Likely
Creative taxonomy
Stop calling every ad “UGC”
Ask three people on your team to describe the same ad and you will get three answers. One says UGC. One says testimonial. One says creator video. Now try to count how many testimonials you ran last quarter — you cannot, because nobody agreed what one is.
A shared taxonomy fixes that. Give every ad the same five labels and you can compare what you make against what your category makes, month after month — and see the concepts you have never tried.
What you get from this page
- The complete 39-label taxonomy Likely uses to classify competitor ads at scale. Copy it into your own briefs.
- The six pairs everyone mislabels, each with the one test that separates them.
- A check on your own labelling — the number that tells you your list has stopped working.
- Three things to do this week, starting with your brief template.
On this page
- What a creative taxonomy is
- Why the vocabulary has to be closed
- The five axes
- Hook — 11 named hooks
- Format — 11 named formats
- Angle — 9 named angles
- What happens when nothing fits
- Funnel stage — 3 labels
- Awareness — 5 labels
- The six pairs everyone confuses
- How we know when the vocabulary is wrong
- How to use this without Likely
The short version
Label every ad on five things: hook (how it opens), format (what the asset is), angle (what it argues), funnel stage (where it sits) and awareness (what the buyer already knows).
Add those five as required dropdowns in your brief template and you can start counting concepts instead of guessing at them. Every label, with definitions, is below.
What a creative taxonomy is
A creative taxonomy is a fixed list of labels you describe every ad with, so that two ads described the same way really are the same thing. It is the difference between a folder of screenshots and a number you can act on.
You already have one. It just isn't written down. Everyone on your team has private words for what an ad is doing — "the testimonial one", "that pain-point angle", "the spec graphic". The moment two people mean different things by "UGC", nothing you count means anything.
The list below is the one Likely runs on. It has been argued about, measured against real ads, and revised twice when the numbers said it was wrong. That is what makes it worth copying rather than inventing your own from scratch.
Why the vocabulary has to be closed
Use one fixed list. If an ad does not fit it, label it "other" — never invent a new label on the spot. That one rule is what keeps your numbers comparable over time.
Open tag lists drift. Give people freedom and within a quarter you have "UGC", "creator video" and "ugc-style" as three separate tags describing one thing. Now your coverage grid has three thin cells instead of one solid one, and the gap you were looking for is hidden by your own labelling.
The rule is simple: the list is the list. When something genuinely does not fit, it gets labelled "other" — nobody invents a new label on the spot. That costs you a little precision today and buys you the ability to compare one period against another, which is the only thing that turns labels into a trend.
“Other” is a measurement, not a bin. Any fixed list will eventually meet an ad it cannot describe, and saying so is more useful than forcing a fit. What matters is how often it happens. A few ads in a hundred is a list working normally. More than one in ten, and the list itself is missing a word.
The five axes
Every ad gets one label on each of five axes. The axes are independent, so a single ad can score anything on any of them.
Every label below carries a sample ad. Hover one, or open it, and you get the frame with the copy set on it, what you would be looking at, what it does to the person watching, what makes it that label, and the label people pick instead when they get it wrong. The definitions are the rule; the samples are for the calls the rule does not settle on its own — and most arguments about labelling are about those.
The samples are drawn rather than captured. A screenshot of somebody’s campaign would put a real advertiser’s name against a technique they never agreed to demonstrate, and it would be out of date the week they changed creative. None of these carries a brand, because the point is the move, not who made it.
| Axis | Answers | Named labels | Catch-all? |
|---|---|---|---|
| Hook | How the first two seconds earn attention | 11 | Yes |
| Format | What the asset physically is | 11 | Yes |
| Angle | The argument being made | 9 | Yes |
| Funnel stage | Where in the buying process it sits | 3 | No |
| Awareness | What the buyer already knows | 5 | No |
Three axes offer an "other". Two do not, and that is deliberate: every ad sits somewhere in a buying process and every ad is addressed to somebody, so on those two there is nothing an ad could honestly fall through to.
