What we found in 1,027 Australian trade video ads

What we found

On one day in September I pulled every trade and home services video ad running in Australia out of the Meta Ad Library↗. Fourteen keyword sweeps, from retaining walls and concreting through to turf, roofing and carports. 1,027 ads, all of them live at the moment I looked. Then I took the 42 that had been running longest, downloaded the video, measured the runtime and counted the cuts, read every frame of the opening and closing three seconds, and transcribed the audio.

Nearly all of them are the same ad. There's a six-beat script sitting under the category, it lands in under 30 seconds, and 19 of the 42 follow it almost exactly. Four businesses in four states are running the same script, two of them word for word and two with a single noun swapped.

The thing I went looking for and didn't find is production value. Owner talking on camera, scripted voiceover over B-roll, captions over B-roll with nobody speaking at all: the three common formats sit at 112, 156 and 165 days of median running time. The owner's face is the shortest-lived of the three. Format isn't what separates them. What they have in common is the script, and that's the whole article.

One category is also empty. Not one Australian business with "retaining" in its name has a video ad that's been running longer than 90 days. Landscaping has 140 ads past that mark and concreting has 51.

How the ads were picked, and what days running means

The sweep was the Ad Library's own filters: Australia, all ads, video, active. Fourteen keywords, one after another, everything they returned. That's where the 1,027 comes from. The corpus numbers below are out of 1,001, because I set aside the 26 that run ten minutes or longer, on the grounds that a ten-minute video isn't competing for the same three seconds. Every ad in the sweep had a readable duration. Nothing was dropped for being unreadable.

Then the important limit, and it's worth understanding before you read a single number. The Ad Library gives a non-political advertiser no spend, no impressions and no click-through rate. The only performance proxy available is how long an ad has been running, and every figure here is days since the ad's stated start date, with all of them still live on the day I looked.

An ad running at 200 days is one somebody keeps choosing to pay for. It's not proof the ad is profitable. A business that stopped checking its account eighteen months ago also produces a long-running ad, and there's no way to tell those two apart from outside. So read longevity as evidence of survival rather than success, and apply that discount to every comparison in here, including the ones I'm most confident about. And the blind spot isn't only on the outside looking in. Meta and Google never find out which click turned into a paid job either, not unless somebody sends that back to them, which is what teaching Google what a good customer looks like→ is about.

The 42 were hand-picked for longevity as well, which makes the 42-ad tables a survivors' sample by construction. Where a claim could be checked against the whole corpus instead I've done that and said which one it is.

The six beats, in order

Nineteen of the 42 run this structure almost to the second, and that group spans 45 to 301 days. The beats are what's constant. The timings below come from the 25 to 30 second version, and the compressed version of the same thing exists at half the length.

One caution on beat 5 before you use it. A cap on intake is a representation about future supply, and if you'd happily take a sixth job after saying you're only taking five, that's the sort of claim the ACCC treats as misleading. ACCC on false or misleading claims↗ It costs you nothing to say the true number instead, and a true number is usually a better line anyway. Nobody believes five.

1. Address and problem, 0 to 3 seconds

   Name the buyer by suburb or by symptom before anything else happens. "Is your yard starting to shift or wash away when it rains?" runs at 208 days. "Is your deck starting to rot, sag or feel unsafe?" runs at 202. GreenBuilt in Canberra opens on the word Canberra and has been running 301 days.

2. The named remedy, 3 to 9 seconds

   The thing that fixes it, named plainly, with the business attached to it. Not a benefit, a noun. "A properly built retaining wall can level your space and protect your property."

3. Scope and area, 9 to 17 seconds

   What they do and exactly where, with the suburb list read out loud rather than left in the body copy. "Across Sunbury, Melton, Keilor, Macedon Ranges and Mickleham." Somebody in Keilor hears their own suburb, which is not the same as reading it in the caption underneath.

