Ad Creative Testing

New ad concepts every month and a read on which one wins

We research the angles, produce a batch of short video ads from your product photos and footage, set up a clean test in your Meta ad account, and send a one-page readout two weeks later.

It starts with a pilot of ten ads for $997. Ad spend stays on your own account.

Two-week readout Example data
ConceptHook rateCost
Problem firstWinner
31%
$24
UnboxingTry new hooks
27%
$31
How it worksTry new hooks
22%
$38
ComparisonDropped
18%
$52

Cost is cost per purchase. Next month keeps the winner and tests four new concepts against it.

An example readout comparing four ad concepts on hook rate and cost per purchase. The numbers are invented to show the layout.

What is ad creative testing?

Ad creative testing is running several different ads against the same audience and the same budget to see which idea sells, then putting the money behind the winner. The thing being tested is the concept, which is the reason a stranger should care. One ad opens on the problem. Another shows how the product works. A third compares it with the cheap alternative. Each concept gets a few different opening lines, called hooks, because the opening seconds decide whether anyone watches the rest.

After a set period you compare the concepts on the same measures: how many people watched past three seconds, how many clicked, and what a purchase cost. The winner gets more budget. The losers are switched off, and what they taught you goes into the next batch.

It is not the same as making more ads. Twenty versions of one idea with a new colour or caption tell you almost nothing. Four ideas that are really different tell you what your customers respond to.

See the loop in 34 seconds

A tired ad, four new concepts, and the readout two weeks later. No sound needed.

  1. 0:00An ad that has run too long
  2. 0:05Why the idea is now the hard part
  3. 0:10Four concepts and three hooks for each
  4. 0:17The readout after two weeks

The product, the ads and every number in the video are examples, not results.

Why does the creative matter more than the targeting?

Because the platform now does most of the targeting for you, and anyone can make a video. Here is what has been measured or said on the record.

49%

of the extra sales that advertising produced came from the quality of the creative. Targeting accounted for 11%. Source: NCSolutions, Five Keys to Advertising Effectiveness, 2023 update, nearly 450 US consumer goods campaigns on TV and digital.

15M

ads were made with Meta’s generative AI tools in a single month, by more than a million advertisers. That was in late 2024. Source: Meta Engineering, 2 December 2024.

“Increased ad diversity can improve people’s experience with ads and drive better advertiser outcomes.”

Meta Engineering, on its Andromeda ad retrieval system, December 2024

Meta’s system now picks which ad to show each person from a far larger pool than it used to. That works in favour of an account with several different ideas in it, and against one that runs a single ad until it wears out.

Making the video is no longer the hard part either. On 6 October 2026 Meta said that generating video from still images is now generally available inside Advantage+ creative. A competitor can turn a product photo into a video in the same afternoon you can. What they cannot copy as fast is knowing which idea works for your product.

Read these with care. The 49% comes from campaigns for groceries and household goods, measured by a company that sells ad measurement. It was not measured on Meta alone and it will not transfer exactly to your brand. Meta’s statements are Meta’s own. Meta has never published an official number for how many different ads an ad set should hold, so we do not quote one.

Source for the October announcement: Social Media Today, 6 October 2026.

What a batch looks like

Every batch starts from four concepts. Each one is a different reason to buy, written as one sentence you can argue with before any video is made.

Each concept then gets three hooks, which are three different ways to open the same idea. That gives twelve video ads. If one concept wins, the hooks tell you which opening carried it.

From the second month the batch also holds four new variations of the previous winner, so the ad that is working gets fresh versions before it wears out.

One month’s batch Example

Insulated bottle4 concepts12 video ads

Problem first3 hooks

  • Warm water by 10am?
  • Your bottle is lying to you
  • Three hours and it gave up

How it works3 hooks

  • Two walls with no air between
  • Why the ice lasts 24 hours
  • Cut one in half and look

Unboxing3 hooks

  • What is in the box
  • First fill, first sip
  • It arrived like this

Comparison3 hooks

  • Ours against a $9 bottle
  • Same ice, six hours later
  • Which one would you carry?
An example batch for an insulated bottle: four concepts, each with three opening lines. The product and the lines are invented.

How is this different from an AI video tool or a creator?

Both are good at producing videos. Neither decides what to test or reads the result, and that is where most of the value sits.

