizing one specific claim or slogan |
Unordered is an AND, not an OR. If you type collagen powder UK, you get ads containing all three tokens somewhere in their text — not ads matching the phrase “collagen powder UK”, and not ads matching any one of them. Long queries therefore narrow fast, and they narrow in ways you did not intend: adding a fourth word often returns four results because the advertiser happened to write “UK delivery” in a different sentence.
Exact phrase respects order. This is the mode for testing whether a specific sentence is common in your category — clinically proven, free shipping over, cancel anytime. It is also the mode that tells you whether a competitor’s line is theirs or an industry cliché: search the line in quotes, and if nine advertisers use it, the line is table stakes and copying it buys you nothing.
There is a practical ceiling worth knowing before you build a query: the Ad Library API caps search terms at 100 characters. The interface will accept more, but the underlying matching does not reward it. Two to four words is the useful range; a sentence is not a query, it is a wish.
The Query Grammar Has No Negation
This is the constraint that shapes every good search, and almost nobody mentions it. The library treats a space as a logical AND and does not support operators. No OR, no minus to exclude, no parentheses, no wildcards.
The consequence is that you cannot subtract an unwanted meaning inside the query. If detox returns a wall of wellness influencers and you want the supplement brands, you cannot write detox -influencer. You have three legitimate ways around it:
- Move the discrimination into the filters. Restrict to one country, one media type, one language. A narrower corpus does part of the work a minus operator would have done.
- Make the phrase more specific.
detox tea 14 daycuts better thandetoxplus a mental filter, because it forces a co-occurrence the wrong advertiser is unlikely to type. - Search, then subtract by hand. Collect the Page names, drop the ones that are not your category, and keep the list. The list is the real output — the ad results are the raw material.
Plan for a noisy first pass. A query that returns 400 results is not better than one that returns 40; it is a query you have not finished writing.
What the Keyword Index Actually Matches
The index reads text. Specifically it reads the ad’s copy — primary text, headline, description — and the advertiser’s Page name. It does not read the image, does not transcribe the video, does not listen to the audio, and does not index the landing page behind the ad.
Three consequences follow, and they are where most search sessions go wrong:
Visual concepts are unsearchable. “Split-screen before and after”, “hand holding product against a kitchen counter”, “creator talking to camera in a car” — none of these can be queried. You can only find them after a text search has surfaced an advertiser, at which point you are browsing again. This is the honest limit of the tool and the reason visual research still requires a human scroll.
The landing page is not part of the ad. If a brand’s copy says “shop now” and the offer only appears on the site, searching for the offer will not find the ad. Search for the words the advertiser was forced to type into the ad itself.
Page name is indexed, so brand-name queries are inflated. Searching a competitor’s brand name returns their ads plus every ad that mentions them — resellers, affiliates, comparison content, and competitors bidding on the name. Deduplicate by Page before you count anything.
One more: language. The index is language-specific in practice. A brand running the same creative in five languages produces five separate text records; a search in English will not surface the German version even though the video is identical. If you are auditing a multi-market advertiser, run the query once per language you care about.
Filter by Filter: What Each One Really Scopes
Filters look like refinement controls. Several of them change the meaning of the result set rather than just trimming it.
Country
The country selector scopes where the ad was shown, not who it targeted. An advertiser running one campaign across the EU appears in every country view, which is why a single advertiser can look like twenty campaigns if you sample country by country.
There is a related distinction inside the data: an ad can be targeted at a country or merely reach it. The library exposes this as a targeted-country toggle — off by default in most sessions. Turn it on when you want ads deliberately aimed at a market, and leave it off when you want the full footprint of a regional campaign.
Ad category
The category dropdown is the most consequential setting on the page and the easiest to leave wrong. “All ads” is the commercial corpus. The other categories — issues/elections/politics, housing, employment, credit — exist because regulators require longer retention for them, and they contain only those ads.
Choose the wrong one and you get a clean, plausible, empty-ish result set that has nothing to do with commerce. It also changes retention: political and issue ads are archived for seven years, commercial ads are not. More on that below, because it is the main reason people think the library “lost” an ad.
Active status
Active, inactive, or all. On commercial ads outside the EU, inactive mostly means gone — the library keeps stopped ads for a limited window for EU delivery and far longer for political ads. So inactive outside the EU is not a history view; it is a thin residue. Do not build a longitudinal study on it.
Inside the EU the picture is better: ads delivered to EU users stay visible for about a year after they stop, with reach data attached. That asymmetry is the single biggest reason EU-based research is easier than US-based research on the same advertiser.
Media type and platform
Media type separates video, image and meme formats — worth doing, because the structural patterns that work in a fifteen-second video do not transfer to a static card, and mixing them makes any comparison meaningless.
