How many variants per creative angle? The 3 then 5 to 8 rule

Plan on 3 variants per angle in round one, then 5 to 8 only on validated angles. An angle that is still unconfirmed after 3 variants gets closed, not extended. The real ceiling is budgetary: each variant must receive at least 2 to 3 times your target CPA in media to be readable. Beyond 8 variants of the same angle, you are no longer testing a message, you are repainting it. What follows is the full mechanism: the stopping rule, the reverse calculation from the media budget, the marginal cost of each type of variant, and the line to write in your quote. For its Smart+ Web campaigns run on a target CPA or minimum ROAS, TikTok recommends in its Smart+ Web best practices a daily budget ideally equal to 30 times your historical CPA, and never below 10 times that CPA.

Are a “variant” and an “angle” the same thing?

No, and confusing the two is what makes production budgets balloon without improving results. The angle is the reason to buy that you put forward: saving time, fear of getting it wrong, price, status. If you need the full definition and typology, it is detailed here with the associated costs. The variant does not change the reason to buy. It changes the way of saying it.

In agency quotes, four words circulate to designate three different things. Concept, angle, variant, version. A media buyer says “creative” for anything that carries an ID in the ad manager. A producer says “deliverable” for anything that comes out of the edit. Between the two, nobody counts the same way, and that is where invoices become impossible to compare.

The rule we apply is easy to check: if you can sum up two videos with the same promise sentence, they are two variants of the same angle. If the two sentences differ, they are two angles. “You save 40 minutes a day” and “You save 40 minutes a day, says a night-shift nurse”: same angle, two variants. “You save 40 minutes a day” and “Your colleagues are already using it”: two angles.

That leaves iteration, which is neither. An iteration is a corrected version after reading the data: you shorten it, move the CTA up, cut the 4 empty seconds. It comes after the test, not before. We cover this separately in creative iteration as a unit of purchase, because it is billed and planned differently from a variant produced upstream.

Variant = same promise, different execution, produced before the test. Iteration = same execution, corrected after the test. Angle = different promise. Three distinct quote lines, three distinct prices.

How many variants per creative angle in practice, depending on the testing round?

3 in round 1, 5 to 8 in round 2 if the angle performed, 0 in round 3 if nothing comes out. The question “how many variants per creative angle” has no single answer because the number is not a quantity: it is a rank. The same angle deserves 3 variants in week 1 and 9 in week 4. Or none.

In round 1, you are not looking for the best execution. You are looking to find out whether the promise interests anyone. Three variants are enough to rule out the hypothesis “it was the execution that failed”. A single variant proves nothing: if it fails, you will never know whether it was the angle or the edit. Six in round 1 is budget frozen on an unvalidated hypothesis. TikTok's ad manager points the same way: its list of recommendations, updated in February 2025, prompts you to add creatives as soon as an ad group has fewer than three, with a target of 3 to 5 unique creatives per group.

Moving to round 2 is decided on a criterion, not on a feeling. We use this one: at least one of the three variants reaches the target CPA or comes within 30% of it, with a hook rate above the floor. If two variants out of three are flat on the floor but the third performs, the angle is validated; what failed was the execution of the other two.

Published benchmarks point the same way. Panja, in its 2026 Meta Ads creatives guide published about two weeks ago, recommends 3 to 5 genuinely different variants per ad set at launch and 10 to 20 distinct concepts in rotation across the account at scale. Arthur Scaleyou Agency, in an article on Meta creative testing published about a month ago, describes an even tighter structure: 3 concepts with 2 hooks each, over a 3 to 7-day wave without touching the settings.

The most useful sentence of the lot comes from MarqOps, picked up by Panja: ten to twenty distinct concepts beat fifty variants of the same idea. That is exactly the mechanism we observe. The Meta auction remains based on estimated value, but a pool of varied concepts gives more material for personalisation: diversity is a creative best practice, not a bidding rule. According to the Engineering at Meta blog (October 2026), Meta Andromeda, Meta's ad selection system, compares ads across a larger set of candidates and has delivered a 2.4% increase in conversions on Facebook since 2025.

