Paste an ad, get a critique
By the end of this lesson you will have pasted a real ad screenshot into free claude.ai and received a structured media-buyer critique: four scored parts and the one change that matters most. It works on your phone, costs nothing, and is worth screenshotting and sending to someone.
Last lesson you described ads in words. That filter is gone now: Claude can look at the ad itself. Open claude.ai on your phone, tap the plus button in the chat box, and attach a screenshot. A free account is all this needs.
Five rules make image chats work properly. They come from how Claude actually reads images, so bake them in from day one:
- Image before text. Attach the screenshot first, then type your prompt under it. Claude reads the message top to bottom; the image should be what the instructions refer back to, not an afterthought.
- Label your images. The moment you send more than one, write "Image 1: the full ad. Image 2: close-up of the price." Unlabeled images get confused with each other.
- You get up to 20 images per message on a free account, each up to 10MB. You will rarely need more than two or three, but a whole competitor teardown fits in one message.
- Crop and zoom for small text. If the fine print, the price, or a badge matters, crop that region on your phone and send the close-up as its own labeled image. A full-screen shot squeezes small text into mush.
- Claude cannot tell whether an image is AI-made, and it cannot generate images either. It judges what is in the frame, not where the frame came from. Never ask "is this ad AI-generated"; the answer is a guess.
The critique prompt
This is the prompt you will reuse for the rest of the course. It gives Claude a job, a fixed rubric, and rules that keep it honest.
Attach one of your three stopper ads from lesson 1, then paste:
You are a senior performance media buyer. You have reviewed thousands of ad creatives and killed most of them. You are blunt and specific, never polite for its own sake.
Image 1 is an ad. Critique it with this exact rubric:
1. HOOK (score 0-10): would this earn a stop from a stranger in under 2 seconds? Justify in one line.
2. CREATIVE CLARITY, MUTED (score 0-10): with the sound off, can someone tell what is being sold and who it is for? Justify in one line.
3. OFFER STRENGTH (score 0-10): what is the actual deal, in the ad's own words? If there is no deal, only a product, score it 3 or below. Justify in one line.
4. CTA (score 0-10): does the viewer know exactly what happens when they tap? Justify in one line.
5. THE ONE CHANGE: the single edit with the biggest impact on performance. One sentence, specific enough to act on today.
Rules: quote the ad's own words whenever you refer to its text. If any text in the image is too small or blurry to read, say so instead of guessing. Do not assume the product is good. Score honestly and let the numbers spread; a polite 7 on everything helps nobody.You should see
Five numbered sections: four scores, each with a one-line reason that quotes words visible in the ad, then one concrete change. Something like "HOOK 4/10: the first frame is a logo, and logos stop nobody" tells you it is working. A warm paragraph about how nice the ad looks tells you it is not.
Now run the rest of your six. The three stoppers should mostly score well on hook; the three you skipped should score badly somewhere specific. When the scores match your own thumb, you are seeing what a media buyer sees. When they do not, read the justification; one of you noticed something the other missed, and it is not always Claude.
Check it
- All four scores present, each justified in one line
- At least one justification quotes words actually visible in the ad
- The scores spread; not everything sits at 6 to 8
- THE ONE CHANGE is a single concrete edit, not "improve the hook"
- Anywhere text was unreadable, Claude said so instead of guessing
If you see
If Claude praises an ad your own thumb skipped, the role has gone soft. Tighten it: add "if this ad is weak, say so plainly; you are rewarded for finding the flaw others miss" and resend. If it says text is unreadable, or invents a price you cannot see either, crop that region, zoom, and resend it as "Image 2: close-up of the offer text". If you get an essay instead of the rubric, reply "answer only in the five numbered sections, nothing else".
Ten minutes in, you have a repeatable critique that most people running ads never get. Grab a screenshot of your favourite verdict and send it to someone who boosts posts and hopes.
What you just learned
- Attach the image first, label every image, and crop small text into its own close-up.
- Free claude.ai takes up to 20 images per message; two or three well-labeled ones beat twenty unlabeled.
- A role plus a fixed 0-10 rubric turns "nice ad!" into a critique you can act on.
- Claude judges what is in the frame; it cannot tell AI-made from photographed, and it cannot make images.
practice
Open Claude Cowork on your own computer and try this on real files. Each module ends with a checkpoint where you submit what you made.
Open in ClaudeNext
Steal the formula