Most "Meta Ad Library guide" posts are reheated screenshots of the search bar and a closing line about "staying ahead of the competition." They never tell you the part that matters: the Ad Library is a transparency tool, not a research tool, and the difference is why you keep getting frustrated with it. This is the reference I wish existed — what Meta actually exposes, what it deliberately hides, how the two APIs really behave, and where the gaps are wide enough to drive a product through.
This is the guide to the Ad Library itself. If you want the founder's playbook for spying on DTC competitors, that lives in a separate piece on the blog. This one is about the tool and its limits.
What the Meta Ad Library actually is
The Meta Ad Library is a public, searchable database of every ad currently running across Facebook, Instagram, Messenger, and the Audience Network. You can reach it at facebook.com/ads/library without an account. For most ads you see the creative, the advertiser, the platforms it runs on, and the date it started. For one category — ads about social issues, elections, or politics — you also get spend ranges, impression ranges, and rough demographic targeting.
That asymmetry tells you everything about why this exists.
Why Meta built it (and why that shapes everything)
Meta did not build the Ad Library so you could swipe your competitor's hooks. It built it under regulatory pressure after 2016, when nobody could audit political advertising on the platform. The EU's Digital Services Act later codified the expectation: very large platforms must keep a public ad repository. The Ad Library is a compliance artifact dressed up with a friendly search bar.
That origin explains every quirk you're about to hit. Political ads get spend and impression data because regulators demanded accountability for who buys influence. Commercial ads — the Gymshark and Ridge Wallet ads you actually care about — get the minimum: the creative, the start date, and the platforms. Meta exposes enough to satisfy "transparency" and not one field more. Once you internalize that the tool was built for auditors and not for marketers, you stop being surprised by what's missing.
What it exposes vs. what it hides
Here is the honest field map for a normal commercial ad in 2026.
| You CAN see | You CANNOT see |
|---|---|
| The ad creative (image, video, copy) | CTR, CPC, CPM |
| The advertiser / Page | Clicks, conversions, revenue, ROAS |
| Platforms it runs on | Exact spend (commercial ads) |
| The date it started running | Exact impressions (commercial ads) |
| Whether it's active or stopped | Targeting and audience |
| Multiple active versions of an ad | Any historical record of past ads |
| Spend/impression ranges (political/issue ads only) | Why one ad outperforms another |
Burn this into memory: the Ad Library has no performance metrics for commercial ads. None. If a third-party tool shows you a competitor's exact ROAS or CTR, it invented the number — those figures live only inside the advertiser's own account and Meta has never exposed them. Honest research starts by admitting what the data can't tell you.
How the API actually works (the part nobody writes)
Most people only use the website. There's also the Ad Library API, and it behaves in ways that will trip you if you assume it's a normal REST endpoint. Five things you need to know:
1. It's two endpoints, not one. There's the data API (ads_archive) that returns structured fields, and there's a separate snapshot/render URL for the actual creative. The data API does not hand you a clean image or video URL — it gives you a ad_snapshot_url pointing at a JavaScript-rendered page. Getting the real creative means going through the render path, not the JSON. People building their first scraper always miss this and wonder why they have metadata but no images.
2. Spend and impressions are always ranges. Even where they exist, you get spend.lower_bound / spend.upper_bound and the same for impressions — never a point value. A range like "$10K–$15K" is the most precise number Meta will ever give you. Design around ranges or don't bother.
3. There is no history. This is the single most important limitation in this entire guide. The API returns the ads that are live right now. There is no endpoint for "what was this brand running in March." When an advertiser turns an ad off, it falls out of the active result set and Meta keeps no public record you can query. If you didn't capture it, it's gone. The history is not a feature you're missing — it doesn't exist, full stop.
4. You can't sort by run-time. There's no "order by longest-running." You can't ask the API for proven winners. You get an unordered (effectively newest-leaning) page of results and have to compute longevity yourself from ad_delivery_start_time. The most useful signal in the whole dataset is the one Meta refuses to sort by.
5. The website is a single-page app. Scraping the public site means waiting for client-side JavaScript to render. A naive curl returns an empty shell. The site is built to be looked at, not queried — which is exactly what you'd expect from a compliance tool.
One more operational reality: access tokens expire every 60 days, regional results vary (an ad can appear under one country and not another, so serious ingestion tries many countries with fallback), and ad_delivery_stop_time === null is how you know an ad is still active. None of this is documented in a way that survives first contact.
How to actually use it well
Despite all that, the Ad Library is genuinely useful if you use it for what it's good at. A workflow that respects the tool's nature:
Read run-time as your performance proxy. You can't see ROAS, but you can see how long an ad has been live. Brands kill losers fast — nobody keeps paying to run an ad that doesn't convert. An ad that's been live 100+ days is a proven winner, even though Meta will never label it that way. This is the closest thing to a performance metric the dataset contains.
Read the distribution, not one number. A single old ad is noise. A brand with twenty ads all running 90+ days has a system. A brand churning fifty new creatives a month is testing aggressively. The shape of the run-time spread tells you their strategy.
Check version counts. When you see multiple active variations of the same ad, that's the brand A/B testing copy or thumbnails. The variants that survive are the ones working.
Snapshot continuously. Because there's no history, the only way to ever have history is to start recording today. Daily snapshots turn the live-only API into a longitudinal dataset — first-seen dates, churn, iteration rate. This is tedious by hand and the entire reason tooling exists.
Treat spend ranges as direction, not truth. A range tells you order of magnitude. Pair it with engagement and run-time to triangulate; never quote the midpoint as a fact.
The real limits
To be blunt about the ceiling: the Ad Library shows you what competitors are running and roughly for how long. It cannot tell you why it works, who it's targeting, or how well it converts. It has no memory. It can't sort by the one field that matters. It serves creatives through a rendered SPA that fights automation. And it actively withholds every performance metric for commercial ads — by design, because it was built for regulators, not for you.
These aren't bugs. They're the boundary of a transparency mandate. Pretending otherwise is how bad research gets made.
The tools that fill the gaps
This is where the ecosystem comes in, and it's worth being fair about who does what. Foreplay is a strong swipe-file gallery for saving and organizing creative. Atria and MagicBrief lean into ad ratings and creative-brief analysis. AdSpy.com and BigSpy run large, broad databases across platforms. Motion analyzes your own ad accounts (complementary, not competitive). PiPiADS and Minea skew TikTok. Each solves a slice. Check their sites for current pricing — it moves.
AdWhispr fills a specific gap: the history and the derived intelligence the API can't give you. You paste a Facebook URL; it ingests that brand's entire Meta ad library and snapshots it daily — so the history Meta doesn't keep becomes yours. Then you interrogate it by chat or inside Claude via an MCP server. The signals it surfaces are the ones you'd otherwise compute by hand: days-running winners read as a distribution, engagement-verified reach (impression range cross-checked against scraped likes and comments), creative-iteration rate from first-seen dates, and spend triangulation that narrows Meta's absurdly wide range using those other signals — with every input cited.
What it refuses to do is the tell. It will never show you a competitor's exact ROAS or CTR, because those numbers don't exist outside the advertiser's account. Any tool that displays them made them up. AdWhispr cites its inputs for every signal instead — and that honesty is the whole point. It's also read-only on competitor data: it reads the public library, it never touches anyone's live ad account.
The Meta Ad Library is a flashlight pointed where regulators aimed it. Knowing exactly where the light stops is what separates real competitive research from screenshot-collecting.
Want the history Meta won't keep and the run-time math done for you? Start with AdWhispr and turn the live-only library into a dataset you can actually interrogate.