4. Why us, in three words, 17 to 22 seconds

   Three short proofs, and almost always process rather than credential. "We show up, communicate clearly and get the job done right." Nobody recites a licence number.

5. The constraint, 22 to 26 seconds

   A cap on intake, a deadline, or scarcity. "We're only taking on the next five jobs" appears in four separate ads in the set, all of them past 90 days.

6. A mechanical CTA, 26 to 30 seconds

   Name the button and the action out loud. "Tap the Learn More button below to book a free design consultation." Not "get in touch". The name of the button.

Four businesses, one script, four states

Here's the part I didn't expect to find.

"Is your yard sloping, cracking, or becoming a safety concern? A properly built retaining wall can level your space and protect your property. Our team will get it done right, but we're only taking on the next five jobs. If you want it done properly, tap the learn more button below to book a free design consultation."

That's Precision Earth Projects, 150 days running, at 17.0 seconds. It is also Masters of Outdoor Living in Sydney, 111 days running, at 16.2 seconds, identical to the word. I checked it as a token set rather than by eye and the overlap is total. South Crete Concreting runs the same script at 156 days with "shed slab" in place of the wall, and Mighty Concrete runs it at 98 days with "driveway". Four businesses, four states, one script, every one of them past 90 days.

I don't think any of them copied another. The likelier explanation is a shared supplier or a script template doing the rounds, and that's both the fairer reading and the more interesting one, because it means somebody has already tested this structure across four trades and four markets and it keeps surviving in all of them.

Which cuts two ways. It's the strongest evidence in the study that the skeleton is real and not a pattern I imagined myself into. And it's the reason to use the skeleton and not the words. If you run that script as written you're the fifth business in a template pool, competing in the same feed with the same sentences, and at least one of the other four sells the thing you sell.

Under 30 seconds, and no cliff after it

The 15 to 29 second band is the one that survives. 122 of the 291 ads in it were still running at 90 days, against 46 of 146 for anything shorter and 64 of 207 for the 30 to 44 second band. Median running time in the band is 67 days, which is the highest of the six.

Now the part that argues against the usual advice. There's no cliff at 45 seconds. The decline is gradual, and the 90-second-plus band is the second best performer of the six at 39 of 100.

What I can't tell you is what those long ads are doing to earn it, and I'd rather say so than guess. Only 2 of the 42 I pulled apart run 90 seconds or more, so the band that does second best is the one I've barely read. The unbroken single take, for what it's worth, lives a band lower: Concrete Logan Experts has been running 188 days on 58 uninterrupted seconds of a man screeding a slab, and an earthmoving ad runs 49 seconds in one shot from the cab of a digger.

So set your ceiling at 30 seconds because that's the optimum, not because longer is fatal. What happens above 90 seconds is a gap in this study rather than a recommendation.

Open on the problem, not on your own face

Twelve of the 42 open on a symptom question, and that group has the highest median of any opening with enough ads behind it to mean anything: 153 days. Seven open by naming a place, at 147 days. Eight open on the owner introducing himself, and that group is the weakest of the three at 111.5 days.

Across the whole corpus the same thing shows up much harder. "Tired of" and "sick of" appear in 97 ads and 67 of them are past 90 days, against 345 of 1,001 for the corpus as a whole. That's the most durable opening construction in the category and it's a problem hook.

Then the detail that makes the point better than any of the numbers. The longest-running ad in the symptom group is Odd Job Todd at 308 days, and it sounds like an owner introduction. Todd says his name and that he owns the place. He says it at 0:04. At 0:00 the ad asks whether you're sick of outrageous pricing, which is why it's coded as a problem opening, and that four-second gap is the difference between the strongest group with real numbers behind it and the weakest. The eight ads that genuinely start with the owner's name and face sit at 111.5 days.

Nine of the 42 open on a person, 8 of them the owner introducing himself and one a customer. The other 33 open on something else: a symptom, a place, an offer, the finished job, a failure, a tease, a credential. Your face isn't the problem. Your face at 0:00 is.

