· 22 min read

Ad Creative Brief: Six Fields That Make Failure Diagnosable

A creative brief is a prediction, not a description. Six fields you can diagnose later, plus five steps for reading a competitor's brief backwards from ads.

A marketer presenting a creative brief to a team in a bright office

Most creative briefs are descriptions of a deliverable. Thirty seconds, vertical, upbeat, product in use, two people, natural light, logo at the end. Every clause is true the moment someone writes it, which is the problem: a brief that cannot be wrong cannot be learned from. When the ad underperforms, nobody knows what to reopen, so the next round starts from a blank page and the same instincts.

A useful brief is a different kind of document. It commits to six things that can each fail independently, and one sentence about what should happen if they are right. Then, when the ad runs, the failure has an address. You do not argue about whether the video was good. You ask which of six fields was wrong, and you change one.

This guide is about that document: what each field is responsible for, how to tell when it is wrong, how to reverse-engineer a competitor’s brief from their live ads, and a one-page sheet you can copy.

A Brief Is a Prediction, Not a Description

Compare two one-line briefs for the same product.

The descriptive one: “Make a 30-second video for cold traffic showing the lunch box in an office, with upbeat music and text overlays.”

The predictive one: “People who bring lunch to work stop because the food is cold at 1pm, not because packing is hard. A visible six-hour thermometer test will make a $64 box worth it against the $14 one they already own. One continuous take, fifteen seconds, no music. If it is right, cold-traffic click-through should beat our current control by enough to justify a second execution.”

The second one is not better written. It is a different object. Every clause in it can be wrong, and — this is the part that matters — each one fails differently:

  • If the audience is wrong, the ad gets cheap clicks from people with no reason to care, and the comments say so.
  • If the insight is wrong, the ad reads as true but never urgent. People agree and scroll.
  • If the claim is wrong, the ad can still earn a click, but the product page has to do work it was never designed for.
  • If the evidence is missing, the ad is believed about as far as the discount.
  • If the format is wrong, the claim gets cut before it lands, and the drop-off is measurable in seconds.
  • If the reference is wrong, you get a competent copy of someone else’s idea with your product pasted into it.

Six fields, six failure signatures. That is the whole argument for writing them down: not documentation, diagnosis.

There is a second job the brief does, and it is easy to miss. It defines what counts as a variant. Two ads built from one brief differ in hook or execution and share a claim; two ads built from two briefs differ in what they promise, which means their results cannot be compared. Most accounts that “test creative” are actually testing briefs against each other without knowing it, which is why the winner never generalises.

The Six Fields, and What Each One Is Responsible For

FieldThe question it answersLooks wrong whenSymptom it produces
AudienceWho is this ad speaking to, in what situation?It describes a demographic rather than a momentCheap clicks, wrong comments, weak downstream
InsightWhat do they already believe that is costing them something?It is a fact about the product or a truismAgreement without urgency
ClaimWhat single thing are we asserting that they would otherwise not believe?It has no named alternative, so it cannot be falseClicks without conviction
EvidenceWhat makes the claim survivable by a skeptic?It is an adjective, or it does not exist yetBelief that stops at the price
FormatWhat constraint forces the claim to be visible?It was chosen before the claim wasThe claim gets cut, or never shown
ReferenceWhat are we deliberately borrowing, and what are we refusing?It is unannotatedA copy of someone else’s ad

The order is not arbitrary. Audience and insight are decided before anything is shot. Claim and evidence are decided together, because a claim without evidence is a wish. Format follows the evidence, not the other way round. Reference comes last, so it cannot substitute for thinking.

1. Audience: a situation, not a segment

The audience field answers who the ad is written for. It is not the targeting setting. The ad set decides who gets served the ad; the brief decides who the words are addressed to. Conflating them is the most common structural error in the document, and it produces ads written for a demographic that no human belongs to.

A test: can you name a place where this person is already talking about the problem? A review section, a specific search query, a forum thread, a comment pattern. If you cannot, you have a market, not an audience.

