Which Facebook Ad Creatives Work? Analyze Images and Copy With AI

Targeting and budget matter, but on Meta the creative, meaning the image or video and the words around it, often decides whether an ad works. The trouble is that Ads Manager shows numbers per ad, not why one ad beats another. AI can look at the numbers and the actual creative together and tell you what the winners have in common.

Who this is for

Anyone running more than a couple of ads who wants to know what's working, what's wearing out, and what to make next.

What you need

• AdChat connected to your AI app or the built-in chat.
• No write access needed. Everything in this guide reads data. You'll only need write access later if you want the AI to build new ads.
• A couple of weeks of data on at least a few ads, ideally in the same campaign so the comparison is fair.

The numbers that matter for creative

• CTR (click-through rate): the share of people who saw the ad and clicked. A good signal of whether the creative grabs attention.
• Cost per result (often called CPA): what you pay for each lead or sale. The number that pays the bills.
• Frequency: how many times, on average, each person has seen the ad.
• CPM: what you pay for 1,000 impressions. Rising CPM can mean Meta thinks people are less interested.

Step 1: Rank your ads

Show performance by ad for the last 30 days across my active campaigns: spend, impressions, CTR, results, cost per result and frequency. Sort by cost per result and mark the top three and bottom three.

This uses get_insights at ad level. What a good answer looks like: a ranked table with clear winners and losers, and a note when an ad has too little spend to judge.

Step 2: Look at the actual creative

Numbers alone don't tell you what to make next. Ask the AI to look at the ads themselves.

For my top three and bottom three ads, show me the creative: headline, primary text, call-to-action button and link. Then look at the images and describe what each one shows.

AdChat pulls the copy with get_ad_creatives and the images with get_ad_image, so an AI that can see images can describe them. For video ads it can fetch video details with get_ad_video, but it can't watch the video the way a person does, so describe the video yourself if that matters.

What a good answer looks like: a side by side description, like "Winners all show the product in use with a person; losers are product-only shots on white. Winning headlines lead with a benefit; losing ones lead with the brand name."

Step 3: Ask for the pattern

What do my best ads have in common that my worst ads don't? Think about the image style, the first line of text, the offer, the length of the copy and the call to action. Give me three clear lessons.

This is where AI earns its keep. You want specific lessons, not "use high-quality images".

Step 4: Check for creative fatigue

Creative fatigue is when people have seen an ad so many times they stop responding. The usual signs are frequency climbing while CTR falls and cost per result rises.

For each active ad, compare the last 7 days with the 7 days before: frequency, CTR and cost per result. Which ads look fatigued?

A rough rule many advertisers use: if frequency keeps rising and CTR drops week over week, start preparing a replacement. There's no magic frequency number. A small retargeting audience will naturally see ads more often than a broad cold one.

Step 5: Break down by placement

The same creative can do very differently in the Facebook feed, Instagram feed, Stories and Reels.

Break down my top five ads by placement for the last 30 days. Where does each one work best and worst?

A square image might shine in feeds and look awkward in Stories. If one placement clearly underperforms, the fix might be a vertical version, not a new idea.

Step 6: Turn lessons into a brief

Based on what's working, write a short brief for three new ads: describe the image I should use, write the primary text and headline for each, and explain which lesson each one tests.

Now you have a plan grounded in your own data.

What to do next

• Build those new ads: new ad from a winning ad. That step needs write access turned on, and new ads are created paused.
• Pause the clear losers: find wasted ad spend.
• Make creative checks part of your weekly check-up.

Common mistakes

• Comparing ads from different campaigns. A retargeting ad will usually beat a cold audience ad. Compare like with like.
• Judging on tiny spend. An ad with $8 spent and one sale isn't a winner yet.
• Only looking at CTR. Clickbait gets clicks. Cost per result tells you whether those clicks turn into customers.
• Waiting too long to refresh. Have the next creative ready before the current one wears out.
• Changing several things at once. If the new ad has a new image, new copy and a new offer, you won't know which one made the difference.

FAQ

Can the AI really see my ad images? If your AI app supports images, yes. AdChat fetches the image so the AI can describe it. If it can't, you'll still get the text, links and numbers.

Can it analyze my competitors' ads? This guide is about your own ad account. AdChat works with the ad accounts you connect.

How often should I refresh creative? It depends on audience size and budget. Small audiences and big budgets wear ads out faster. Let the fatigue check in Step 4 tell you.

Can AdChat make images or videos for me? No. It can suggest what to make and write the copy. You create the image or video, then AdChat can build the ad from an image link or an existing post.

Try it free for 7 days. Sign up and find out what your best ads have in common. Questions? Email [email protected].