Every label has two parts, and the second one looks strange. You will see
Relatable pain next to pijnpunt. The first is the name you
read; the second is the identifier the database stores, and it is Dutch because the
people who first coded the vocabulary were.
By the time the product spoke English there were thousands of rows using those identifiers. Renaming them would have meant migrating every labelled ad, every gap, every brief and the model rubric in one move — and buying nothing a lookup table does not already give you. So the identifier stayed put and the English name became a display layer over it.
If you are building your own vocabulary, steal that: name a concept once, then never rename it. Your wording will change more often than you expect, and every rename is a migration you did not plan.
Hook — 11 named hooks
The hook is how the first two seconds earn the next two. It is about the opening move only, not what the ad goes on to argue.
Curiosity gap curiosity
Opens a loop the viewer needs closed, withholding the subject itself.
See an example
First frame. A hand closing a mirrored cabinet. Three bottles on the shelf behind it, none of the labels readable.
What it does to you. The itch to know which two. You keep watching to close a loop you did not open.
Why this one. You cannot tell what is being sold. The distance between “three things” and which two is the reason anyone watches the second sentence.
Not this if. Name the product in the opening line and the loop closes. That is a Claim.
Contrarian contrarian
Opens by stating the opposite of what the category normally says.
See an example
First frame. Someone setting an exfoliant down on the counter and looking straight at the lens. No music.
What it does to you. A small sting of being told off. You stay to find out whether they can back it up.
Why this one. The category says exfoliate more. Opening against the received advice is the entire move.
Not this if. It has to contradict what the category says, not what the viewer feels. Opening on a frustration the viewer already has is Relatable pain.
Pattern interrupt interrupt
A visual or audio jolt, with no argument attached yet. Stops the scroll on disruption alone.
See an example
First frame. A pan on a hob. The oil goes up. Whip pan, no captions, nobody speaking.
What it does to you. Startle, and nothing else yet. Your thumb stops before you have decided anything.
Why this one. Nothing has been argued yet. The opening buys attention on disruption alone.
Not this if. The moment the jolt carries an argument — “this is what your pan is doing right now” — the hook is that argument. Interrupt is the jolt with nothing attached.
Question vraag
Opens by asking the viewer something directly.
See an example
First frame. A jar held up to a window, turned slowly. Text on screen, no face in shot.
What it does to you. You answer it in your head before you can stop yourself, and now you are in the ad.
Why this one. Second person, answerable, and it makes the viewer go and check something.
Not this if. A question that already contains its answer is a Claim wearing a question mark.
Fact statistiek
Opens on a number or a checkable claim about the world.
See an example
First frame. A number typed on screen over a macro shot of a mascara wand.
What it does to you. Mild alarm about something you own. It is a claim about the world, so it does not read as a pitch.
Why this one. A claim about the world rather than about the product, and one you could go and verify.
Not this if. “Our formula lasts twelve hours” is about the product. That is a Claim.
Relatable pain pijnpunt
Opens inside the problem, described as the viewer actually experiences it.
See an example
First frame. A booking confirmation, then a banking app, then the same person still looking at it.
What it does to you. Recognition. Nothing has been offered, so there is nothing to resist yet.
Why this one. It describes the viewer's situation and offers nothing yet. Recognition first, product later.
Not this if. Put the relief in the same breath and you have moved to a Value teaser.
Claim claim
A flat assertion by the brand about the brand. "The strongest one you can buy."
Narrowed August 2026. Anything that withholds to earn a click is a value teaser; anything in a customer's voice is a customer quote.See an example
First frame. The product still in frame, one line of type under it, held for two seconds.
What it does to you. Flat and checkable. You want that number or you do not, and you know within a second.
Why this one. A specific assertion about the thing being sold, stated flat, with no setup.
Not this if. If the assertion is about the world rather than about the product, it is a Fact.
Value teaser value_teaser
A benefit withheld and dangled. "The reason our shoes last 800km."
See an example
First frame. A full inbox. Hard cut. The same inbox at zero.
What it does to you. Want, before cost. The gap between the two shots is the entire pitch.
