OpenAdLibraryOpenAdLibrary
Ad Creative & Funnels

How to Find Winning Ads: A Signals Framework and Its Limits

A strong ad isn't the prettiest one in the feed. This framework shows how to read longevity, spread, variant clusters and landing pages to shortlist ads worth studying, and what those signals can't show.

Concept illustration: a grid of native ad cards with one winning creative highlighted and traced by data lines

Most ad research dies on the same hill: people confuse visible with successful. They scroll a feed, screenshot the ads that look slick, and call it a day. The problem is that the slickest creative in the feed might have burned a test budget and gotten killed last Tuesday. The ugly, text-heavy advertorial sitting next to it might have been running for four weeks.

You cannot tell those two apart with your eyes. The difference lives in the signals: how long the ad has run, how widely it spread, how many variants surround it, and what happens after the click. This is the framework for reading those signals, so you stop collecting screenshots and start shortlisting ads with sustained investment behind them. None of these signals shows profit; they tell you where to look.

The examples below are historical captures from the OpenAdLibrary collection in June 2026. They illustrate the signals; they are not current data or evidence of performance. Separating a long-observed ad from a one-day test needs repeated observations over time, from any honest source.

What "finding a winning ad" actually means#

For an advertiser, a winning ad is one that meets their own targets, such as breaking even or better. From the outside you can't see that. What you can do is filter captured ads by evidence of sustained investment rather than surface appeal: longevity, publisher spread, variant testing and landing-page quality, read together. When those four line up, the ad is worth studying closely; it is still not proven profitable, and the right response is to learn the structure and test your own version, not copy it.

That definition sets the bar on purpose. You are not hunting for inspiration; you are studying decisions other buyers made, from the outside, without seeing their results. Every signal below is a way of watching those decisions.

Four signals to read together#

No single number proves an ad is profitable, and you cannot see anyone's ROAS from the street. Four observable signals, read together, narrow down which ads deserve study. Treat them as triangulation: each is weak alone and more informative in combination, but even together they don't measure profit.

Signal What it measures What it can't show Where to read it
Longevity Days or weeks between first and last observation Spend, profit or continuous delivery; gaps may be capture gaps First-seen vs. last-seen dates
Spread How many publishers and placements carried it Whether wide distribution means scaling, testing or cheap inventory Publisher count, placement diversity
Variant testing Whether the ad is one of many related variations Which variant performed Clusters of near-identical creatives by one brand
Funnel seriousness Whether the landing page looks like a built-out funnel Conversion rate or offer quality The traced landing page / pre-lander

The mistake amateurs make is grading a creative in isolation. An ad worth studying is not just a great image: it is one the advertiser kept showing, on many sites, in many versions, pointing at a funnel they clearly invested in. Read the system, not the picture.

1. Longevity: how long an ad was observed#

Time is the first filter. Many native buyers review results often and pause what isn't working, so an ad observed for weeks has at least survived those reviews. That is useful information, but it is not evidence of profit: brand budgets, slow optimization, small always-on spend and capture gaps can all produce a long run.

Here is an honest caveat about our own data. In June 2026, the longest observed spans in our captures were about 28 days, because that was roughly how far back our capture window reached per creative. The classic "90-day winner" is industry lore, not something we can confirm from the outside. What captures can show is which creatives were seen across the whole observation window; they can't show whether an ad ran continuously between sightings or what it earned.

One example from June 2026: a SmartAsset finance advertorial was observed across the entire window:

SmartAsset IRA tax advertorial running 28 days straight
Caption: SmartAsset's 'Ask a Pro: How Can I Avoid Paying Taxes on IRA Withdrawals?' ad, observed across a 28-day capture window on Outbrain by OpenAdLibrary, June 2026 (historical example)

That is a longevity read: the ad was observed for the whole window, which makes it worth studying. It doesn't show that the campaign was profitable; a well-funded advertiser can run content marketing on a long budget for reasons other than direct return. The practical move: sort by observed run duration first and let time do the initial filtering, then check the other signals. Our ad longevity guide covers the mechanics, including the always-on brand-ad false positive.

2. Spread: how widely an ad was seen#

A test campaign may sit on a handful of placements. An ad seen across dozens or hundreds of publishers has wider distribution, which often means more budget behind it. That can be scaling, but it can also be broad targeting on cheap inventory or a well-funded test, so spread suggests investment, not that the ad works.

Read it two ways. Breadth is the number of distinct publishers carrying the creative. Quality is which publishers: an ad on major news brands and large portals may reflect a bigger or different budget than the same ad on 80 long-tail arbitrage domains. Neither tells you the price paid or whether the back-end math works, so treat spread as a sign of investment, not of profit.

This historical solar ad uses a UK subsidy angle. It was observed over 27 days, which shows persistence, not that the funnel paid:

Solar home battery subsidy native ad, 27 days running
Caption: 'Solar home batteries: Electricians agree about 1 thing' by Solar Battery Subsidy, observed over 27 days on Taboola by OpenAdLibrary, June 2026 (historical example)

If you research home and garden offers, check subsidy and rebate claims carefully: they must match a real program the reader is actually eligible for.

