The ad diagnosis grid: angle, hook, format
THE SHORT ANSWER
An ad is classified on three independent axes: angle (4 types — objection, desire, proof, mechanism), hook (10 archetypes, judged from the opening alone) and format (14 production formats, split by medium). Angle and hook alone give 40 combinations. Most accounts, counted honestly, occupy three or four of them.
Every tool that looks at your ads has a grid it sorts them into, and almost none of them will show you what it is. That is backwards. If software is going to tell you something is wrong with your creative, you are entitled to see the ruler it measured with — otherwise you are taking a verdict on faith.
So here is ours, in full: the three axes, every allowed value, and the rule that makes it decline to answer. It is the same taxonomy our classifier reasons with, not a simplified marketing version of it. Take it and run it by hand if you like; the last section explains how.
AXIS 01 / 03
Angle — what kind of argument is this?
The angle is the core persuasion strategy: not what the ad is about, but what kind of case it is making. Alongside the type, a short brand-specific label names this particular message — “no logo elegance”, “italian popularity” — so that two mechanism ads in the same account remain distinguishable.
TABLE SCROLLS SIDEWAYS →
| TYPE | WHAT IT DOES | HOW TO RECOGNISE IT |
|---|---|---|
| Objectionobjection | Answers the doubt the reader already has. The ad’s job is not to create desire but to clear the thing blocking it — price plausibility, durability, fit, whether it actually arrives. | Reads like a reply to a question nobody asked out loud: “real leather under €80 — seriously?” |
| Desiredesire | Paints the wanted outcome, identity or aspiration. Nothing is being disproved; something is being wanted. Most lifestyle creative lives here, which is also why most accounts are over-weighted in it. | Strip the product out and the ad still sells a version of the viewer’s life. |
| Proofproof | Evidence carries the message: reviews, sales counts, demonstrations, before-and-after, third-party validation. The persuasion is borrowed from someone who already bought. | There is a number, a named person, or a visible demonstration doing the work. |
| Mechanismmechanism | Explains how it works or what makes it structurally different — process, material, engineering. The move that still lands in categories where every promise has been maxed out. | You could draw a diagram of the claim. It has parts. |
AXIS 02 / 03
Hook — how does it open?
The hook is the opening attention device, and it is judged from the first visual or first line only. This is the axis that goes stale fastest: an archetype your audience has been shown by every brand in the category for six months is not neutral any more, whatever it does in someone else’s case study.
TABLE SCROLLS SIDEWAYS →
| ARCHETYPE | WHAT IT IS | HOW TO RECOGNISE IT |
|---|---|---|
| Pattern interruptpattern_interrupt | An unexpected visual or statement that breaks the scroll. | Something is wrong with the frame in a way the eye has to resolve. |
| Big claimbig_claim | A bold headline promise, stated flat and early. | The first line is the largest thing the ad will ever say. |
| Problem agitateproblem_agitate | Names a pain and then twists it. | The reader feels slightly worse at second three than at second one. |
| Calloutcallout | Addresses a specific audience directly. | A group is named, and everyone outside it is being let go on purpose. |
| Founder POVfounder_pov | First-person founder or maker voice. | Someone says “I” and means it about this product. |
| Curiosity gapcuriosity_gap | Withholds information to pull the view. | The sentence is grammatically finished and informationally isn’t. |
| Social proofsocial_proof | Others already use it or love it. | The subject of the opening sentence is other people, not you. |
| Negationnegation | Defines the product by what it is not. | The first thing you learn about it is an absence. |
| Before / afterbefore_after | Transformation or contrast. | Two states, and the ad is the arrow between them. |
| Shockshock | A provocative or surprising open. | You would not say it at a dinner party, which is the point. |
AXIS 03 / 03
Format — how was it produced?
Format is the production shape, and it is the one axis where the medium constrains the answer: a video ad gets a video format, a static gets an image format, and that invariant is enforced rather than trusted. It is also the axis most often confused with angle — “UGC” is a format, not an argument.
Video (8)
- video_voiceover
- Narrated over footage.
- video_pitch
- Direct talking pitch of the product.
- video_longform
- Extended story or explainer.
- video_founder
- Founder on camera.
- video_faq
- Question-and-answer structure.
- video_unboxing
- Opening or handling the product.
- video_advertorial
- Editorial, story-style ad.
- video_offer
- Discount, bundle or promo focus.
Image (6)
- image_benefit
- Single benefit or feature hero.
- image_comparison
- Us-versus-them or before-after still.
- image_socialproof
- Testimonial, review or UGC still.
- image_offer
- Discount, bundle or promo still.
- image_carousel
- Multi-card layout.
- image_advertorial
- Text-heavy editorial still.