Why does your media budget, not your creativity, set the number?

A variant only really exists if you can afford to pay its reading floor. As long as a creative has not spent enough to produce an interpretable signal, it is not tested: it is simply live. And an untested live creative costs as much to produce as a tested one.

The floor we use on our e-commerce accounts: 2 to 3 times the target CPA per variant before any decision. Target CPA of €25? Each variant must have spent €50 to €75 before anything is said about it. Below that, you are reading noise. It is the same reasoning as Arthur Scaleyou's, who puts a test wave at €200-300 for 4 concepts with a target CPA of €25.

Hence the reverse calculation, which is the only honest way to answer the volume question:

Variants per angle = (Wave budget ÷ reading floor) ÷ number of angles tested.

A real example, February 2025, DTC accessories account, target CPA €22. Wave budget over 7 days: €1,400. Reading floor: €55 per variant. 1,400 ÷ 55 = 25 readable creatives. Divided by 5 angles in test = 5 variants per angle. The client wanted to order 12 per angle “to maximise the chances”. Sixty creatives for €1,400 of media comes to €23 per creative: none could have been judged. We produced 25. Three were validated, two became winners that lasted 11 weeks.

If the calculation gives you fewer than 2 variants per angle, the problem is not production: you are testing too many angles for your budget. Reduce the number of angles, not the floor. And if you are looking for the upstream reasoning, how much to put into production in total each month depending on your media tier, it is laid out in the calculation of the creatives budget by tier.

These tiers match what practitioners publish. Thortank, in a Meta scaling guide from €100 to €1,000 a day published about two months ago, describes 15 to 30 active creatives in rotation to sustain €1,000/day, renewed every 30 to 45 days, with 5 to 15 new ones in test in parallel. Adscale, in an article from about three months ago, sets the limit at 4 to 6 creatives per ad set.

What exactly do you vary, and what does not count as a variant?

Six axes produce information; a seventh produces none. Changing the colour of a banner, adding an emoji, swapping the music: these are not variants, they are versions. They take up an ad set slot, they use up reading floor, and they answer no question. We have refused them in briefs since 2024.

The marginal cost percentages come from our own 2025-2026 production rate cards on vertical UGC creatives. They are not published elsewhere and they vary with the level of finish: on a heavily edited mother creative, a hook swap costs proportionally less; on a raw creative filmed on a phone, it costs almost as much as the original because the creator has to be called back.

A point of method that many miss: you only vary one axis at a time in round 1. If you change the hook and the presenter between two variants, you have two different creatives and zero learning. In round 2, by contrast, combinations are fair game: the goal is no longer to learn, it is to cover the space of the validated angle.

The hook remains the most profitable axis to multiply, because it is the only one you can multiply without re-shooting. You still need to know what you are measuring on it and what it really costs: that is the subject of our article on measuring and costing a video hook.

The reading thresholds for a variant, according to Panja (2026 Meta Ads creatives guide, published two weeks ago): hook rate floor at 25%, hold rate between 40 and 50%, CTR alert below 1%, frequency alert above 3.

How much does a variant cost compared with the mother creative?

Between 5% and 110% depending on the axis, with a clear break at the moment you have to switch a camera back on. Anything that can be sorted in the edit stays under 35%. Anything that requires a new shoot goes back up to almost 100%, because you pay again for the creator, the briefing time, the equipment and the lead time.

This is the trade-off clients grasp fastest. When we present a wave quote, the line “5 angles × 1 mother creative” and the line “18 variants in the edit” are not at the same unit price, and they should never be. For the unit price of the mother creative by volume ordered, we have published the cost-per-creative grid by volume; here we are only talking about the differential.