A general comparison. Tools and creators differ, so check the one you are considering.
Compared onAI video toolFreelance creatorsCoremagna creative testing
What you getAs many videos as you care to generate, within minutes.A handful of videos from real people, usually within a few weeks.A planned batch every month: four concepts with three hooks each, cut for feed, Stories and Reels.
Who picks the conceptsYou do.You do, or the creator improvises.We do, from your reviews, your past results and the ads your competitors run. You approve them.
Who sets up the testYou do.You do.We do, in your ad account, or we hand a setup sheet to your media buyer.
Who reads the resultYou do.You do.We send one page that says what to scale, what to switch off and what to test next.
People on cameraAvatars or generated presenters.Real people.Product shots and demos by default. Real people only from your own footage or a creator you choose.
Your partPlanning, prompting, editing, launching and analysing.Briefing, chasing, editing, launching and analysing.Approve the concepts, then read one page.

How a month works

Five steps. The pilot runs the same way with a smaller batch.

  1. Research

    We read your reviews, your last three months of ad results and the ads your competitors are running in the Meta Ad Library. The output is four concepts, each with one sentence on why it might work.

    Days 1 to 3
  2. Sign-off

    You see the concepts, the hooks and every product claim before anything is produced. Anything you cannot stand behind comes out.

    Day 4
  3. Produce

    Twelve short video ads from your product photos and footage, plus four new variations of last month’s winner. Each one is cut for feed, Stories and Reels.

    Days 5 to 10
  4. Launch

    We set the test up in your ad account, or hand the files and a setup sheet to your media buyer. Every concept gets the same audience and the same budget, so the difference you see comes from the creative.

    Day 11
  5. Read

    After 14 days you get one page: hook rate, click rate and cost per purchase for each concept, what to scale, what to switch off, and the brief for next month.

    Day 25

What the readout tells you

Four numbers for each concept and one decision. Every number comes from your own Ads Manager, so you can check it.

Hook rate
3-second video plays ÷ impressionsWhether the opening stops the scroll. It is the first number to settle, and it needs the least budget.
Hold rate
ThruPlays ÷ 3-second video playsWhether people who stopped kept watching. Meta counts a ThruPlay when a video plays to the end or for at least 15 seconds.
Click rate
Link clicks ÷ impressionsWhether the ad made people want to see the product page.
Cost per purchase
Spend ÷ purchases Meta attributes to the adThe number that pays the bills, and the noisiest one. The readout shows how many purchases it rests on, so a result built on six sales is not read like one built on sixty.
Verdict
Scale, try new hooks or dropOne line for each concept, and a brief for the next batch that follows from it.

What the videos look like

Stills from finished product videos in our portfolio, made from product photos with AI video tools and edited by hand. They show the production quality. They are not test results.

Watch these and the rest of the portfolio

What we need from you

Enough to make the ads and read the numbers. Nothing that lets us spend your money or see your customers.

For a test

  • Partner access to your Meta ad account at advertiser level, or none at all if your media buyer launches the ads
  • Product photos, any footage you have, your logo and brand colours
  • Ad results for the last three months read access or an export
  • The claims you can prove ingredients, test results, guarantees, review counts
  • One person who approves concepts and claims

We never ask for

  • Admin control of your Meta business account
  • Your card or billing details ad spend stays on your account
  • Customer lists or any personal data
  • The admin login for your store

What we cannot promise and what we can

Nobody can promise that an ad will sell. We promise different ideas, a fair test and an honest reading of it.

No return promises

We will not quote a return on ad spend before the test has run. The readout shows what happened, including the concepts that lost.

No invented customers

No AI presenter poses as a buyer, and no review or before and after is made up. Claims come from you, with proof.

You approve every ad

Concepts, scripts and claims are signed off before production, and the finished ads before launch.

Your account stays yours

We work as a partner on your ad account. Billing, pixels and customer data stay with you.

No lock-in

The pilot is a one-time fee. The monthly plan runs month to month and stops with 30 days’ notice.

Every file is yours

The videos, the briefs and the readouts belong to you and can keep running if we stop working together.

What it costs

Two prices. Ad spend is separate and goes straight from you to Meta.

Monthly testing

$1,997a month

A new batch every month and a readout on the last one.

  • Four new concepts, researched and briefed
  • 12 new video ads: four concepts with three hooks each
  • Four new variations of last month’s winner
  • Every ad cut for feed, Stories and Reels
  • The test set up in your ad account, or handed to your media buyer
  • A one-page readout and the next brief
Book a call

Pilot

$997one time

Ten ads delivered within 14 days: five concepts with two hooks each. We set up the test and send a readout once it has run for two weeks. The quickest way to see whether this suits your account.