Why three axes and not one
Three axes rather than one, because the interesting failures live between them. An account can look varied — twelve creatives, four photographers, three formats — while every single ad runs a desire angle opened with a big claim. That is not twelve tests. That is one test with twelve outfits, and no amount of new footage will change its result.
The grid makes that visible as arithmetic instead of intuition. Four angles times ten hooks is forty combinations before format enters at all. Most accounts, counted honestly, are sitting in three or four of them.
The rule that makes it say “unknown”
Every dimension carries a confidence, and the classifier is instructed to leave a dimension as unknown with low confidence when the creative genuinely does not settle it — rather than invent a value to look complete. The example that forces the rule: a video ad is classified from its poster frame, and a single still very often cannot prove voiceover versus talking pitch.
“Honesty over guessing: if a dimension is genuinely indeterminable, set its value to unknown and confidence ≤ 0.3. Never invent a tag to look confident.”
This matters more than it sounds. A grid in which every ad received a confident tag looks like better data and is worse data — the guesses are indistinguishable from the readings, and every count built on top of them inherits the error silently. One honest blank costs you a cell. A confident wrong tag costs you the table.
Running the grid by hand
Five steps, a spreadsheet, an afternoon. No tool required — and doing it manually once is the fastest way to find out whether the output is worth automating for your account.
- 01
Pull thirty ads, not five
Export the creatives that carried real spend in the last ninety days. Fewer than about twenty and you are reading noise; a single week of data will hand you a confident answer about nothing.
- 02
Tag each ad on all three axes before looking at any numbers
Angle, hook, format — in that order, one pass per axis. Doing all three per ad tempts you to make them agree. Doing one axis across all thirty ads keeps the definitions stable, which is the only thing that makes the counts comparable later.
- 03
Write “unknown” when you mean it
A poster frame often cannot settle voiceover versus pitch. An ad can genuinely sit between desire and proof. Recording that honestly costs you one cell and saves the whole table — a grid where every ad got a confident tag is a grid that has stopped measuring anything.
- 04
Count before you judge
How many of your thirty are desire? How many hooks are big_claim? Most accounts discover the concentration here — not that the ads are bad, but that eleven of them are the same ad wearing different photography.
- 05
Only now bring in performance
Group spend and result by angle, then by hook. You are looking for two things: a cell that carries spend without carrying results, and a cell with no ads in it at all. The second one is usually the more interesting.
What the grid cannot tell you
What the grid does not do is rank. There is no “best” angle and no hook that is correct in general — a proof angle is inevitable for a supplement and slightly absurd for a founder’s first product with no customers yet. The taxonomy names what an ad is doing. It does not tell you whether that was the right thing to do.
It also cannot see the two things that most often explain a result: where your market sits on the awareness ladder, and what your audience has already been shown to exhaustion by everyone else in the category. Those are separate diagnoses, and a tag on a single ad will never contain them.
And the honest boundary: we do not claim that classifying your ads this way makes them perform better. It makes what you have been doing legible. What you do with that is the part nobody can automate for you.
Questions people actually ask
- What is an ad angle?
- An angle is the core persuasion strategy of an ad — what kind of argument it is making. In this taxonomy there are four: objection (answers a doubt), desire (paints a wanted outcome), proof (lets evidence carry it) and mechanism (explains how it works). The angle is not the topic and not the format; two ads about the same product, shot identically, can run opposite angles.
- What are the hook archetypes?
- Ten opening attention devices: pattern interrupt, big claim, problem agitate, callout, founder POV, curiosity gap, social proof, negation, before/after, and shock. The hook is judged from the first visual or line only — what happens in second four does not change which archetype the ad opened with.
- How do you diagnose a Meta ad?
- Classify it on three axes — angle, hook, format — then compare the distribution of those tags against where your spend and results actually landed. The diagnosis is rarely about a single bad ad. It is usually a concentration: most of the account running one angle, or spend piling into a hook that stopped working two months ago.
- Does the grid tell you which ad will win?
- No. It tells you what you have been making and where the gaps are. Whether a different distribution produces better results is a question your own tests answer, not a taxonomy — and we do not claim that applying this grid improves performance.
- Can I use this without Trailmaker?
- Yes, and it is a reasonable afternoon’s work with a spreadsheet: three columns, thirty rows, the definitions above. The grid is not the product. What gets hard by hand is doing it again every month and remembering what the last four rounds already told you.
- Why does the classifier answer "unknown"?
- Because a wrong tag is more expensive than a missing one. The classifier is instructed to return "unknown" with low confidence when a dimension is genuinely indeterminable from the creative and copy, rather than invent a value to look complete. A single video poster frame frequently cannot prove voiceover versus pitch, and the honest answer keeps the rest of the table trustworthy.