Plan, make and call your creative tests

Creatify Agent writes the test plan, makes the variants, launches them once you approve and calls the winner with a 95% range. It reads your ad platform’s own results; there’s no separate A/B tool. Start free, no card.

The rule for a winner, written before launch

Hypothesis, one variable, budget, length and decision rule, sized to the budget you actually have.

Ad Agent /Creative test plan

Plan my next creative test for the dew drops. $100 a day, $35 target per purchase.

Worked for 1m 52s

Read the product page: $36, 4.8★ from 2,361 reviews
Picked one variable: the claim in the first 3 seconds
Sized it: $33 a day per variant, 14 days
Wrote the rule for calling a winner
Done

Run it as a Meta A/B test, not in your main campaign: Meta treats three near-identical openings as one and piles spend onto a single ad.

If the first 3 seconds lead with a proof claim, hook rate rises about 20%. Winner: hook rate +15% at 95%, 10+ purchases at $42 or less, no call before day 7.

Creative test plan · Hook claimPlan · Nov 2–15

⚠️ You didn’t ask, but a $35 target on a $36 product leaves close to $0 from the first order. Push the bundle too.

Suggested follow-ups

Plan only: nothing in your account changes.

Three hooks on one body

Only the first 3 seconds change, so the test reads the hook and nothing else.

Ad Agent /Hook variants

Make all three variants.

Worked for 6m 05s

Cut one 15-second 9:16 body: routine, texture, result
A: a drop catching light on the cheek, no claim
B: “Sephora’s #1 Niacinamide Serum” on screen
C: “97% agreed… more glowy”, quoted from your page
Done

Three 15-second videos, identical after 0:03: same voiceover, captions, CTA and landing page.

Claims are quoted word for word from the product page. File names match the test plan, so results map back to each hook.

Hook test · A, B, C · 15 s · 9:163 videos · ready to download

Saved in this chat. Nothing in your ad accounts changed.

Suggested follow-ups

Making ads changes nothing in your ad accounts.

Each hook on its own budget

No A/B test in Ads Manager? The agent gives every hook its own budget, so Meta can’t move spend off one of them.

Ad Agent /Meta launch

Launch the three hooks on Meta.

Worked for 1m 31s

Uploaded the 3 videos with one ad text for all
Picked the Purchase event on your pixel
Gave each hook its own campaign at $33 a day
Put all three launches on one approval card
Done

Even spend per hook, same audience and landing page. Audiences can overlap, so it isn’t a true split; a Meta A/B test is.

Recommended actions3

Review each before approving. AI can make mistakes.

Dismiss allApprove all

Dew Drops hook test – A (no claim)Sample StoreLaunch Campaign

Sales · Purchase · Advantage+ audience · 1 video ad.

Daily budgetUSD 33/day

Dew Drops hook test – B (Sephora’s #1)Sample StoreLaunch Campaign

Daily budgetUSD 33/day

Dew Drops hook test – C (97% glowy)Sample StoreLaunch Campaign

Daily budgetUSD 33/day

Waiting for your approval. Each goes live once you approve and Meta’s ad review passes.

Suggested follow-ups

Nothing changes in your ad accounts until you approve it.

Every morning: ready to call, or not yet?

A routine reads the test daily and says so plainly when there isn’t enough data to call it.

Ad Agent /Hook test check

Hook test check · daily 9:00 AMRoutine

Worked for 1m 12s

Read 5 days of spend, hook rate, CTR and purchases
Checked each hook against the 10-purchase floor
Worked out a 95% range on hook rate vs A
Checked spend stayed even across the three
Done

Not enough data yet: 15 purchases across three hooks after 5 days, none with 10. C leads on hook rate, but not at 95%.

Spend so far$497of $1,400 · day 5 of 14

Purchases15A 4 · B 5 · C 6

Hook rate · C vs A+18%95% range −4% to +41%

CallNot yetno call before day 7

Next check tomorrow at 9:00 AM. Nothing in your account changed.

Suggested follow-ups

Read only: nothing in your account changes.

A winner with a 95% range, not a hunch

Each variant gets a range on cost per purchase, and thin data gets “not proven”, even when it looks far behind.

Ad Agent /Hook test verdict

Call the winner of my hook test. Target is $35 per purchase.

Worked for 1m 40s

Read 9 days of spend, CTR and purchases for 3 hooks
Worked exact 95% ranges on cost per purchase
Rechecked A vs B: about 98% confidence, not 99%
Checked how much more spend B would need
Done

Hook A wins: its whole range sits under your $35 target. B’s range spans it. C has 6 purchases at $70 each.

Hook test · Meta, 9 days · sample

Variant	Spend	CPA (95% range)	Verdict
A · “I stopped using primer”Problem · 52 purchases	Spend$1,180	CPA (95% range)$22.69 ($17–$30)	Winner
B · “Derm reacts”Authority · 27 purchases	Spend$1,090	CPA (95% range)$40.37 ($28–$61)	Testing
C · “Bundle and save”Offer · 6 purchases	Spend$420	CPA (95% range)$70.00 ($32–$191)	Cut

Even $3,000 more wouldn’t settle B, so cap it. Stop C on cost, not proof. Next: new problem hooks vs A.

Suggested follow-ups

Read only: nothing in your account changes.

What a clean creative test looks like

Four decisions, made before a dollar is spent. Skip one and the result can’t tell you what worked.