3. Variant testing: discipline is a fingerprint#

Look at a single advertiser's body of work, not one ad. Buyers who test methodically run families of creatives, not one: ten headlines on one image, one headline across five thumbnails, A/B/C/D advertorial intros feeding a single offer. That iterative testing is the operational signature of someone working methodically.

Captures make this pattern visible. In June 2026, the brand "My IQ" appeared repeatedly among creatives observed for the full 28-day window, with slight variations on the same quiz hook: "The Best IQ Test 2025," "What's Your IQ Level? Find Out Now," "Take a 3m quiz to get your IQ," "Start The IQ Test." Same offer, same funnel, a rotating headline matrix, all observed across the window on Microsoft Audience Network. That looks like a test rig. It doesn't tell you which variants won: the ones still running may be winners, or simply the ones nobody paused.

When you cluster an advertiser's creatives and see disciplined variation, you learn that they are testing seriously, because dabblers do not build test matrices. Within the cluster, the variant observed longest and spread widest is the first one to study. It did not necessarily beat the others: you can see what kept running, not the experiment's results.

This is also where you learn how the angle works: which hook kept running, which proof element repeated, which emotional frame was reused. We go deep on decoding those patterns in How to Analyze Winning Native Ad Creatives (Hooks, Angles, Advertorials), and on sourcing fresh angles for your vertical in How to Find Winning Native Ad Angles for Affiliate Campaigns.

4. Funnel seriousness: follow the click#

The ad is half the campaign. The landing page is where the money is made or lost, and it is the half many researchers never look at. An ad pointing to a thin, broken or placeholder page is probably a test. An ad pointing to a polished pre-lander, advertorial and offer sequence with built-out proof elements and visible CRO work points to a funnel someone built to convert, though not necessarily one that does.

Health offers are a good place to practice this. Here are two curiosity-gap hooks from our June 2026 captures (historical examples; their health claims are not verified):

Native health advertorial about hearing devices
Caption: 'Americans Are Ditching Hearing Aids for This New Device' by Nebroo, observed over 26 days on Taboola by OpenAdLibrary, June 2026 (historical example)
Native health advertorial about medications and memory
Caption: 'MDs Identify 10 Medications Now Attached to Memory Problems In Seniors' by Vital Guardian, a Taboola creative observed over 3 days by OpenAdLibrary, June 2026 (historical example)

When captured, the Nebroo creative had been observed for 26 days and the Vital Guardian one for only a few. Neither observation shows how either funnel performed, and both headlines make health claims that would need competent and reliable scientific evidence. An ad library tells you an ad exists; you also need to know where the click goes. For native ads, OpenAdLibrary follows the redirect trail to the landing page where one was captured, without clicking the live ad, so you can read the funnel the way the advertiser designed it.

Funnel quality doesn't tell you whether the offer itself is sound. A great ad on a great-looking page can still sit on top of an offer that does not pay or convert. Separating a genuine converter from a well-dressed dud is its own discipline, covered in Offer Validation: How to Tell if an Affiliate Offer Actually Converts.

The research workflow, step by step#

Signals are the theory. Here is the repeatable process.

Pick the network where your competitors actually buy. Start where your vertical lives, not where research is easiest. Category mixes differ by network, so check which networks carry ads for your offer type; a tool's capture counts describe its own coverage, not a network's market share. Filter by longevity first. Sort captured ads by observed run duration. On your first pass, ignore the day-zero launches and focus on the ads that persisted across the window. You are separating long-observed ads from brand-new tests. Cluster by advertiser. Group those ads by who runs them. The brands with the most long-observed, widely spread creatives (the My IQs and SmartAssets of your vertical) are the ones to study first. Read the variant family. Inside a serious advertiser's cluster, find the longest-observed, widest-spread creative. That is the one they kept investing in; you can't see whether it won. Trace the click. Follow it to its landing page and read the whole funnel: pre-lander, advertorial, offer, proof, CTA structure. Identify the real advertiser. Native ads often route through trackers and intermediaries that obscure who is behind the spend. Resolving that chain to the real operator (not the publisher, not the network) lets you build a full picture of one playbook. Log and re-check. Re-run the same query weekly and watch which creatives persist, which disappear and which new variants enter. Persistence over time is a stronger reason to study an ad, not confirmation that it is profitable.

The same workflow can be pointed at physical and digital products instead of offers, as laid out in How to Find Winning Products with Native Ad Data (2026 Workflow).

Why native ad data is uniquely readable#

Much native advertising is unusually observable compared to walled-garden social, not by regulation but by architecture. Native ads run on the open web across thousands of independent publishers, so the ads readers see are publicly served. OpenAdLibrary collects them from the networks' public placements and follows their redirect trails without clicking live ads. There is no single closed feed hiding the inventory, but no outside collection is complete either.