VersionExampleWhat is wrong with it
SegmentWomen 25–44, interested in cooking and healthy eatingDescribes a media buy; nobody recognises themselves
SituationOffice workers who bring lunch three days a week and whose bag leaks by noonNames a moment, a friction and a consequence
Situation plus historySame, and they already bought an insulated bag that did not hold temperatureAdds the thing they will compare you against

The third version is the one that produces copy. It tells you what the first line should say, what the visual should show, and what objection the evidence has to answer, because the audience has already tried something.

The failure signature is specific: strong click-through, weak downstream, and comments from people explaining that this is not for them. That is not a creative problem. The ad was written for someone who was not in the room.

2. Insight: a belief that is currently costing them something

An insight is not a fact about your product, and it is not an observation about the market. It is something the audience already believes or does that is currently producing a worse outcome than they want. The word “currently” is doing the work: if the belief is not costing them anything today, it will not move them tomorrow.

Three rules, in order of how often they are broken.

It has to predict a behaviour. State the insight, then answer: what does this person do differently because they believe it? “People care about quality” predicts nothing. “People who pack lunch have already solved for healthy and are now optimising for still edible at 1pm, so they buy insulated bags and then discover the bag does not hold temperature” predicts a purchase, a return and a search query.

It has to be findable in public text. A brief that cites where the insight came from is a brief someone can argue with, and being arguable is the point. Reviews, comment threads, support tickets, search autocomplete and forum posts are all free and all quotable. An insight with no source is a hunch with better formatting.

It cannot be about the product. “Our box holds 1.2 litres” is a specification. “Our box is the only one at this price with a gasket” is a claim. Neither is an insight.

The failure signature: the ad feels true and does nothing. People agree with it in the comments and keep scrolling, because you have told them something they already knew in a pleasant way. The other signature is worse — the ad is built on a problem the audience does not have, so the response is polite confusion.

3. Claim: one sentence, comparative, falsifiable

A claim needs a named alternative. What would they buy, or keep doing, if this product did not exist? Without an alternative, a claim cannot be false, and a claim that cannot be false is a slogan.

ClaimAlternative it beatsFalsifiable?
”The last lunch box you will ever need”None statedNo
”Keeps food cold at 1pm without an ice pack”The ice pack they currently useYes
”Forty percent lighter than the steel box at the same volume”A specific competing formatYes, if the weights are real

Three rules for the claim field.

One claim per asset. Two claims is two ads. The moment you write “and also”, you have started a second brief.

The claim must be about a difference, not a category. Every advertiser in a category can say the category’s claim, which is why category claims stop registering — the same mechanism covered in why fatigue moves through metrics in a fixed order. If your claim would be true of your three closest competitors with the logo swapped, it is wallpaper.

The claim has to survive being said out loud by a customer. If a real buyer would not repeat it to a friend without embarrassment, it is marketing language, and the ad will need to spend its whole length overcoming the tone.

The failure signature is a click without conviction: click-through is acceptable, add-to-cart is thin, and the product page is doing the persuading. Or the ad only works while a discount is running, which means the discount is the claim.

4. Evidence: the cheapest thing that survives a skeptic

Evidence is where briefs die. Not because teams do not want it, but because the evidence does not exist and nobody wants to say so, so the brief asks the editor to manufacture the feeling of proof with pacing and text overlays.

There is a hierarchy, and each step down costs belief:

TierEvidence typeExampleWhat it costs to produce
1DemonstrationOne continuous shot of the thermometer at hour sixA camera, a clock, an afternoon
2Measured number with a stated method”41°F at six hours, measured with a probe at the centre”A measurement and one sentence
3Material or specification that implies the outcomeDouble-wall vacuum, 18/8 steel, gasket sealNothing, if it is true
4Third-party testA lab or publication result you can citeUsually money
5Customer quote with specifics”Third winter, still cold at lunch”A customer who said it
6Adjective”Premium insulation”Nothing, and it is worth nothing

Two rules follow.

If the evidence does not exist, the brief is not ready. It is a product or operations request wearing a creative hat. Say that in the document. A brief that says “evidence: to be confirmed” will be resolved by whoever is holding the camera, and they will resolve it with tier six.

The format must be able to show the evidence. A temperature claim cannot be proven in a montage. If the evidence needs a clock, the format needs a clock in frame. This is the one place where format is not a preference.

The failure signature is the most expensive one: good click-through, good add-to-cart, weak purchase and elevated returns. They believed the ad. The product did not keep the promise the ad made on its behalf. The second signature is a comment section full of “does it actually…”, which is the audience telling you the claim arrived without its proof.