Platform covers Facebook, Instagram, Messenger, Audience Network and Threads. Narrowing to one platform is rarely about the platform; it is about removing placements that render the creative differently, so you compare like with like.
The date window is not a launch filter
This is the trap that produces bad analysis. The date range filters on when the ad had impressions, not on when it started. Setting an end date does not mean “ads that launched before that date” — it means “ads that were still delivering on or before that date.”
The practical workaround is to set a start date and read the “Started running on” line on each card yourself. If you want ads that survived at least 90 days, query with a start date roughly 90 days back, then discard every card whose start date is later than that. The filter gets you a candidate set; the card gives you the truth.
The URL Is the Asset, Not the Screenshot
Every filter you set is written into the address bar. That makes a search a reproducible object: bookmark it, paste it into a doc, send it to a client, reopen it next month and get the same query against fresh data.
The parameters worth recognising, because they tell you what the interface is doing:
| Parameter | What it controls |
|---|---|
q | Your search words |
search_type | page, keyword_unordered or keyword_exact_phrase |
view_all_page_id | The numeric Page ID for an advertiser view |
country | Two-letter market code, or ALL |
ad_type | all, or one of the regulated categories |
active_status | active, inactive or all |
media_type | image, video, meme or mixed |
publisher_platforms[0] | Placement filter |
content_languages[0] | Ad language |
start_date[min] / start_date[max] | Impression window |
Two habits follow. First, name your saved links by the question they answer, not by the keyword — “UK collagen claim density” survives six months; “collagen” does not tell you anything later. Second, keep one link per market rather than rebuilding the query by hand each time; country is the parameter people most often forget to reset, and a forgotten country silently changes every number you write down.
Screenshots are still useful, but for a different job: a screenshot is evidence of what an ad looked like, and ads disappear. The URL is the method; the screenshot is the receipt.
Reading a Result After a Search
Once a query returns something, three fields do most of the work — and they are the same three whether you arrived by Page or by keyword.
Library ID. A permanent identifier for the ad. Copy it into your notes. It is the only reliable way to get back to a specific ad, and the only stable key if you are building any kind of sheet. Without it your record is “that green one from the shoe brand”, which is not a record.
Started running on. The launch date, and the closest thing to a performance signal the library will ever give you. Age is not proof of profit — an ad can survive on inertia, on a tiny budget, or because nobody remembered to pause it — but a long-running ad is at minimum an ad nobody chose to kill, which is more than you can say for the ones that vanished.
Variant count. When several ads share one creative and one text block, the card says so and lets you expand the group. This is the most underrated field in the interface: it tells you which concept the advertiser is willing to duplicate across audiences and ad sets. One concept with nine variants is a decision. Nine concepts with one variant each is still a search — and it tells you the category has no settled angle, which is a finding in itself.
If the ad ran in the EU, a fourth block appears: estimated reach, age and gender split, and the audience the advertiser actually chose. Comparing chosen audience against delivered audience is the closest thing to a free media-plan critique — advertisers routinely select a broad band and Meta concentrates delivery somewhere narrower.
A Twenty-Five Minute Search Protocol
The point of a protocol is that it stops you from browsing and calling it research. This one produces a written answer.
- Write the question first. “Which UK supplement brands are running a specific-days claim” is a question. “Look at supplements” is not, and it will produce a folder of screenshots.
- Pick the query type from the question. One advertiser → Page. A claim or slogan → exact phrase. A category you cannot name yet → unordered keywords.
- Draft the query at two to four words. If the first pass returns more than a few hundred results, add a discriminating word rather than scrolling. If it returns zero, drop to two words and check the country setting before concluding the market is empty.
- Scope deliberately: one country,
All ads, active only, one media type if you are comparing creative. - Read the oldest five cards, not the newest. Newest-first is the default and the least informative ordering, because the newest ads are the ones least likely to survive.
- Log Page name, Library ID, start date, variant count, one-line angle. Five fields, one row per ad, in a sheet — not in a chat window.
- Save the URL under the question it answers.
- Repeat the same URL in two weeks and diff the list. New Page names are entrants; missing ones either stopped or renamed; unchanged rows are the incumbents.
Run this across three related queries and you have a category map that no single browsing session produces.
Three Patterns That Only Search Can Find
Claim density. Search a claim in exact-phrase mode across one market and count distinct Pages rather than ads. Ten advertisers making the same promise means the promise is commodity positioning — matching it is the fastest way to become the eleventh identical ad. One advertiser making it means the claim is either proprietary or unproven, and the difference is usually visible in whether they also show evidence.
Offer vocabulary. Search the words that carry the offer — free shipping, money back, cancel anytime, ships in. What you get is the price of entry in that category, expressed in the language advertisers actually use. It also shows you the phrasing nobody is using, which is often the cheapest available differentiation.