Regulation is pushing the largest platforms the same way. Under Article 39 of the EU's Digital Services Act, very large online platforms and search engines that present ads must keep a public, searchable ad repository with API access. Enforcement is real: in December 2025 the European Commission fined X €120 million for three DSA breaches, one of which was its advertising repository, and on the same day it accepted TikTok's binding commitments to provide a compliant repository; AliExpress's commitments, including a searchable ad repository, were made binding in June 2025. These rules cover designated platforms in the EU, not the whole industry, but they are making more ad data public.

Common mistakes that produce false positives#

Even with the framework, a few traps catch people.

Grading the prettiest creative. Design polish tells you nothing reliable about profitability on its own. The framework exists because your eyes will mislead you.

Trusting a single signal. A long-running ad on one cheap publisher might be a forgotten always-on placement. A wide-spread ad that just launched might be a well-funded test about to flop. Demand convergence.

Mistaking brand ads for performance ads. A big brand may run a creative for weeks for awareness, not ROI. A carmaker's model-launch ad, such as Honda promoting the City on Taboola, can stay visible for weeks because the goal is reach, not direct response; its run length shows presence, not spend or return.

Honda City auto brand ad on native inventory
Caption: A Taboola finance ad, 'IRS Forgives Millions By June 30th Tax Deadline' by Fresh Start Information, captured by OpenAdLibrary, June 2026 (historical example; the ad's claim is not verified)

Longevity is more informative for direct-response advertisers who review results often, such as the Fresh Start tax-relief creative above, but even there it shows persistence, not profit, and the ad's own claim is not verified here. Brand awareness plays follow different rules.

Copying creatives instead of learning the structure. The win is not the exact image. It is the angle, hook, and funnel logic. Copy the mechanism, not the pixels.

Researching once. A promising ad found today is a snapshot. The real intelligence is in the trend line: which ads persist, spread or disappear over weeks.

Where OpenAdLibrary fits#

This framework works with any honest data source. OpenAdLibrary is one such source, and we sell it. It captures public native ads across Taboola, Teads/Outbrain, MGID, Revcontent, MediaGo, Yahoo and MSN; stores the creative image; classifies the ad-tech supply chain to help resolve the advertiser behind an ad; and follows redirect trails to captured landing pages where it can, all without clicking live ads. Observed run length and spread are shown directly, so long-observed ads are easy to find.

The free plan shows the 20 newest Meta, TikTok and native ads, with ad detail and native redirect trails, no card required. Search, filters, the API and the MCP server require the Founders plan at $29.99/month; the other tools we checked list plans from $79.99 to $399 a month (October 2026). Founders also adds Creative Studio's AI and publishing tools and Copy DNA if you want to move from research to building your own ads. You can explore the underlying capability on our native ad spy tool page.

The framework is the skill. The data is the leverage. Put them together and "find winning ads" stops being a hopeful scroll and becomes a process you run every week.

Start free: see the 20 newest ads, no card required, and start reading the signals; upgrade to Founders when you need search and filters.

Frequently asked questions

What is the single best signal that an ad is winning?
There isn't one you can see from outside. Longevity is the most useful first filter, because an advertiser who reviews results often has at least kept the ad running for weeks. But a long run can also reflect brand budgets, slow optimization, small always-on spend or gaps in capture, so it is a reason to study an ad, not a measure of profitability.
Can you tell if an ad is profitable without seeing the advertiser's account?
No. You can narrow down which ads deserve study by combining signals instead of trusting one: how long the ad has been observed, how many publishers carry it, whether it sits inside a family of variants, and whether the landing page is a built-out funnel. When all four point the same way, the ad is worth studying closely, but its profitability remains unknown.
How is finding winning ads different from just browsing an ad library?
Finding promising ads means filtering by signals (sorting by observed run duration, clustering by advertiser, tracing clicks to landing pages), while browsing just shows you what exists. A framework separates brand-new tests from ads with sustained investment behind them; it still can't show which ones are working.
Do I need a paid spy tool to find winning ads?
Not to learn the signals. OpenAdLibrary's free plan shows the 20 newest Meta, TikTok and native ads, with ad detail and native redirect trails, no card required. Searching and filtering the library (for example by days running), the API and MCP require the Founders plan at $29.99/month, which is what systematic research needs.
Which native networks should I research first?
Start where your vertical lives and your competitors actually buy, not where research is easiest. Category mixes differ between networks, so check which ones carry ads for your offer type; a tool's capture counts per network describe its own coverage, not each network's market share.
The OpenAdLibrary Team
Written byThe OpenAdLibrary Team
Ad intelligence & native advertising research

We build OpenAdLibrary, the open ad-transparency platform. Every day our systems capture live native ads across Taboola, Outbrain, MGID, Revcontent, Teads, Yahoo and MSN, identify the real advertiser behind each one, and follow the click to its landing page. These guides distill what we see in that data so you can research the market faster.