Small-sample warning on this section, because it's the softest thing in here: the hooks were coded by hand on 42 ads, two of the nine categories have a single ad in them, and I wouldn't defend the ranking below the top three. The corpus-wide "tired of" number is the one to trust.

What production value does not explain

The three common delivery formats, by median days running: captions over B-roll with no voiceover at all, 13 ads, 165 days. Scripted voiceover over B-roll with no face, 13 ads, 156 days. Owner speaking on camera, 12 ads, 112 days.

Owner on camera comes last. It also holds the second-longest ad in the entire set, a father and son standing in front of a fence at 510 days, and the spread inside every one of those groups is enormous. Twelve ads per cell and no spend data doesn't let anybody claim a format effect, and the honest conclusion is that there isn't one to claim. Format doesn't separate the survivors in this data. Which isn't the same as showing it doesn't matter: 12 ads a cell with no spend figures couldn't detect a real difference if there was one.

What the survivors have in common is the script. Which cuts against the advice going around, though not as far as I'd like: nothing in this data says rough wins, and nothing in it says structure wins either. What it says is that every one of these ads is rough, and the ones still running share a structure.

The polish, for the record, is almost non-existent. All 42 are 720p or smaller, one of them 480 pixels wide, and nothing in the set is 1080p. 38 of the 42 are taller than wide, 35 of those exactly 9:16, and the other 4 are square. Not one is landscape. One roofing ad has been running 103 days with an un-removed CapCut watermark burned into its final frame, complete with the export ID, which means it was cut on a phone in a free app by somebody who never checked the file, and it's outlasted most of the category.

The shot rhythm splits clean in two with almost nothing in the middle. Either a single unbroken take, or a montage cutting every 1.2 to 2.6 seconds. And the B-roll is the same three things in the same order nearly every time: machinery, then hands, then the finished thing. Augers going into pier holes, a screed pulling across wet concrete, a boot on a rotten deck board, then the wide shot of the finished job with the house behind it. Two ads in 42 use a drone. None of them use a gimbal you could detect.

All of which is inside reach of a phone and an afternoon.

What the long-running ads never do

The absences are sharper than the patterns, and they're easier to act on.

Nobody says what the job costs. 128 of 1,001 ads contain a dollar figure and nearly all of them are a price on a commodity service, a roof clean or a driveway seal, rather than the price of a build. Exactly one ad in the corpus leads on the price of a build, and it's three weeks old, which is far too young to read as anything. Related: you might be advertising for price shoppers→.

Nobody claims to be the best. Unqualified superlatives turn up in 43 ads, 4.3% of the corpus, and none of the long-running specialists use one. Nobody talks about scale either. No team-size boasts, no jobs-per-week, no growth story. The ads that keep running are small-business ads and they sound like it.

Nobody explains the engineering. Not one of the 42 mentions concrete strength, pier depth, drainage detail or geofabric. The closest any of them get is a four-noun process list, "excavation, footings, reinforcement and a clean, durable finish", which is a rhythm rather than a technical argument. If you've been sitting on the belief that your buyer wants the spec, 42 of the longest-running ads in your category disagree.

Nobody says what they won't do. Scope in these ads is always an inclusion, the list of work they take on, and never a refusal, which is the opposite of what the jobs you don't want are the ones you're asking for→ argues and is the one genuinely untested gap in the whole pattern.

Nobody mentions a competitor, by name or by implication. And almost everything is captioned on screen, including the ads that also have a voiceover. 13 of the 42, close to a third, have no voiceover whatsoever and carry the entire message in burned-in text. Which means a script still matters when nobody says a word of it out loud.

Where the offer lands, and what the last three seconds do

The offer is late. In the skeleton ads the constraint lands at 22 to 26 seconds of a 30-second ad, and it's almost always a cap on intake rather than money off. Scope, on the other hand, is early: what they do and where they do it is settled by 17 seconds.

