AI content

AI product photos and ad creatives: 20 new pieces a month

This playbook makes 20 AI ad creatives a month from your product photos. They are AI product photos, short videos, ad images, catalogue images and social posts. A person checks each piece against the real product. You label AI content and test the pieces in ads. AI for Marketing Ops runs it for you with our AI content creation service.

Book a discovery call

Updated:

Who it is for
Ecommerce marketing teams that run Meta or Google ads and have real photos of their products.
Time
About 5 working days in week 1, then about 1 hour a week while the ads run.
Outcome
20 approved and labelled pieces, an ad test, and a list of the pieces to keep.
Tools
Your product photos · An AI image and video tool · Meta Ads Manager · Google Merchant Center · A shared spreadsheet

Steps

  1. 01 Day 1

    Choose products and list their facts

    Pick 5 products that already sell and collect your best photo of each. The AI works from these photos, so each piece is only as true as the photo behind it.

    List the facts the AI must never change: shape, colours, materials, sizes and the text printed on the product.

    What to do

    • Choose 5 products with steady sales.
    • Collect a sharp photo of each product on a plain background.
    • List the colours, materials, sizes and printed text of each product.
    • Save the photos and the facts in one shared folder.

    When it is done: A folder with 5 products, their best photos and a list of facts for each.

  2. 02 Day 1

    Plan 20 pieces by channel

    Decide where each piece goes before you make it. A product page, an ad and a catalogue each have their own size and rules.

    Give 4 pieces to each product. Each piece tries one idea, for example the product in use, a close detail or the product next to an object for size.

    What to do

    • Make one row per piece in a spreadsheet: product, channel, size and idea.
    • Plan a mix of product images, short videos, ad images, catalogue images and social posts.
    • Use the 4:5 format for Facebook Feed images. Meta lists 1440 x 1800 pixels for this placement.
    • Write one ad text per product now. Every piece of that product uses the same text, so the test compares only the pieces.

    When it is done: A spreadsheet with 20 rows, one for each piece.

  3. 03 Days 2 to 4

    Make the pieces from your real photos

    Make each piece with an AI image or video tool, always from the real product photo. A tool that draws the product from words alone invents the details.

    Keep the main product image for Google Shopping clean. Google rejects main images with added text, prices, logos or watermarks, so this text goes only in ad images.

    What to do

    • Upload the real product photo to the tool for every piece.
    • Make 2 or 3 versions of each piece and keep the one closest to the real product.
    • Show the product on screen from the first second of each video.
    • Name each file with the product, the channel and the piece number.

    When it is done: 20 draft pieces, saved by product and channel.

  4. 04 Day 5

    Check every piece against the real product

    A person compares each piece with the real product before anything goes live. AI tools change small details, such as a logo, a seam or a colour, and the buyer sees the difference when the parcel arrives.

    The review is also where you remove anything dishonest. A piece that shows a fake customer or a fake review never goes live.

    What to do

    • Compare the shape, colours, logo and printed text with the real product.
    • Read every word in the picture and in the ad text.
    • Reject any piece with a fake review, a fake customer or a copy of another brand’s ad.
    • Mark each piece as approved or rejected in the spreadsheet, with a reason.
    • Make new versions of the rejected pieces until 20 pieces pass.

    When it is done: 20 approved pieces, and a note on why each rejected version failed.

  5. 05 Day 5

    Label AI content and keep the metadata

    Mark each piece as made with AI, inside the file and where people see it. Google Merchant Center says every AI image must carry metadata that marks it as made with AI, such as the IPTC DigitalSourceType tag. You must not remove that metadata.

    Meta adds an "AI info" label to ads made or changed a lot with its own AI tools. Meta says it is starting to detect ads made with other AI tools and to label them too.

    What to do

    • Check that each AI image still carries the IPTC DigitalSourceType tag after you edit or compress it.
    • Send AI images to Merchant Center only in image_link, additional_image_link or lifestyle_image_link.
    • Add a short label such as "Made with AI" where a person or a scene looks real.
    • Keep a log of the tool and the date for each piece.