1 · HypothesisIf we change X, metric Y moves by about Z%“Written down with the reason, so a loss teaches you something too”
2 · One variableA hook, a claim, an offer or a format“Everything else stays identical: body, voice, copy, CTA and landing page”
3 · Budget and lengthEnough for 10+ conversions per variant“At least 7 days to cover a full week; rarely more than 28”
4 · Decision ruleThe winner’s 95% range has to clear the target“Set before launch, with a floor on conversions and no early calls”

How tests usually go wrong

✕Ten near-identical variants, so each gets too little spend to read
✕Winner called on day two because one ad is ahead
✕Hook, offer and format changed at once, so nobody knows what worked
✕Variants briefed out and back a week later

How Creatify Agent runs them

✓Three to five genuinely different concepts or hooks, sized to the budget
✓No call under 10 conversions; the winner’s 95% range has to clear your target
✓One variable per A/B test, written down with the hypothesis before launch
✓Made in the same chat, then launched into your account on an approval card

Meta A/B test, ABO or your main campaign?

Where a test runs decides what it can tell you. The agent picks one of three setups and says why.

One campaign · 3–5 distinct concepts

New concepts: inside your main campaign

Meta now reads what each ad is about and finds people for it, so distinct concepts added to one broad campaign test themselves. Splitting them into thin ad sets starves each of the ~50 conversions a week it needs to learn.

Even split · hypothesis written first

One variable: a Meta A/B test

A claim, an offer, a price or a landing page needs a clean read. Meta’s A/B test splits the audience evenly; Meta suggests about $100 a day per variant, so the agent sizes the plan to your budget.

Graduate on ~95% confidence

Already run a test + scale pair? Keep it

If your ABO testing campaign feeds a scaling campaign and it works, the agent won’t restructure it mid-flight. A winner moves over only after 7 days, 10+ conversions and a range that beats your target.

Exactly what it does in a test

The plan, the variants and the call happen in chat. Anything that touches your ad account waits on an approval card.

What the agent sticks to

✓Not a separate testing toolThere is no built-in A/B engine. The agent plans the test, does the math and reads your platform’s own results.
✓Nothing runs until you approveEach launch, pause or budget move sits on a card with its reason and numbers. Unanswered cards expire after 72 hours.
✓Budget steps, not jumpsScaling a winner goes up about 20% at a time. Raises over +30% on Meta, +20% on Google or +50% on TikTok are flagged.
✓Fatigue checks on a scheduleSet a weekly routine and the agent checks every winner for rising frequency and falling CTR, then reports by email, Slack or chat.

Creatify Agent and creative testing

Hypothesis, variable, budget, length, decision ruleWritten in chat

Hook, format and concept variants (video and image)Made in chat

Spend, CTR, purchases, CPA, ROAS, frequencyRead

Video watched 25–100% (Meta)Read

Winner call with 95% rangesIn chat

Launch variants as new ads or a new campaignAfter you approve

Pause the loser, move budget to the winnerAfter you approve

Set up a Meta A/B test (Experiments)You, in Ads Manager

DeletesNever; it pauses

Trusted by brands, agencies, and platforms.

Creative testing questions

Running different versions of an ad (concepts, hooks, formats, offers) against each other with real spend to learn which one gets results at the lowest cost, then putting more budget behind it. A good test starts with a hypothesis, changes one thing, and sets the rule for calling a winner before launch.

Usually three to five, and they should be genuinely different. Each variant needs enough spend to collect conversions, so on a small budget fewer variants read faster than many thin ones. Near-identical variants mostly compete with each other.

Enough for each variant to reach at least 10 conversions, and more if you want to detect a small difference. Meta suggests about $100 a day per variant for its A/B test. The agent works the numbers from your target cost per purchase and daily budget, and says plainly when the budget is too thin to answer the question.

At least 7 days, to cover a full week, and rarely more than 28, after which seasonal shifts muddy the read. Learning phases matter too: about 3–7 days on Meta and around a week on TikTok.

When the leader has at least 10 conversions and its 95% range on cost per purchase beats your target, or beats the other variants. Under 10 conversions, the agent says “insufficient signal” rather than picking a winner, even when one ad looks far ahead.

It depends on the question. Distinct new concepts can go into your main broad campaign, where delivery spreads spend across them. A single variable such as a claim, offer or landing page needs a Meta A/B test for a clean split. If you already run an ABO testing campaign that feeds a CBO scaling campaign and it works, keep it.

No. Creatify doesn’t run its own split-testing engine. Creatify Agent designs the test, makes the variants, launches them in your ad account when you approve, and reads the platform’s results to call the winner with confidence ranges. For a true split on Meta, you create the A/B test in Ads Manager.

No. It doesn’t score ads with a synthetic panel or a prediction model. It can review a draft against what has worked in your account and in competitors’ long-running Meta ads, but the call is made on real results.

It reads results from Meta, TikTok, Google Ads, Snapchat, Reddit, OpenAI Ads and AppLovin, and can launch new ads on all of them except AppLovin, always after your approval.

You can start free with no card. Connecting an ad account is free; the agent runs on credits, and the Free plan includes 10 a month. Free video exports carry a watermark.

Plan your next creative test

Paste a product link or your last test’s results, and get a plan or a verdict in one chat.

Plan, make and call your creative tests

The rule for a winner, written before launch

What a clean creative test looks like

How tests usually go wrong

How Creatify Agent runs them

Meta A/B test, ABO or your main campaign?

New concepts: inside your main campaign

One variable: a Meta A/B test

Already run a test + scale pair? Keep it

Exactly what it does in a test

What the agent sticks to

Creatify Agent and creative testing

Trusted by brands, agencies, and platforms.

Creative testing questions

What is creative testing?

How many creatives should I test at once?

How much budget does a creative test need?

How long should a creative test run?

When can I call a winner?

Should I use a Meta A/B test, ABO or CBO for creative testing?

Does Creatify have built-in A/B testing?

Can it predict which ad will win before I spend?

Which ad platforms does it work with?

Is it free?

Plan your next creative test