The most-copied close is the free quote, and it isn't close. That phrase turns up in 132 ad bodies across the sweep and 50 of those are past 90 days. Design consultation turns up in 11, and 3. So the interesting thing isn't that the consultation wins, because it doesn't. It's that the one script four businesses are running closes on a design consultation instead, which puts it deliberately in the minority. I'd take that ask over the quote myself, because a quote is a number and asking for a number invites a comparison of numbers. That's a preference though, not something this data settles.

Then the single most copyable craft detail in the whole study, and I haven't seen it written down anywhere. Four of the 42 end by drawing the lead form. They overlay a mock-up of the Meta instant form, the empty boxes for full name, email and phone number, over the final shot of the finished job, with an arrow pointing at the button. They're teaching the viewer what the next screen looks like before they tap. It costs nothing and almost nobody does it.

On where the ads point: across the 327 trade ads running 45 days or longer it's the business's own website 55%, the Meta instant form 31%, Messenger 9% and an Instagram profile 6%. Among the ads past 180 days the instant form share rises to 36%. Worth saying, because the instant form gets treated as the downmarket option and in this category it's what a third of the survivors use. It does need setting up properly, which is a separate matter and the lead form settings that decide whether your leads are real→ covers it. The 9% pointing at Messenger is the one I'd argue with, for reasons in a Messenger conversation isn't a lead→.

And the buttons, since somebody always asks: Get quote and Learn more are tied on 84 uses each, well clear of See details on 43 and Send message on 16.

Nobody has won the retaining wall category

Retaining walls has the lowest survival rate of the nine categories in the sweep. 116 ads mention retaining walls and 22 of them are past 90 days, and when you look at who those 22 belong to it's landscapers, pool builders, fencers, a pressure-washing business and a couple of firms selling marketing to tradies.

Seven Australian businesses in the sweep have "retaining" in their name. The longest-running ad among them is JB Retaining and Earthworks at exactly 90 days. The next is Retaining Wall Group Vic at 48. After that it's 18 days, 12, 10, 9 and 8. The longest-running ad from anyone positioned on retaining walls at all is Masters of Outdoor Living in Sydney at 111 days, and that's a general landscaper running the same script as the other three.

Compare that to next door. Landscaping and outdoor living: 140 ads past 90 days. Decking and patios: 65. Concreting: 51. Turf: 49. Fencing: 43. The neighbours are crowded and this category is bare.

Two young ads in it are worth watching, both under seven weeks. One opens on the wait for a quote rather than on walls at all, promising a same-day callback, which is a hook about the buying experience instead of the product. The other tells the failure as a slow technical reveal: that crack didn't appear overnight, it started as a hairline. That's the same material as a scare ad, played completely straight, and it's the more premium register of the two.

Somebody else has noticed the gap, incidentally. There are ads in the corpus running at retaining wall builders, selling them marketing. So the window is open rather than unnoticed.

What this can't tell you

A day is one day. Every ad here was live on a single afternoon in September, which means the whole study is a snapshot of survivors and says nothing at all about the ads that ran, worked, spent their budget and stopped. Those don't appear in the library as active and they're invisible to this method entirely.

Longevity is the weak link and I'd rather over-say it than under-say it. No spend, no impressions, no result. An ad that's been running 300 days might be a machine printing money or a forgotten line item. The one thing it definitely is, is an ad nobody turned off.

The keyword sweeps overlap, so an ad selling a backyard build can land in three categories at once and the category counts shouldn't be added together. The hook coding is by hand on 42 ads and two of its nine cells hold a single ad each. And the 42 were chosen for age, which is exactly the bias you'd expect to produce a tidy-looking pattern, so the corpus-wide numbers are the ones I'd bet on and the 42-ad tables are the ones I'd treat as illustration.