5. Format: the constraint that tests the idea

Format gets chosen first in most briefs and should be chosen fourth. Its job is not to fit a placement. Its job is to make the claim hard to fake.

Claim typeFormat that proves itFormat that cannotProduction tier
Holds temperature over timeOne continuous take with a visible clockFast-cut montageLow
Fits a real bagTop-down shot inside the bag, uneditedStudio flat layLow
Easier than the alternativeSide-by-side of the same task, same personSingle-hero product shotLow
Better materialMacro, raking light, no colour gradeLifestyle scene at golden hourMedium
Cheaper over a yearStatic comparison with two numbersVideo with musicLowest

The table is deliberately biased toward cheap formats, because cheap formats are the ones that can carry evidence. A $300 shoot does not prove more than a phone on a tripod. It proves that someone had $300.

One length rule: the claim has to be legible in the first three seconds, or the claim is too complicated for the format and needs rewriting, not more seconds. A fifteen-second cut that opens with a problem statement has four seconds left to prove something. That is usually enough for a demonstration and never enough for a backstory.

The failure signature is a drop at second four. The hook did its job and the claim could not continue it, which means the two were never the same idea. This is the difference between a hook that has been verified across several ads and a hook that was written to be a hook.

6. Reference: three annotated, one anti-reference

References are the most abused field, because an unannotated reference is an instruction to copy. The document says “like this one” and the editor reproduces the structure, the pacing and the framing, and the product arrives as a guest in someone else’s ad.

Annotate every reference with three things: what is being borrowed, at what level, and what is explicitly not being borrowed.

  • “Borrow the proof device: the clock in frame. Do not borrow the actor, the kitchen or the music.”
  • “Borrow the pacing of the first three seconds. Do not borrow the claim.”
  • “Borrow the framing of the comparison. Do not borrow the category.”

Add one anti-reference: an ad in your category that is popular and wrong for your purposes, with the reason. This is the field that stops a team defaulting to the market leader’s format by habit, and it is the only place in the brief where you are allowed to be rude about someone else’s work — with a reason, about the work, not the brand.

Prefer references from outside the category. Inside references are where category wallpaper comes from; the ads everyone has already seen produce the ads everyone is already ignoring.

The failure signature is easy to spot in review: the deliverable is the reference with a different product in it. That means reference replaced claim, and the brief had six fields but only one of them was load-bearing.

The Seventh Line: The Prediction

The six fields describe the ad. One more line makes the brief a prediction, and without it the six fields are still just a well-organised description.

Write it in this shape: if [audience] [behaviour], then [metric] should move [direction] against [comparator] within [window].

“If office workers with a leaking bag see the six-hour test, then cold-traffic click-through should beat our current control within seven days, and add-to-cart should hold.”

You do not need a number before you have a baseline. “Our current control” is a legitimate comparator, and it is honest. What you cannot do is leave the line out, because the line is what turns a result into a lesson. For the budget side of that sentence — how much a verdict costs and how long a window it needs to mean anything — the arithmetic lives in creative testing as a sampling problem.

Reading a Competitor’s Brief Backwards

This is the practical use of the six fields, and it is the reason the structure is worth memorising. A live ad is the output of a brief you cannot see. Five steps get you most of the way back.

Step 1: Build the set. Twelve to twenty live ads from three advertisers in your category, from the free Ad Library. Record the full primary text, the format, the start date and the market. The full text matters because the feed truncates it; the library does not. The filters and the walkthrough are in the free Ad Library guide, and the search syntax pulls a category in one query instead of twenty page visits.

Step 2: Cluster by claim, not by look. Two ads with different actors, different music and the same promise are one brief. Two ads with identical visuals and different promises are two briefs. This is the step everyone skips, and it produces the most useful single observation in the whole exercise: an advertiser running fourteen ads with one claim is defending something. An advertiser running fourteen ads with fourteen claims is still shopping.

Step 3: Fill the fields backwards. Claim first, because it is the most visible. Then evidence, because you can see what they show. Then insight, because the evidence only works if the viewer already believes something — your job is to name that belief. Then audience, from who is named in the first line. Then format, from what the claim forced them to shoot. Reference last, from which of their ads look like siblings.

