UGC questions › Testing ads on Meta
TL;DR
Your industry decides how long a UGC ad runs far more than anything in the ad itself. We tracked 1,368 Meta video ads from 70 advertisers. 87% of local-service ads were still running after 90 days, but only 43% of app ads were. Short videos, short text, more versions, creator-handle ads and question hooks made no clear difference.
Key findings
TL;DR: Our first study looked closely at 365 ads. This one tests popular rules about long-running ads on a much bigger set.
Our first longevity study coded 365 creatives by hand for format and hook. It left most of the data unused. The Ad Library records every ad's start date, stop date, landing link, call-to-action button, copy and video length, and nobody we could find had tested the common rules for long-running ads against those fields with proper controls.
Those rules show up across most UGC advice pages: keep videos under 15 seconds, keep copy short, launch lots of variants of each concept, run ads through creator handles, open with a question and send traffic to Messenger. They're usually stated as fact, and none of the pages we reviewed shows a sample size behind them.
So we tested eight of them. We also tested our own pilot findings from October, and most of those failed too. We report that below, because a null result is still a result.
TL;DR: We used Meta's public ad library. We tracked ads from launch, including ads that were turned off, so we could see how long they really ran.
We used two datasets, both from the Meta Ad Library and both US-only.
| Dataset | What it holds | Size | What it can answer |
|---|---|---|---|
| Launch cohorts | Every video ad that started in two windows (March 15 to April 20, 2026 and June 1 to July 12, 2026), including ads that later stopped | 1,368 ads, 70 advertisers | How long ads run before they're switched off |
| Market snapshot | Every active video ad across 54 market searches in late October and early November 2026 | 25,697 ads, 10,209 advertisers | How old the ads running right now are |
The cohorts are the important part. Because they include ads that were switched off, we can measure survival directly instead of only looking at the winners still running. Of the 1,368 ads, 571 had stopped by October 20, 2026 and 797 were still running.
We measured run length in calendar days, from the start date to the stop date, or to October 20, 2026 for ads still running. We used Kaplan-Meier survival estimates, which handle ads that are still running without pretending they stopped. Every confidence interval resamples advertisers, not ads, because ads from the same advertiser behave alike. We also ran each factor through a Cox model three ways: raw, with industry and advertiser size controlled, and within the same advertiser.
Ad Library run length tells you an advertiser kept an ad on. It doesn't tell you the ad was profitable, because Meta shows no spend or results for US commercial ads (Mockingbird explains what the start date does and doesn't show). We treat a long run as a sign that the advertiser saw enough to keep paying for it, and nothing more.
TL;DR: About 6 in 10 ads in our group were still running after 90 days. Our group leans toward big advertisers, so treat these numbers as a way to compare groups.
77% of the cohort ads were still running at 30 days, 63% at 60 days, 60% at 90 days and 57% at 180 days.
| Days since launch | Share still running | 95% CI |
|---|---|---|
| 30 | 77% | 67% to 86% |
| 60 | 63% | 50% to 78% |
| 90 | 60% | 46% to 74% |
| 180 | 57% | 43% to 71% |
These figures run high for one reason. Our advertiser panel was hand-picked and leans toward large, established accounts, and big advertisers keep ads on longer than the average account. Use them to compare groups, which is what the rest of this article does. Don't quote them as "60% of UGC ads last 90 days".
The live market tells a similar story from the other side. Of 25,697 video ads running across 54 markets, 29% were 90 days old or more, 11% were 180 days or more and 3.7% had been running for a year or longer. The median live ad was 38 days old. About one live ad in 27 is a year old.
TL;DR: Local services, like home repair and clinics, keep ads running the longest. Apps and software turn ads off the fastest.
Local-service advertisers keep video ads running about twice as long as app and SaaS advertisers, and this was the one result that held up under every check we ran.
| Industry (launch cohorts) | Ads | Advertisers | Still running at 90 days | 95% CI |
|---|---|---|---|---|
| Local services | 151 | 13 | 87% | 69% to 96% |
| DTC and ecommerce | 469 | 21 | 64% | 37% to 92% |
| Info products and coaching | 344 | 18 | 49% | 22% to 79% |
| Apps and SaaS | 301 | 15 | 43% | 24% to 67% |
Local services beat apps and SaaS by 45 points (CI +14 to +66) and info and coaching by 39 points (CI +4 to +67). The gaps between ecommerce, coaching and apps are real in the table but too uncertain to rank. Finance advertisers reached 91% at 90 days, but only 3 of them were in the cohort, so we leave them out.
The live market data points the same way with far more advertisers behind it.
| Industry (live ads, 54 markets) | Live ads | Advertisers | 365+ days old | Median age |
|---|---|---|---|---|
| Home services | 2,324 | 1,370 | 5.5% | 60 days |
| Dental, clinics and telehealth | 3,908 | 1,995 | 5.4% | 54 days |
| Travel, events and media | 1,685 | 926 | 5.2% | 32 days |
| Home, lifestyle and apparel (ecommerce) | 3,820 | 1,454 | 4.0% | 36 days |
| Coaching, courses and creators | 3,014 | 1,174 | 3.5% | 37 days |
| Supplements and wellness (ecommerce) | 2,455 | 918 | 3.2% | 40 days |
| Finance, lending and accounting | 2,129 | 930 | 2.6% | 31 days |
| B2B services and agencies | 1,778 | 822 | 2.6% | 28 days |
| Food and drink (ecommerce) | 1,653 | 644 | 2.5% | 39 days |
| Apps and SaaS | 2,931 | 1,174 | 1.8% | 31 days |
| All markets | 25,697 | 10,209 | 3.7% | 38 days |
A home-services ad running today is about twice as old as a B2B agency ad, and twice as likely to have passed a year. Our industry guide and home services page cover why. Most of the reasons come down to audience size: a roofer in one metro area has a small, slow-changing audience and one offer, while an app is chasing new users across the whole country with constant feature changes.