What survives all of that is the structural stuff, because it doesn't depend on longevity meaning anything: four businesses are running one script, the category films at 720p on phones, nobody explains the engineering, nobody names a price for a build, and a category with 116 ads in it has no specialist ad older than 90 days. Those are facts about what's in the feed, not inferences about what works.

Shoot one this week

None of this needs us and none of it needs a production crew. The version you can do yourself takes an afternoon and a phone.

1. Write the six beats before you touch a camera

   In that order, out loud, with a timer running. If it comes in over 30 seconds, cut the middle rather than the ending. It's the one part of the ad this study can actually describe, and the part every survivor has in common.

2. Start on the problem, not on yourself

   Point the phone at the thing that's failing and ask the question a customer would ask about it. You go in at four seconds, once the problem has earned you the attention.

3. Say the suburb names yourself

   In the audio, not only in the caption. Seven of the 42 open by naming a place, at a median of 147 days.

4. Say your three proofs as process, not as credentials

   How you turn up, how you communicate, how you finish. A licence number is a fact about a filing cabinet.

5. Say a true number

   If your intake is four jobs a month, say four. The most-copied line in this category is a scarcity claim most of the people running it can't honestly make, and yours can be.

6. Name the button

   Out loud, in the last three seconds. Then draw the form over the last shot with an arrow at it, because only four ads in the top 42 do it.

7. Film it vertical on your phone and don't grade it

   Every one of the 42 longest-running ads in this category is 720p or smaller, not one is landscape, and one has a CapCut watermark in the final frame.

The ad is the smallest part of this

Here's the thing the study can't measure, and I think it matters more than the six beats. An ad is a promise made in 28 seconds. Everything that decides whether anybody believes it happens after the tap, and none of it is in the video.

Somebody who stops on your ad doesn't ring you. They check you. The website, the reviews, the Google profile, the photos, whether the number in the header actually dials. That's a professionalism test being run on assets most businesses have never looked at in that order. When I went through 31 of these businesses one at a time, 13 were running two or more versions of their own name, address or phone number across their own website, Facebook and Google, and 5 had a phone or email link that didn't work at all. Most of them I found by stopping on their ad in the first place, which is that study's main limit and also the thing worth noticing here: the spend was live while all of that sat there. The rest of the list is in auditing 31 trade businesses→.

So the ad is one surface. The brand is what makes somebody decide you're the professional one before you've said a word, and the website decides what your business is even allowed to say on your behalf, which is most of template websites vs custom built→. Being findable in the suburb you just read out in the ad is a different job again, and that one's local SEO for tradies→.

Then the part nobody sees. Whether an enquiry becomes a record that gets chased or an email that gets lost. Whether the review gets asked for. Whether the second call happens at all. The practical half of that is five automations every small business needs→, and the honest accounting of what looking after it costs is what is your monthly actually for→.

None of those are separate purchases in any useful sense, and buying them one at a time is how a business ends up with suppliers who've never spoken to each other and a phone that still doesn't ring. Which is the argument for assessing the whole thing at once, and it's why this article can't honestly end at shoot a video. Shoot the video. Then go and look hard at what it's pointing at.

One more thing

The uncomfortable read on all of this is that the template is winning because there isn't much competition for it. Four businesses running one script across four states is a pattern that survives because the thing it's competing against is mostly a logo on a black background, which is what auditing 31 trade businesses→ turned up, and a boosted post that was never written as an ad.

Which means the bar is low and it's also moving. When the fifth and sixth business run the same 17 seconds, that script stops working in exactly the way the first four found out it worked. The skeleton stays useful. The words are already used.

If ads are the part of this that's biting for you, why ads stop working for trade businesses→ comes at the same ground from the other end. If you want the whole method rather than one channel, it's written out end to end in the Foundational Marketing Blueprint→, free, whether you ever hire anybody or not. And if you're weighing up whether to talk to anybody at all, nobody buys from a demonstration→ is why we do the looking before there's a call rather than after it.