AI Ads: How to Make, Test and Run Them on Meta

Cover card with the title AI Ads: How to Make, Test and Run Them on Meta, in white text on a dark background with a cyan glow

AI ads are ads that AI helps make, run or show. Most advertisers use AI to make many images and videos fast, then test them against each other. AI speeds up the making. Your own test decides what wins.

For help setting up the workflow, see my AI sales enablement page.

What are AI ads?

The phrase covers three different things, and most articles mix them up:

Meaning What it is Who does the work
Ads made with AI AI writes the text and generates the images and video You, with an AI tool
Ads run by AI The ad platform picks who sees which ad and how much to bid The platform
Ads shown inside AI Sponsored placements inside an AI assistant’s answers The AI company

The first is the one you control, so it is the one this guide covers. The second is already normal: Meta’s own tools decide a lot of delivery for you. The third is new. A BCG article from January 19, 2026 says OpenAI announced it would begin testing ads inside ChatGPT for U.S. users, and that Walmart and Amazon have put ads inside their own shopping assistants. If you sell products, watch it, but you cannot run a test there the way you can on Meta today.

How do you make AI ads on Meta?

Use the creative tools already inside Meta’s ad platform. On Meta’s Q3 2026 earnings call (October 21, 2026), its CFO said nearly 10 million advertisers used at least one of its generative AI creative tools in the quarter. She also said adoption of its video generation features grew 25% from the previous quarter, helped by ready-made videos that can be generated in seconds from a single image upload.

That tells you what the tools are built for: take one image you already have and turn it into more ads. A practical order looks like this:

  1. Start from a real product or offer photo. AI changes and extends what you give it. A weak starting image gives you more weak ads.
  2. Generate a few versions, not fifty. Change the hook, the background or the angle, and keep the rest the same.
  3. Turn your best image into a short video. Meta’s tool does this from one upload, so a video test costs you very little extra work.
  4. Write the text yourself first, then ask AI for variations. A general assistant like Claude or ChatGPT is good at ten ways to say your offer. For starter prompts, see AI prompts for business.
  5. Look at every ad before it runs. Odd hands, warped text and made-up details are easy to miss.

When I need a list of prospects or investors, I ask ChatGPT, Gemini, Claude, Perplexity and every other AI model the same prompt and compare the answers. Do the same with ad ideas: give two assistants the same brief and keep the better angle.

Do AI ads work better than normal ads?

Not on their own. That is the honest answer, and the sources back it up. On the Q3 2026 call, Meta gave adoption numbers for its creative tools, but no conversion result for them. The conversion lifts it did mention came from its ad-ranking model and a separate business assistant, which are different tools.

So an AI ad is not better by default. What AI changes is the cost of trying more ideas. Five versions used to mean five photo shoots. Now it means an afternoon. That only pays off if you have a way to find the winner, which is the next section.

I come at this the way I did at my past companies. When we disagreed about an ad, we looked at the data and A/B tested it. Whatever the data says wins, not who yells the loudest or who has the fanciest title. AI does not change that rule. It just gives you more things to test.

How do you test AI ad creative fast?

Change one thing at a time, give each version enough time and budget to produce results, and judge by the number that matters to you, not by which ad you like. A simple test plan:

Step What to do What to avoid
1. Pick the goal One number: cost per lead, cost per sale or return on ad spend Judging by likes or views
2. Pick one variable The hook, the image, the offer or the video length Changing the image, text and offer together
3. Make a few versions Two to four versions that differ only in that variable Fifty versions that split the budget
4. Run them side by side Same audience, same budget, same dates Running one this week and another next month
5. Wait Let each version collect results before you decide Calling a winner after two days
6. Keep the winner, test the next variable New round, new version against the old winner Starting from scratch every time

On patience: a common paid-ads mistake is impatience. When you change an ad, it does not change the account overnight. We used to panic two days later and change it back. Give changes time to work before you decide something failed.

On risk: my number one rule when someone new touches an ad account is do no harm. Historically people made a million changes at once, restructured everything, and then everything crashed. Test new AI creative next to your current best ad. Do not replace the ad that is already paying for itself.

Whatever you test, track every step from click to customer. In a class I taught, I used a $100 Facebook ad as the example: I want to know how many people clicked, how many came to the site, how many signed up, how many booked a demo, how many showed up and how many became customers. If I spend $100 and get $105 back, I made money. That is all you are looking for, and you cannot see it unless you track the whole path.

Video: The Sales & Marketing Playbook Video · 74 min The Sales & Marketing Playbook I show how I track every marketing dollar from the first click to a paying customer, so you can tell which ads earn their budget. Watch the video →

Which AI tools make ads?