Why this one. The payoff arrives first. What it costs you arrives later, if at all.
Not this if. A teaser promises a result. A Claim states a property. “Empty inbox” is a result; “nine days of charge” is a property.
Customer quote customer_quote
The ad opens in a customer's words, on screen or read aloud.
See an example
First frame. A sentence in quote marks on a plain card, a first name credited underneath.
What it does to you. Borrowed trust. Somebody who is not the brand went first.
Why this one. The first voice belongs to a customer and is marked as belonging to one.
Not this if. A creator saying the same sentence about their own experience is a POV. The quote has to be attributed away from the speaker.
POV pov_hook
The first two seconds are shot as the viewer's own point of view.
Not the POV format. This is the opening only. See the confusion pairs.See an example
First frame. Handheld and low, from the eyeline of someone still in bed. A clock in shot.
What it does to you. You are put inside a morning you recognise, and the clock does the rest.
Why this one. You are placed inside a moment and told whose eyes you are behind.
Not this if. POV is the vantage point, not the content. If the first thing said is a number, the hook is a Fact told from a POV.
Meme meme
A borrowed internet format that expects recognition to do the work.
See an example
First frame. Two lines of heavy white type on a still, then a cut to the product.
What it does to you. Recognition of the template itself. The joke walks past your ad defences.
Why this one. It borrows recognition that already exists. The template does the stopping, not the writing.
Not this if. Being funny is not the same as borrowing a format. An original written scene is a Skit.
Format — 11 named formats
Format is what the asset physically is — the thing a producer would have to schedule. It is the axis that decides what an idea costs to make.
Testimonial testimonial
A person recommending the product to camera.
See an example
What you would see. One person, front camera, at home, no cut for the first eight seconds.
What it does to you. Somebody reporting back rather than performing. The thin production is the signal.
Why this one. The asset is built around a person's account of their own result.
Not this if. Customer quote is a hook and Testimonial is a format — different axes, so an ad can be both. Label the hook on how it opens and the format on how it is built.
GRWM grwm
Get-ready-with-me. The viewer is taken through a routine in real time.
See an example
What you would see. A mirror in morning light. A routine started and talked through. The product arrives fourth.
What it does to you. Company rather than persuasion. You are watching a routine that happens to contain an ad.
Why this one. The routine is the structure. Remove the product and the video still has a shape.
Not this if. A routine that is only the product is a Product demo. GRWM earns its name from everything around it.
Day-in-life day-in-life
A person is followed through a stretch of their day.
See an example
What you would see. Coffee, commute, desk, gym, evening. Six locations, the time stamped in the corner.
What it does to you. A life to measure yourself against, with the product as a detail inside it.
Why this one. Time passing is the spine.
Not this if. One continuous routine in one place is GRWM. Day-in-life has to move.
Before/after before-after
Two states shown in sequence. The change is the whole creative.
See an example
What you would see. The same wall, same angle, same light, two states cut hard against each other.
What it does to you. The cut does the arguing. You believe your own eyes faster than anybody's voice.
Why this one. The argument lives in the comparison of two frames, not in what anyone says.
Not this if. Transformation is an angle and can be argued in words with nothing shown. This format has to show both states.
Unboxing unboxing
The product is revealed from its packaging on camera.
See an example
What you would see. A box on a table, tape split with a key, tissue, the product lifted clear.
What it does to you. Anticipation and a small payoff. You are watching somebody else get the thing.
Why this one. The subject is the reveal.
Not this if. Once they start using it and the point becomes how well it works, it has become a Product demo.
Voiceover voiceover
Narration over footage the narrator does not appear in.
See an example
What you would see. Macro product b-roll and hands. Nobody appears on camera at any point.
What it does to you. Authority without a face, which means the writing has to carry all of it.
Why this one. The voice does the arguing and nobody performs on screen.
Not this if. A face on camera saying the identical words is a different format. This one is defined by the absence.
Skit skit
A scripted scene with characters playing parts.
See an example
What you would see. Two people, an office set, a setup line and a turn. Played, not documented.
What it does to you. Entertained first, sold second. That is the trade the format makes.