The contractual lock nobody reads

A standard UGC creator package in France generally includes one edited video, two to three alternative hooks filmed the same day, and a set of retakes. Beyond that, you pay again. And the real issue is not the editing: it is the rights.

• Variants derived from the same rushes: covered by the original licence in the vast majority of contracts, as long as the duration and territory do not change. Check all the same that the word “variations” or “derived versions” appears in the contract; if it does not, you are in a grey area.
• New shoot with the same creator: new service, new licence, new fee. We see re-shoots negotiated at 40 to 60% of the original fee once the relationship is established.
• Extended running of a winner: this is where it costs. A creative that runs for 11 weeks often goes beyond the 3-month licence signed at the outset. Budget the renewal from round 2, not when the ad is switched off.

The item brands systematically underestimate: the briefing time for variants. Producing 8 variants of an angle requires 8 precise descriptions. If nobody writes them, the editor improvises, and you get the same video 8 times with different banners. Allow one hour of creative strategist time for 4 to 6 variants briefed properly.

At what point does one more variant stop paying off?

Beyond 8 variants of the same angle running at once, you are buying against yourself. Your variants do not fight each other in the auction (Meta only retains one per advertiser), but they fragment the budget and stop each one from leaving the learning phase. The system spends your budget separating near-twins instead of looking for new buyers.

Three signals tell you that you have passed the useful point. Frequency rises while reach stagnates. The average CPA of variants 5 to 8 converges towards that of variants 1 to 4 without ever beating it. And the ad manager concentrates 80% of the spend on two creatives, with the others never getting through the learning phase.

The Media Buyer pointed out in an issue from about six months ago that creative fatigue often sets in from the second time the same creative is shown to the same user. Between very similar variants, the effect is comparable: the user cannot tell your hook A from your hook B, they recognise the same ad. We have documented the real thresholds and the diagnostic errors in this article on video ad fatigue.

The preconception we dropped in 2025

For two years, the dominant practice was to produce 10 hooks on every validated video. We did it. It works, but only on an angle that is already winning, and the returns fall off quickly. On our accounts, hooks 1 to 3 bring most of the gain, hooks 4 to 6 bring something marginal, hooks 7 to 10 bring nothing measurable. We stopped selling them.

What we do not know, and what nobody really knows: which of the six axes will win on a given account. On a food account, it is almost always the visual proof. On a B2B service account, it is the presenter. On a fashion account, it is the format. There is no cross-cutting rule, and any agency that tells you in advance “the hook will make the difference” is selling you a hunch, not data.

A practical trade-off when budget is left over: a new angle is worth more than a ninth variant. The marginal cost of a new angle is higher (new mother creative, often a new shoot), but the expected gain is out of all proportion. That is what we observe in practice: a portfolio of varied angles feeds delivery better than a pile of near-twins.

How do you write the number of variants in a brief and in a quote?

One quote line per variation axis, never a single global “variations” line. As long as the quote says “20 creatives”, you do not know what you are buying and neither does your provider. When it says “5 mother creatives + 6 hook swaps + 4 proof variants + 5 4:5 reframes”, everything can be checked at acceptance. Google states in its October 2026 Demand Gen guide that advertisers who adopted creative variety, with horizontal, vertical and square formats, achieved on average more than 23% additional conversions.

In the brief, we add a field called “variant plan” containing four pieces of information: the angle in one sentence, the number of variants ordered, the variation axis chosen for each, and the reference mother creative. Four lines per angle. It takes ten minutes and eliminates 90% of editing back-and-forth.

• Angle: “You stop losing 40 minutes a day looking for your references.”
• Mother creative: 32-year-old creator, desk, 28 seconds, screen demo.
• Variant A: hook axis, opening on the figure instead of the problem.
• Variant B: proof axis, replacing the screen demo with a filmed customer review.
• Variant C: format axis, 15-second version, cut-heavy, for Reels placements.