Competitor-of-competitor. Search a theme, collect the Page names, then open each in Page view. You will find advertisers you did not know existed and, more usefully, you will see which of them are running the theme as their whole account versus as one test among forty. That ratio is the difference between a positioning decision and an experiment, and it is invisible if you only ever look at brands you already knew.
A fourth pattern sits on top of the other three: the re-run diff. Because the URL is stable, the same query run monthly produces a crude changelog of a market — who entered, who left, whose angles held. That is a time series built out of one bookmark, and it is the only free way to get one.
Where Search Stops Working
The search layer has hard edges. Knowing them keeps you from over-claiming in a deck.
- No negation and no boolean logic. You cannot express “this but not that” in a query. Every exclusion is either a filter or a manual pass.
- No visual or audio search. The index is text. Creative concepts have to be found by scrolling once you have a shortlist of advertisers.
- No performance data on commercial ads. No spend, no impressions, no clicks, no conversions outside the EU reach block. Everything about effectiveness is inference from age and duplication.
- No long history outside the EU. Commercial ads that stop running largely disappear. A “before and after” study of creative needs a schedule of your own captures, because the library will not hold the before.
- No sorting by age. You read start dates off the cards. At thirty results a page, that caps how much you can do by hand in one sitting.
- One market at a time. A regional campaign has to be re-queried per country, and the same ad will show up repeatedly across those queries. Deduplicate by Library ID or your counts will be inflated.
None of these are reasons to skip the library. They are reasons to stop treating a search result count as a market size.
When the Query Becomes the Bottleneck
Manual search is excellent at answering one question and terrible at answering the same question every week across ten markets. The work is not the reading — it is rebuilding the query, re-scoping the country, and re-copying rows into a sheet, forty times.
That is the layer where a structured database earns its place. TrendTrack indexes 10M+ Meta ads across markets, so a theme can be searched once and compared across countries without re-running the query per market or reconciling duplicates by hand. The same index covers the store side — traffic, catalog, apps — through the free Chrome extension, which is where most people start before deciding whether the full platform is worth it.
Verified 2026 · 20% off Starter, Pro and Business for your first 3 months
The promo is simple enough to state in one line: TrendTrack promo code is WINNING - 20% off all plans for the first 3 months. If you are still deciding whether you need anything beyond the free library, the free-plan breakdown covers what stops working without a paid seat — and if you would rather query the index from your own tools than from a browser tab, the MCP setup is the shorter path.
FAQ
Is the Meta Ad Library search free?
Yes. The interface at facebook.com/ads/library is free and does not require an ad account, a Business Manager, or even a login for standard commercial ads. You can search by advertiser or by keyword across any supported country. Login is only required for certain political and issue ads, which need identity verification. The free tier’s limits are analytical, not financial: no spend, no impressions and no click data for ordinary commercial ads.
What is the difference between keyword and exact phrase search?
Unordered keyword search returns ads that contain all of your words in any order, so collagen powder also matches an ad that says “powdered collagen”. Exact phrase search returns only ads where the words appear in that exact sequence, which is what you want when you are testing whether a specific claim or slogan is common in a category. Use unordered to map a market and exact phrase to size a single message.
Why do my search results include ads that do not contain my keyword?
Three usual causes. The index also matches the advertiser’s Page name, so a brand whose name contains your term will appear with unrelated copy. Ads can match on a substring or a stemmed form of your word. And if the ad runs in several languages, the version you are seeing may not be the version that matched, because each language variant is indexed as its own text record. Check the Page name before assuming the search is broken.
Can I exclude a word from a Meta Ad Library search?
No. The query grammar treats a space as a logical AND and supports no OR, no minus and no parentheses, so exclusion has to happen outside the query. The workable substitutes are narrowing the country, language and media type filters, making the phrase specific enough that the wrong advertiser would not type it, or collecting Page names and removing the irrelevant ones by hand before you analyse anything.
Why do I see different results when I change the country?
The country selector scopes where an ad was shown, not who it targeted. One Europe-wide campaign therefore appears in every country view, while a campaign aimed at a single market appears in only one. A separate toggle distinguishes ads targeted at a country from ads that merely reached it, and leaving that toggle off is the usual reason a regional campaign looks larger than it is.
Can I search ads that have stopped running?
Partly, and it depends on where they ran. Ads delivered to the EU stay visible for roughly a year after they stop, with reach data attached. Political and issue ads are retained for seven years. Ordinary commercial ads outside the EU largely disappear once they stop, so the inactive view is thin rather than historical. If you need a record of a competitor’s creative over time, capture it on a schedule instead of assuming the library will keep it.