What you can seeWhat it tells youWhat it does not tell you
The promise repeated across adsTheir claim, and that it is defendedWhether it is profitable
The thing they show to prove itTheir evidence tierWhether the number is real
Who is named in line oneTheir audience definitionWho actually converts
The shooting structureTheir format constraintTheir production budget
Ads that resemble each otherTheir reference setWhether the resemblance was intended

Step 4: Date the cluster. How old is the oldest ad in the cluster, how many variants does it have, and how recently was a new one added? A cluster with six variants in three weeks is an active test. A cluster with one ad from fourteen months ago is a legacy asset that may never have been audited. The distinction between “still running” and “still defended” is covered properly in how to read ad longevity, and it is the difference between copying a live bet and copying an abandoned one.

Step 5: Find the empty slot. Build a small census: rows are claims, columns are evidence tiers, cells are counts across all three advertisers. Two findings matter.

A row where every advertiser makes the same claim at tier five or six — a claim the whole category asserts and nobody proves — is the most valuable finding in the table. The claim is already validated by the category’s spend; the proof is missing. Proving it is cheaper than inventing a new claim and better than shouting the old one louder.

An empty row is not automatically an opportunity. It is often empty because it does not work, because the claim is unfalsifiable, or because the product cannot deliver it. Treat an empty row as a question, not a gap.

A worked example

Everything here is hypothetical, built to show the shape of the table rather than to describe a real advertiser.

AdvertiserCluster (claim)Evidence tier observedInferred insightVariantsOldest ad
A”Cold at lunchtime without ice packs”1 — visible test”My bag does not actually hold temperature”95 months
B”Cheaper than buying lunch”2 — a monthly number”Packing is a cost decision, not a food decision”36 weeks
C”Premium insulation, designed in Copenhagen”6 — adjectiveNone observable123 weeks

Read across the table. A has one claim, nine executions and a five-month-old anchor: they are defending a proven angle, and the proof device is the thing to study. B has a cost claim with a real number and only three variants in six weeks: an early test, possibly a second angle rather than a second brief. C has twelve ads, no evidence tier above an adjective and a three-week history: that is a team producing volume against a brand claim, and the volume is not evidence of success. The empty slot is C’s column — a durability claim with no proof — and the reason to take it is not that C is wrong, but that A has already taught the market to want the proof.

The One-Page Sheet

One page is not a style preference. A brief longer than a page gets summarised in conversation, and the summary is whatever the most senior person remembers. What follows is the whole document.

BRIEF — [campaign] — [date] — [owner]

AUDIENCE
  Who, in what situation:
  What they have already tried:
  Where this shows up in public text:

INSIGHT
  What they believe that is currently costing them:
  What behaviour it predicts:
  Source:

CLAIM
  One sentence:
  The alternative it beats:
  What would make it false:

EVIDENCE
  Tier (1-6):
  Exactly what is shown or stated:
  Exists today? Y / N -> if N, this is not a creative request

FORMAT
  Structure:
  Why this structure proves the claim:
  Length ceiling: __ seconds, claim legible by __ seconds

REFERENCE
  1. Borrow: ___  Do not borrow: ___
  2. Borrow: ___  Do not borrow: ___
  3. Borrow: ___  Do not borrow: ___
  Anti-reference and why:

PREDICTION
  If [audience] [behaviour], then [metric] moves [direction]
  against [comparator] within [window].

KILL LINE
  We reopen this brief if:

The kill line is the field most teams omit and the one that keeps the document honest. Write the condition under which you would abandon the claim, not the ad. “If add-to-cart stays flat across two executions with tier-one evidence, the insight is wrong, not the edit.”

Which Field Failed

The reason to keep six fields separate is that the ad’s symptoms point at different ones. Read the symptom, reopen one field, change one thing.