Why this one. Someone is acting, and the scene was written for this ad.
Not this if. A Meme borrows a template that already exists. A skit writes its own.
POV pov
The whole ad is staged as the viewer's own viewpoint.
See an example
What you would see. Camera at eye height, hands entering the bottom of frame as though they were yours.
What it does to you. Simulation rather than observation. You are doing it, not watching it.
Why this one. The camera position is the decision. It is a way of shooting, not a way of opening.
Not this if. POV is also a hook. The hook is the first two seconds; the format is the whole asset. A POV hook on a Testimonial format is a normal ad.
Product demo product-demo
The product itself is the entire creative — hands operating it, or an animated reveal.
Added 2026. Every other format names a presenter-led device. An ad with no presenter had no name, and 13% of one sample fell through to "other" because of it.See an example
What you would see. The product in use, macro, real time, doing the thing it exists to do.
What it does to you. Proof by demonstration. Nothing to believe, only something to watch.
Why this one. Function is the subject and the camera stays on it.
Not this if. If the camera is on the person's reaction rather than the product's behaviour, it is a Testimonial.
Feature callout feature-callout
An annotated information graphic: labelled specs, leader lines, icon rows, side-by-side tables.
See an example
What you would see. The product static in centre frame, labels landing one at a time on the beat.
What it does to you. A list you can scan. It assumes you may already be sold and just need the spec.
Why this one. The asset is a list, and the edit is built to carry a list.
Not this if. A demo shows it working; a callout names it. Plenty of ads do both — label the one the edit is built around.
Static static
An unannotated still whose subject is the product or the scene.
Narrowed 2026. It used to mean "the file does not move", which made it absorb every still on the platform.See an example
What you would see. One frame. Product, headline, logo, and no motion anywhere.
What it does to you. It has to work in the half second before the thumb moves. All of it is composition.
Why this one. Every other format needs time to exist. This one has none.
Not this if. A sequence of stills is not a static. If there is a cut, it is a video.
Format is the axis that costs money. A feature callout is an afternoon in Figma. A day-in-life needs a creator, a script and a shoot. We scored all eleven on a 0–100 production difficulty scale — see what each format costs to make.
Angle — 9 named angles
Angle is the argument. Not what the ad shows, but what it is asking you to believe. Two ads can share a format and a hook and still be making completely different cases.
Problem probleem
Names a problem and positions the product as its resolution.
See an example
How it is built. Most of the runtime spent on the mess: the failure, the thing not working, in detail.
What it does to you. Discomfort held a beat longer than is comfortable, so the product lands as relief.
Why this one. The weight of the ad sits on the problem. The product arrives late and briefly.
Not this if. If the problem is only a runway for the change, and the change gets the screen time, it is a Transformation.
Transformation transformatie
Argues from a before-state to an after-state. The gap is the pitch.
See an example
How it is built. Where you are, then where you could be, with the route between them shown.
What it does to you. Aspiration with a path attached. The distance is the offer.
Why this one. The subject is the delta itself.
Not this if. Before/after is the format that shows a delta. Transformation is the argument, and can be made in words with nothing shown.
Education educatie
Teaches something that stays true whether or not you buy.
See an example
How it is built. A diagram, a demonstration, somebody explaining a mechanism you did not know about.
What it does to you. You end it cleverer than you started, and you owe that to a brand.
Why this one. The viewer leaves with something usable even if they never buy.
Not this if. Teaching one fact to set up a claim is a Fact hook. Education means the teaching is the body of the ad.
Social proof social-proof
Other people's adoption is the argument. Reviews, counts, crowds.
See an example
How it is built. Review screenshots stacked, a counter climbing, a wall of faces.
What it does to you. Safety in numbers. The decision has already been taken by people like you.
Why this one. The reason to act is other people, in numbers or in volume of reviews.
Not this if. One person's account is a Testimonial. Social proof needs the crowd.
Scarcity schaarste
Limited time, limited stock, or a closing window.
See an example
How it is built. A stock counter, a date, a sold-out label creeping across the variants.