On the deliverables side, demand three things: a file naming convention that contains the angle and the variation axis, masters without burned-in subtitles, and the rushes of unused alternative hooks. That last point is worth money: they are your future round 2 variants, and if you do not get them back, you will pay for them again.

If your brief does not yet have a solid structure, the nine-field format we use is published in full here; the variant plan has a field to itself.

And in organic, is the variants-per-angle ratio the same?

No, and it even reverses. In paid, the budget limits the number of variants you can test. In organic, distribution costs you nothing; what is scarce is the publishing slot and the attention of an audience that already follows you.

Operational consequence: in organic, you test 1 variant per angle and many angles. Twenty videos, twenty different promises, and you see which one lands. A follower who sees the same idea twice in three days unfollows; they are not under advertising pressure, they are under pressure from you.

But when an angle takes off in organic, the logic reverses again and becomes more aggressive than in paid: you turn it into a series. Ten, fifteen, twenty episodes of the same angle, with a recurring structure. What would tire out an ad set feeds an account, because the organic audience comes for a format, not an offer.

Do not transpose the paid grid onto your editorial calendar. We regularly see brands apply “3 variants per angle” to their TikTok posts and produce three near-identical videos in a week. The account stagnates, the brand concludes that the format does not work, when it is the variant rule that has been copied in the wrong place.

What to check before the next wave

The right answer to “how many variants per creative angle” is checked against three figures from your own account, not against a benchmark. Your target CPA, your wave budget, your number of live angles. The rest follows.

Put these three figures on a sheet of paper before sending any brief. If the calculation gives you 2 variants per angle across 6 angles, that is your plan, not 8, even if your provider offers a package of 8. If the calculation gives you 12, do not order 12: order 6, open two more angles with the rest, and keep the alternative hooks in reserve for round 2.

One last thing, which appears in no grid. The factor that makes the number of profitable variants vary most is neither budget nor sector: it is the quality of the starting rushes. A creator who delivered thirty minutes of usable material gives you eight solid variants at 15% marginal cost each. A creator who delivered forty usable seconds gives you two, and the third will be filler. The number of variants is decided on the day of the shoot, not on the day of the edit.

Frequently asked questions

How many variants should I produce if I am testing 5 angles with a €1,000 media budget?

With a target CPA of €25, the reading floor is €50 to €75 per variant. €1,000 divided by €60 gives about 16 readable creatives, or 3 variants per angle across 5 angles. Ordering more means some creatives will never pass the decision threshold and so will never really be tested.

Does a hook swap count as a real variant?

Yes, provided the opening line changes in logic and not just in wording. Replacing “You lose 40 minutes a day” with “40 minutes. That is what you lose every day” is not a variant, it is a rephrasing. A hook that opens on a figure when the original opened on a problem, on the other hand, tests something.

Should I stop an angle that has produced no result after 3 variants?

Yes, if it has received its full reading floor, that is 2 to 3 times the target CPA per variant. An angle tested with an insufficient budget is not a dead angle, it is an untested angle. A properly funded angle with no signal gets closed, archived, and can be re-tested six months later with another offer.

How many variants per angle to sustain €1,000 of daily spend?

Thortank, in a Meta scaling guide published about two months ago, describes 15 to 30 active creatives in rotation for this level of spend, renewed every 30 to 45 days. Spread over 4 to 6 validated angles, that makes 4 to 6 variants per angle in rotation, plus 5 to 15 new creatives in test each month.

Are variants included in a UGC creator's package?

A standard package generally includes one edited video, two to three alternative hooks filmed the same day and a set of retakes. Variants derived from the same rushes are covered by the original licence in most contracts. Any new shoot is a separate service and licence, often negotiated at 40 to 60% of the original fee.

Is a ninth variant or a new angle better?

A new angle, almost always. Its production cost is higher because it needs a mother creative and sometimes a shoot, but the expected gain is greater. Bidding systems select on the diversity of options: ten distinct concepts produce more signal than fifty variants of the same idea.