    When it is done: Labelled files and a log of how each piece was made.

  6. 06 Week 2

    Test the AI ad creatives in Meta ads

    Put the approved pieces into ads and compare them under the same conditions. Meta Ads Manager has an A/B test tool that compares versions of an ad.

    Change one thing per test: the piece. Keep the audience, the budget, the ad text and the dates the same, so the result tells you which piece works.

    What to do

    • Group the pieces by product, so each test compares pieces of the same product.
    • Use the same audience, budget and dates for every version.
    • Add the new catalogue images to your product feed.
    • Do not change the ads while the test runs.

    When it is done: A live test for each product, with one piece changed per version.

  7. 07 Week 4

    Keep the winners and plan the next 20

    Read the results and keep the pieces with the lowest cost per purchase. Stop the pieces that lose.

    Use what you learned to plan next month. The ideas that won for a product are the start of its next 4 pieces.

    What to do

    • Rank the pieces of each product by cost per purchase.
    • Keep the best pieces running and stop the rest.
    • Note which idea won for each product.
    • Plan next month’s 20 pieces around the ideas that won.

    When it is done: A list of winning pieces and a plan for the next 20.

Mistakes to avoid

  • Letting the AI change the product. If the colour or the shape in the ad differs from the real product, the buyer receives something else.
  • Putting prices, text or logos on the main Google Shopping image. Google rejects these images, so keep that text for ad images.
  • Losing the AI tag. An edited or compressed file can lose its metadata, so check the file after each export.
  • Changing many things in one test. When the picture, the text and the audience all change, you cannot tell what won.
  • Showing an AI person as a real customer, or adding fake reviews to a piece.

What to measure

What to measure in each monthly round
MetricHow to measureWhen
Pieces approvedCount the approved rows in the spreadsheet, out of 20Day 5
Pieces testedCount the pieces that ran in an ad testEnd of week 2
Click-through rate per pieceClicks divided by impressions, in Meta Ads ManagerWeek 4
Cost per purchase per pieceAd spend divided by purchases, in Meta Ads ManagerWeek 4
Rejected product imagesImages that Merchant Center does not approveEvery week

Rather have us run it

We run this playbook with you or for you

Sources

  1. Google Merchant Center Help: AI-generated content
  2. Google Merchant Center Help: Image link [image_link]
  3. Meta: Expanding GenAI transparency for Meta’s ads products
  4. Meta Ads Guide: image ad specs
  5. Meta Business Help Center: About A/B testing

Questions

Common questions

Can AI product photos replace a photo shoot?

For new scenes, sizes and formats, yes, if you start from one sharp and true photo of each product. The AI changes the scene around the product. It must not change the product. For a new product with no photo, you still need one real photo first.

Do I have to label AI images and ads?

Google Merchant Center says every AI image must carry metadata that marks it as made with AI, such as the IPTC DigitalSourceType tag. Meta adds an "AI info" label to ads made or changed a lot with its own AI tools. Meta is starting to detect other AI tools. Laws differ by country, so check with your legal adviser. AI for Marketing Ops labels all AI content clearly.

Why 20 pieces a month?

Twenty is the number of pieces we test in a month. In this playbook that is 4 pieces for each of 5 products. It is our working number, not an industry benchmark. With a small ad budget, test fewer pieces, because each piece needs enough impressions before you compare it.

Can AI for Marketing Ops run this playbook for us?

Yes. Our AI content creation service makes the pieces from your product photos. We define the buyer and the angles first. On a monthly plan, we make new pieces to test every month, with a 3-month minimum. A senior person on our team reviews every piece, and nothing goes live without your approval.

Book a 30‑minute discovery call

The call is free and online. Tell us which products need new images or videos. We check your store before we talk. You keep a short plan from the call, even if we do not work together.

Book a discovery call

30 minutes · Free · Online