What it does to you. Time pressure rather than desire — the fear of the choice being taken away.
Why this one. The reason to act now is supply or time running out.
Not this if. A discount that is simply available is an Offer. Scarcity has to end.
Offer aanbod
The offer is the argument. "Here is 50% off."
Added 2026. Half the ads sitting in Problem turned out to be plain discount ads. A discount is not an argument in the sense every other label means.See an example
How it is built. Price, bundle and guarantee on screen at once, all of it legible.
What it does to you. Arithmetic instead of emotion. You are being asked to do a sum.
Why this one. Commercial terms are the argument.
Not this if. If the terms expire and the expiry is the point, that is Scarcity.
Authority autoriteit
An expert, a credential, or a test result carries the claim.
See an example
How it is built. A named professional, captioned with the credential, explaining the why.
What it does to you. Deference. You are asked to trust training rather than somebody's experience.
Why this one. Standing to judge is what carries the claim.
Not this if. A famous face with no relevant expertise is Social proof or Lifestyle. Authority needs the credential to be the reason.
Comparison vergelijking
Argues against a named or clearly implied rival.
See an example
How it is built. Two products side by side, one dimension named, the difference shown.
What it does to you. A verdict you can check. The rival being in frame is what makes it feel fair.
Why this one. The argument does not work without the other thing in frame, named or not.
Not this if. Comparing your own before and after is Transformation. Comparison needs a rival.
Lifestyle lifestyle
Identity is the argument — who you become, not what it does.
See an example
How it is built. The product inside a life: a kitchen, a coastline, good light, no claims made.
What it does to you. Identity. Nothing is argued, so there is nothing to argue with.
Why this one. No functional claim is made anywhere. Identity is the entire pitch.
Not this if. The moment a benefit gets named, you have left Lifestyle.
What happens when nothing fits
Hook, format and angle each end in a label that means "none of these". It is not a dumping ground. How often it gets used is the one number that tells you whether the rest of the list is still any good.
Sooner or later a fixed list meets an ad it cannot describe. Someone tries a way of opening a video that did not exist last year, or an asset arrives that is three familiar things at once. There are only two ways to handle it: put the ad under the label that fits worst, or say plainly that the list has nothing for it.
The first is tidier and quietly ruins the count. A label that absorbs everything unfamiliar stops describing anything in particular, and by the time you notice, every comparison you have made using it is wrong. So the honest option gets a name of its own — and then it gets counted.
What to watch is the share, not the ad. One ad you cannot classify tells you nothing at all; ads are strange sometimes. Two or three in a hundred landing in "other" is a healthy list doing its job. More than one in ten is a signal that the list is missing a word, and the fix is to work out which word and add it — not to widen a nearby label until it swallows the gap.
In the data this label is stored as overig, which is Dutch for "other",
for the same reason every identifier on this page is Dutch.
Funnel stage — 3 labels
Funnel stage says where in the buying process the ad sits. It is deliberately coarse: three buckets, no catch-all, because every ad sits somewhere.
| Label | Slug | What it means |
|---|---|---|
| TOF | tof | Top of funnel. Reaching people who are not looking for you. |
| MOF | mof | Middle of funnel. Building the case with people already considering. |
| BOF | bof | Bottom of funnel. Closing someone who is close to deciding. |
Awareness — 5 labels
Awareness says what the buyer already knows, following Eugene Schwartz's five stages. It sits alongside funnel stage rather than replacing it, because the two answer different questions — and their disagreement is usually the finding.
| Stage | Slug | What they already know |
|---|---|---|
| Unaware | unaware | Doesn't know the problem exists. The ad has to introduce it. |
| Problem aware | problem_aware | Feels the problem, has no idea a category of solution exists. |
| Solution aware | solution_aware | Knows the category, does not know this brand. |
| Product aware | product_aware | Knows the brand, still hesitating. |
| Most aware | most_aware | Wants to buy, waiting for a reason or an offer. |
Here is why both axes exist. A bottom-of-funnel ad aimed at a product-aware buyer is retargeting, and it works. The same bottom-of-funnel ad aimed at someone who has not yet accepted the problem is wasted spend. Funnel alone cannot tell those two apart.