What you observeField to reopenWhat you probably wrote
Low click-through, and drop-off at second threeFormatA structure that could not carry the claim
Normal click-through, comments saying “not for me”AudienceA segment instead of a situation
Agreement in comments, no urgency, flat responseInsightA truism, or a product fact
Click-through fine, add-to-cart thinClaimA category claim with no alternative
Good add-to-cart, weak purchase, high returnsEvidenceProof that did not survive contact with the product
Strong first run, nothing repeatableEvidenceA one-off performance rather than a mechanism
Ads look like the category leader’s adsReferenceAn unannotated reference

Two of those rows are not creative problems at all, and saying so is part of the brief’s job. The returns row is a product or expectation problem. The repeatability row is a “you got lucky with one asset” problem. A brief that absorbs those failures and promises to fix them with a new edit will absorb them again next month.

What a Brief Cannot Do

It cannot invent differentiation. If the product is the same as three others at the same price, the sixth field will not save it. The honest brief says the claim is weak and asks for a product decision.

It cannot create evidence. It can only demand it, and it should refuse to proceed without it. Everything else in the document is downstream of that refusal.

It cannot replace the test. A perfect brief with an unreadable sample still produces no verdict, which is a budget question, not a document question.

It cannot survive being an approval document. This one is worth naming plainly. The moment a brief goes into a review cycle for sign-off, its incentive changes: a falsifiable prediction invites argument, and argument delays launch, so the fields get softened until nobody objects. What comes back is a description of a deliverable — the document this article started with. If the brief has to be approved, approve the budget and the evidence tier, and leave the claim to the person who will be held to it.

Final Word

The brief is the cheapest document in the campaign and the only one that decides what the campaign can learn. Six fields, each falsifiable, each with its own failure signature, and one line that says what should happen. Write it in a page, keep it arguable, and change one field per round — because the difference between a team that gets better at creative and a team that gets luckier is whether the last ad left behind an address for its failure.

Apply WINNING for 20% off TrendTrack →

Verified 2026 · 20% off Starter, Pro and Business for your first 3 months

Step 2 of the reverse-brief is the step that runs out of spreadsheet. Clustering by claim across three advertisers is twenty ads and an afternoon; keeping the cluster current across a whole category is a standing query, and that is where a database beats a copy-paste. TrendTrack indexes more than 10 million Meta ads with full text, start dates, formats and markets, so a claim census can be filtered instead of transcribed. TrendTrack promo code is WINNING - 20% off all plans for the first 3 months on the paid panels, and the free Chrome extension covers the per-ad reading that fills the sheet. Start with the extension setup guide or check what the free plan includes.

FAQ

What should an ad creative brief contain?

Six fields and one prediction. Audience as a situation rather than a demographic, insight as a belief that is currently costing the audience something, a single falsifiable claim with a named alternative, evidence at a stated tier that already exists, a format chosen to make the claim visible, and annotated references with one anti-reference. The prediction states what should move, against what comparator, within what window.

How long should a creative brief be?

One page, because anything longer gets summarised in conversation and the summary keeps whatever the most senior person remembered. Length is not the constraint that matters; the constraint is that every field contains an assertion someone could disagree with. A one-page brief with six falsifiable lines beats a five-page brief with a mood board and no claim.

Is the brief’s audience the same as the ad set targeting?

No, and confusing them is the most common structural error. The ad set decides who is served the ad. The brief decides who the ad is written for, which controls the first line, the casting and the objection the evidence has to answer. When the two diverge you get cheap clicks from people who were never going to buy, and the comments will tell you so.

Can I write a brief by reverse-engineering competitor ads?

Partly. You can recover their claim, their evidence tier, their format constraint and their reference set from public ad data, and clustering their ads by claim tells you which angle they are defending. You cannot recover their insight with confidence, because you are inferring a belief from an execution. Use the reconstruction to find what the category asserts without proving, then write your own insight and your own evidence.

How do I know which part of the brief was wrong?

Match the symptom to the field. Drop-off at second three points at format. Comments saying the product is not for them point at audience. Agreement without urgency points at insight. Clicks without add-to-cart point at claim. Good add-to-cart with weak purchase points at evidence, or at the product. Change one field at a time, or the next result will be as unreadable as this one.

Does a brief need a hypothesis if I am going to test anyway?

Yes, because the test only produces a lesson if something was predicted. Without the line, a losing ad is a losing ad and the next round starts from instinct. With the line, you learn whether the audience, the insight, the claim or the proof was wrong, which is the difference between one bad ad and one bad assumption you keep repeating.

    Share:
    Back to Blog

    Related Posts

    View All Posts »