TL;DR: No clear rule. Very long videos looked weaker at first, but the gap went away once we compared ads in the same industry.
Not as a general rule. Under-15-second videos reached 90 days 65% of the time, against 50% for videos of 60 seconds or more, but the confidence interval on that gap runs from -17 to +44 points.
| Video length | Ads | Advertisers | Still running at 90 days |
|---|---|---|---|
| Under 15s | 197 | 28 | 65% |
| 15 to 30s | 301 | 42 | 67% |
| 30 to 60s | 459 | 52 | 66% |
| 60 to 120s | 204 | 37 | 59% |
| 120s and longer | 137 | 19 | 37% |
Only the longest videos stand out, and that doesn't survive controls. Raw, a video over 120 seconds was 2.27 times as likely to stop on any given day as a 15 to 30 second one. With industry and advertiser size controlled, that falls to 1.34, with an interval that includes no effect. Within the same advertiser it falls to 1.00. The long videos come mostly from coaching and app advertisers, who switch everything off sooner.
App and SaaS advertisers show the opposite pattern. Their under-15-second ads reached 90 days 27% of the time (64 ads, 7 advertisers), against 45% for their 60 to 120 second ads. Our length guide covers how to pick a length for your offer instead of following a single rule.
TL;DR: Long text looked worse at first. But most long-text ads came from coaching brands, which turn off ads sooner anyway.
The raw numbers say yes and the controlled numbers say no. Ads with 600 or more characters of primary text reached 90 days 43% of the time (220 ads, 25 advertisers), against 79% for ads with 300 to 449 characters (164 ads, 28 advertisers).
That gap is mostly industry. 41% of coaching ads use copy of 600 characters or more, against 3% to 13% in every other industry, and coaching ads stop sooner whatever their copy length. Within coaching, long and medium copy performed about the same (49% against 44%). Within the same advertiser, the Cox estimate was 0.70 with an interval from 0.22 to 2.18, which means we can't tell the two apart. Ecommerce showed a large gap in the same direction, but it rests on 6 and 10 advertisers and its interval crosses zero.
TL;DR: No. Making lots of versions of one ad, or launching it again later, didn't help it last longer.
No. Concepts with 10 or more versions (same advertiser, same opening copy, same video length) reached 90 days 32% of the time, against 68% for concepts run as a single ad. Within the same advertiser the difference disappears (Cox 0.95, interval 0.64 to 1.42).
Relaunching a concept, which we defined as the same concept starting again more than 30 days after its first version, didn't help either: 53% of relaunched concepts reached 90 days against 63% of the rest, with an interval from -30 to +12 points.
The variant counts come with a catch. We counted versions at the October snapshot, and an ad that survives longer has more time to collect versions, which blurs cause and effect. What the data rules out is the simple claim that launching many variants of one concept makes it last. The ad fatigue guide covers when a fresh concept beats a new version of an old one.
TL;DR: We found no clear difference. Our sample may be too small to spot a small one.
We couldn't find a difference. Partnership ads, the ones Meta flags as branded content run through a creator's handle, reached 90 days 54% of the time (166 ads, 16 advertisers), against 61% for brand-page ads (1,202 ads, 69 advertisers). The interval runs from -40 to +28 points.
Among the 15 advertisers that ran both kinds, partnership ads looked shorter-lived (47% against 70%), but that interval also crosses zero. All this sample can say is that partnership ads don't obviously last longer, and it may be too small to see a modest effect in either direction.
TL;DR: No. Year-old ads open the same way new ads do, and where the ad sends people made no clear difference.
The opening line doesn't separate year-old ads from new ones. We took the first spoken sentence from 1,322 ads with real speech in the market data and compared ads running a year or more with ads under 90 days old.
| Opening pattern | Share of 365+ day ads | Share of under-90-day ads | Difference (95% CI) |
|---|---|---|---|
| Opens with a question | 21% | 18% | +3 (-4 to +10) |
| Opens with "if you" or "your" | 8% | 13% | -5 (-10 to 0) |
The "if you" gap shrinks to -4 points, with an interval that crosses zero, once we compare like for like within industry. Numbers in the first line, "POV" openers, first-person openers and openers that name an audience all sat inside the noise too.
The landing link didn't explain much either. In our October pilot, year-old ads looked far more likely to send people to Messenger (9%), an Instagram profile (7%) or nowhere at all (14%). That didn't hold up in the full data, where year-old ads were 2.1% Messenger, 2.1% Instagram profile and 4.9% link-less, and each of those groups rests on one or two advertisers. Call-to-action button type, emoji, question marks, prices in the copy and urgency words like "ends today" also showed nothing that survived within industry or within advertiser.
TL;DR: Compare your ads with others in your own industry. Treat ad "rules" as ideas to test, not facts.
Compare your ads against your own industry, and against your own account's history. The difference between a home-services account and an app is larger than any creative rule we tested.
TL;DR: Ad run time shows an advertiser kept an ad on. It can't show sales. Our group of 70 advertisers is also small.
You can download the tables for every cut from our study data page.
On Google
Add us as a preferred source and Google shows our UGC ad research more often in Top Stories and AI answers.
Book a 30-minute call. We review your current ads and your offer and see if we're a fit. No retainer required.
Book a Call