And a category that only ever speaks to solution-aware and product-aware buyers has left the top of its own funnel unclaimed. That is a gap a coverage grid built on funnel stage alone simply cannot see.
The six pairs everyone confuses
This is the part of any taxonomy that is genuinely hard. Six pairs account for most mislabelling — here is the line that separates each one.
1. Claim, value teaser, or customer quote?
The test: does the opening withhold something to earn the click? Then it's a value teaser. Is it in a customer's voice? Then it's a customer quote, whatever it says.
2. POV the hook, or POV the format?
The test: same word, two axes, different questions. Hook asks how it opens. Format asks how it is staged throughout.
3. Problem, or offer?
The test: strip the discount out. Is there still an argument left? If not, it's an offer.
4. Static, feature callout, or product demo?
The test: if text and pointers carry the meaning, it's a feature callout. If it moves and the product is the only subject, it's a product demo. These are three different jobs to brief, which is exactly why they are three labels.
5. Comparison the angle, or feature callout the format?
The test: a spec sheet that names a rival is both — comparison on angle, feature callout on format. The axes are independent, so you are not choosing between them.
6. Funnel stage, or awareness?
The test: they are supposed to disagree sometimes. When they do, that is the finding — not a labelling error to reconcile.
One rule that saves most arguments: a fact belongs on exactly one axis. The offer lives on angle, never on format — putting it in both would collapse a format × angle grid onto its own diagonal and every cell would agree with itself.
How we know when the vocabulary is wrong
A missing label does not always announce itself. Sometimes the share of ads landing in "other" climbs and the gap is obvious. More often those ads get absorbed by the nearest surviving label, which quietly gets fatter while the "other" rate stays calm. The first is caught by watching that rate. The second is caught only by asking whether a label has started to mean too many things at once.
Two labels were added in 2026 because measurement demanded it, and both stories are worth telling because they show what a taxonomy audit actually looks like.
The hook axis was broken, and the catch-all said so
Across 670 labelled ads from one large European retailer, 36.8% came back
overig on hook — against 0.8% on format and 4.6% on angle. That
alone was the alarm. But the more telling number was next to it:
claim held 268 of those 670 labels.
That is the shape a missing label makes. The nearest surviving word absorbs everything
that has no name of its own. Four hooks were added — value teaser, customer quote, POV
and meme — and claim was narrowed to what it should always have meant.
The format axis had a hole nobody could see
This one surfaced as two symptoms that looked unrelated. First, 25 of 200 ads (13%)
came back overig on format — all video, all showing hands operating a
product or an animated reveal, and the model returned "other" at 0.8–0.9 confidence.
It wasn't confused. The word simply did not exist.
Second, and it took longer to spot: static held 47%. It was the one label
defined by a medium rather than an execution, so it absorbed every still by default.
Of 94 stills, 25 were annotated information graphics and 27 were plain packshots —
different things to make, and a brief has to be able to ask for one by name.
How to use this without Likely
Take the list. Put it in your brief template as five dropdowns and make them required. That single change gets you most of the value on this page, and it costs you nothing.
Concretely, three things you can do this week:
- Add the five axes to your brief template. Five required dropdowns, no free text. Now every brief your team writes is countable.
- Label your last 30 launched ads. One pass, one person, an hour. Then cross format against angle in a spreadsheet and look at the empty cells. That is a coverage grid, built by hand.
- Watch your own catch-all rate. If more than one in ten of your ads won't fit the list, the list is missing a word for your category. Write it down and keep the definition tight.
The tedious part isn't the vocabulary. It's labelling several hundred competitor ads the same way, twice, months apart, so the comparison holds. That's the part we built a product for — but the vocabulary works fine in a spreadsheet, and you should have it either way.
Keep going
Three things that build on this
See this applied to your category.
Name two competitors and Likely labels every live ad they have on all five axes. About twenty minutes, no ad account connection.
Run my €0.99 audit