Start with the pilot

Ad spend

Paid by you to Meta, never through us. For a test worth reading we suggest about $100 a day while it runs. With less, we rank concepts on hook rate and click rate and say so in the readout.

Prices are in US dollars. The cost of generating the videos is included. There is no setup fee and no minimum term on the monthly plan.

How far would cost per purchase have to fall to cover it?

We do not forecast results. What you can check is whether the sum makes sense for your account before you spend anything.

If a winning concept lowers your cost per purchase, you get the same number of sales for less. The saving is your monthly ad spend multiplied by the size of the drop. Move the slider to what you spend on Meta each month.

If the answer is a large number, this service is too early for you, and the page should say so.

13.3%

At that spend, a 13.3% lower cost per purchase covers the monthly fee.

Monthly fee $1,997. This is arithmetic, not a forecast.

AI-made ads and the rules

An ad made with AI is still an ad, and the existing rules apply to it. In the United States the FTC’s rule on reviews and testimonials has banned fake ones since October 2024, and it names reviews written by AI for people who do not exist.

In the United Kingdom fake reviews have been banned outright since April 2025, and the advertising code requires that an ad does not mislead. In the European Union the transparency rules of the AI Act began to apply in August 2026. They require realistic AI-generated video to be disclosed as such.

In practice that means product shots and demonstrations are the default. If a presenter appears, they speak as the brand and never as a customer. We do not write reviews, invent results or fake a before and after. Where a platform or a law asks for an AI label, the ad carries one.

We are not lawyers and this is not legal advice. The final call on any claim is yours.

Common questions

Anything not covered here, ask on the call.

No. A test can show that none of the new concepts beats what you already run, and sometimes it does. What we promise is four ideas that are different from each other, a test set up so the result can be read, and a readout that says plainly what won and what did not.
For a pilot we suggest about $100 a day for the 14 days the test runs, paid by you to Meta. That is our working rule and not a Meta requirement. With less, the test still ranks concepts on hook rate and click rate, but cost per purchase will rest on too few purchases to trust, and the readout will say so.
No. We set up the test, read it and tell you what to scale. Day-to-day buying stays with you or your media buyer. If an agency runs the account, we send them the files and a setup sheet and work from their numbers.
Yes. We build them from your product photos and any footage you have, using AI video tools, and a person edits every one. You can see finished examples on the product videos page. Where a real person is needed on camera, we say so and use footage from you or from a creator you choose.
Meta first: Facebook and Instagram, in feed, Stories and Reels. The same files can be cut for TikTok and YouTube Shorts if you ask. The readout covers Meta.
A concept is a reason to buy: it solves this problem, it works this way, it beats that alternative. A variation is the same reason with a different opening line, caption or cut. Testing variations of one concept tells you which wording is best. Testing concepts tells you what your customers care about, which is the bigger question.
The share of people who were shown the ad and watched at least three seconds of it: three-second video plays divided by impressions. It is the earliest sign of whether an opening works, and it needs far less budget to measure than cost per purchase.
Then you have learned four things not to spend more on, which is worth knowing. The readout says so in plain words, and the next brief moves to different ground. We do not dress up a flat test as a win.
You do. Every video, brief and readout is yours to keep and to keep running, whether or not you continue with us.
A link to the product, the photos and footage you have, a list of claims you can prove, and either partner access to the ad account or a media buyer who will launch the ads. A 30-minute call covers the rest.

Sources

Every figure and rule on this page, with where it came from. Checked on 11 October 2026.

  1. NCSolutions, Five Keys to Advertising Effectiveness, 2023 update. Creative 49%, brand 21%, reach 14%, targeting 11% and recency 5% of incremental sales, from nearly 450 US consumer goods campaigns on TV and digital.
  2. Meta Engineering, the Andromeda ads retrieval engine. More than a million advertisers made more than 15 million ads with Meta’s generative AI tools in a month. Published 2 December 2024.
  3. Social Media Today, Meta’s announcements at Advertising Week 2026. Video generation from static assets in Advantage+ creative is generally available. Published 6 October 2026. This is press coverage and not Meta’s own page.
  4. US Federal Trade Commission, final rule banning fake reviews and testimonials. Announced August 2024, in force from 21 October 2024.
  5. UK Competition and Markets Authority, guidance on fake reviews. The ban took effect in April 2025.
  6. UK Advertising Standards Authority, the CAP Code.
  7. European Commission, the AI Act. Transparency rules apply from 2 August 2026.

Start with ten ads and one honest readout

Bring a product link and your last three months of ad results to the call. We will tell you whether a pilot makes sense for your account, and say so if it does not.