Pick the tool by the job, and start with the one you already have access to. Search results for this term list generators such as Recraft, Arcads, invideo and Predis.ai, and Meta builds its own into Ads Manager. I have not tested each of these, so treat the table as a way to sort the options, not a ranking.

Job Kind of tool Good first test
Turn one product photo into more ad images and a short video Meta’s built-in creative tools Use the same photo you run today and compare the new versions to your current ad
Write ten versions of the hook and the text A general assistant such as Claude or ChatGPT Give it your offer, your customer and one past winner
Make videos with a presenter or voice-over A specialist video generator Make one video, then check the voice and the claims word by word
Make ads for many sizes and placements A design tool or the ad platform’s resizing Check that no text is cut off

The tool matters less than the process. A better generator does not fix an offer nobody wants.

What should you check before an AI ad goes live?

Check three things: the platform’s label, the law on fake testimonials, and how your customers feel about AI.

Meta’s label. Meta says it labels ads made or significantly edited with its generative AI creative tools. A newsroom post first published February 3, 2025 and updated June 1, 2026 says the label sits behind the ad’s three-dot menu when the tools significantly edit the image or video, and appears next to the “Sponsored” label when the tools add an AI-generated photorealistic human. Meta also says it will label ads edited with third-party AI tools using industry-standard signals. You do not add this label yourself, so do not assume your ad will look unlabeled.

Testimonials. Do not use AI to invent a customer. The FTC’s rule on consumer reviews and testimonials took effect October 21, 2024. In its own Q&A, the FTC says the rule has no blanket ban on AI-generated avatars, but an avatar giving a testimonial is prohibited if the underlying testimonials are fake or false. Using a celebrity’s avatar to praise your product without permission would also violate the rule. AI is fine for a background or a layout. It is not fine for a made-up happy customer.

Your customers. A Smartly study, cited in that BCG article, found 69% of consumers feel manipulated when brands use AI in advertising without disclosing it. That is one survey, so do not treat it as a law of nature, but it is a good reason to be open about it.

Because rules and labels change, read Meta’s and the FTC’s pages again before a big campaign. If your ads touch regulated areas, ask a lawyer.

What mistakes ruin AI ad tests?

The biggest one is treating the ad as the whole job. I split paid advertising into three buckets: pre-click (the campaign and the keywords), the click (your landing page), and post-click (what happens after they become a lead). Most people only focus on pre-click, and they fail at the other two. AI makes the first bucket cheaper. It does nothing for a slow landing page or a lead nobody calls back.

Other common mistakes:

  • Testing too many things at once. You get a pile of results and no lesson.
  • Judging too early. Two days of data is a mood, not a result.
  • Dropping the old winner. Keep your best ad running while the new ones prove themselves.
  • Skipping the review. Read every ad, every claim and every image closely. You are responsible for what runs under your name.
  • Chasing cheap clicks. A cheap click often means little intent behind it, so judge by cost per lead or sale.

What should you do next?

This week, take your best current ad, make three AI versions that change only the hook, and run them against it with one number to judge by. Write down the date you will check the result. Then add the next variable.

For the wider picture, see AI marketing tools for a small business, AI for business, AI lead generation and how to measure AI ROI. If you want a person to build the workflow and train your team to run it, start on my AI sales enablement page or my AI consulting page.

Sources

FAQs

What are AI ads?

AI ads are ads that AI helps make, run or show. Most advertisers mean the first: using AI to produce ad images, videos and text quickly. The other two are AI that picks who sees an ad, and ads that appear inside AI assistants such as ChatGPT.

Can you make AI ads on Meta?

Yes. Meta's own ad tools can generate creative and turn a single image into a video. On its Q3 2026 earnings call, Meta said nearly 10 million advertisers used at least one of its generative AI creative tools in the quarter.

Do AI ads perform better than normal ads?

Not automatically. On that same call, Meta gave adoption numbers for its creative tools but no conversion result for them. AI makes more versions faster, and only your own test shows which version earns more.

Do you have to label AI ads?

On Meta, the platform adds an AI info label itself when its tools significantly edit an image or video, or add an AI-generated photorealistic human. The FTC also bars fake or false testimonials, and AI does not change that.

How many AI ad versions should I test at once?

Start with a few versions that differ in one clear way, such as the hook or the offer. Too many versions split your budget until none has enough results to compare. Add more only after you have a winner to beat.

Get new resources in your inbox

Consulting

Grow your business with a founder who’s done it

David coaches entrepreneurs and advises companies as a consultant and fractional Chief AI Officer.

Speaking

A speaker your audience will remember

Motivational keynotes & workshops on entrepreneurship, AI, financial literacy, and more.