To create ads with AI, start from a written brief (offer, audience, proof, brand rules), use AI to generate many hooks, copy and visual variations from that brief, have a human check every claim and the brand fit, then launch the variations in a structured test and keep only the ones that drive qualified leads or sales. AI makes the volume cheap. It does not make the strategy, and it does not know which ads produce revenue unless you feed results back to it.
This guide gives you a seven-step workflow, prompts you can copy, a brand rules checklist, the platform labeling rules for Meta, TikTok and Google, and a test plan with real budget math.
What can AI actually do in ad creation today?
AI tools now cover most of the production chain. The useful split is between what saves time and what still needs a person.
| Task | What AI does well | What still needs a human |
|---|---|---|
| Hooks and headlines | Dozens of angles in minutes | Picking the angles that match a real pain |
| Primary text and descriptions | Length variants per placement | Checking every claim is true and allowed |
| Static images | Backgrounds, resizes, layout variations | Using real product and real people where it matters |
| Video | Image-to-video, AI avatars, voiceover, captions | Script judgment and disclosure |
| Translation and localization | Fast drafts in many languages | Native review (Mexico vs Colombia vs Spain) |
| Performance analysis | Summarizing which variants won | Connecting wins to revenue, not clicks |
The ad platforms also build AI into their own tools. Meta's generative AI features in Ads Manager include background generation, image expansion, image generation and text variations. Google's Performance Max asset generation can suggest text, image and video assets from your website, and Google says you can review every generated asset before it runs.
How to create ads with AI: a seven-step workflow
- Write a one-page brief. Offer, price or starting price, audience, top three objections, proof you can legally show, and the one action you want (form, WhatsApp, call).
- Load your brand rules. Tone, banned words, colors, fonts, logo usage, and claims legal has approved. AI follows rules you write down; it ignores rules that live in someone's head.
- Generate angles before assets. Ask for 15 to 20 hooks grouped by angle: pain, outcome, objection, comparison, social proof, urgency. Keep 4 to 6 angles.
- Produce variations per angle. For each angle, make 2 to 3 formats: static image, short vertical video, and a carousel or text-heavy variant.
- Human review. Check claims, prices, disclaimers, faces, hands, text rendering and logo accuracy. Remove anything that implies a testimonial you do not have.
- Label and launch. Apply AI disclosure where the platform requires it, name every ad with angle and format so you can read results later, and launch in a test structure.
- Feed results back. Pass the winners and losers, with cost per qualified lead or cost per sale, into your next brief. This is the step most teams skip.
Which prompts work for ad copy and visuals?
Good prompts are specific about the audience, the constraint and the format. Three templates to adapt:
Hooks: "You write direct-response ads for [business] selling [offer] to [audience] in [country]. Their main objection is [objection]. Write 15 hooks under 12 words, grouped by angle (pain, outcome, objection, proof). No claims about results we cannot prove. No exclamation marks."
Primary text: "Using hook #4, write three versions of Meta primary text: 40 words, 90 words and 150 words. End each with one call to action: 'Send us a WhatsApp'. Use 'tú', neutral Latin American Spanish. Include the starting price [price]."
Visual: "Vertical 9:16 image for a Reels ad. Real product photo [attached] on a clean kitchen counter, morning light, space at the top third for a 6-word headline. No text in the image. No people."
Two habits improve output fast: paste in your three best-performing past ads as examples, and ask the model to explain which angle each variation uses. That makes the review step quicker.
What brand rules should every AI ad follow?
| Rule | Why it matters | How to check |
|---|---|---|
| Only approved claims | AI invents numbers and guarantees easily | Compare against a written claims list |
| Real prices and terms | Wrong prices create complaints and ad rejections | Match the landing page on launch day |
| No fake testimonials | The US FTC rule on fake reviews covers AI-generated testimonials | Every quote traces to a real customer with permission |
| Logo and color accuracy | Image models distort logos | Overlay the real logo file after generation |
| Readable text | Generated text inside images is often misspelled | Add text in an editor, not in the image prompt |
| Disclosure when required | Platforms label or reject undisclosed synthetic media | Platform checklist below |
The FTC's rule on consumer reviews and testimonials, effective October 21, 2024, bans fake consumer reviews and testimonials, and the FTC said AI-generated reviews are covered. If you sell in Mexico or elsewhere in LATAM, local consumer protection rules also apply, so check with counsel.
Do AI-generated ads need a label on Meta, TikTok and Google?
- Meta. Meta's help center says ads fully created or materially edited with Meta's own generative AI features get an "AI info" label, shown next to "Sponsored" when the image includes an AI-generated photorealistic human. In its February 2025 announcement (updated June 2026), Meta said it would also detect ads made with third-party AI tools through industry signals and label them in "About this ad". Political and social issue ads have their own stricter disclosure requirement.
- TikTok. TikTok's ad disclaimers page lists AI-generated, synthetic or manipulated media as a mandatory disclaimer case when content is fully AI-generated or significantly modified by AI.
- Google. Assets from Performance Max generation are reviewed by you before they run, and they must follow standard Google Ads policies.
Policies change often. Check each platform's current page before launch rather than relying on any article, including this one.
How should you test AI-generated ads?
Test angles first, then formats, then small edits. Testing 40 tiny variations at once spreads the budget so thin that nothing gets enough data.
Example (hypothetical numbers): a lending company in Mexico spends $60,000 MXN a week on Meta lead ads. Its cost per qualified lead (a lead the sales team accepts) is about $600 MXN. It wants each test cell to reach roughly 20 qualified leads before deciding.
| Item | Calculation | Result |
|---|---|---|
| Budget per cell for 20 qualified leads | 20 × $600 MXN | $12,000 MXN |
| Cells the weekly budget supports | $60,000 ÷ $12,000 | 5 cells |
| Test design | 4 new AI angles + 1 current best ad as control | 5 cells, 1 week |
| Decision rule | Keep angles at or below control's cost per qualified lead | Usually 1 to 2 winners |
Notice that the decision uses qualified leads, not cost per lead. AI ads with bold hooks often bring cheaper leads that sales rejects. Our guide to optimizing Meta ads for qualified leads explains how to send that signal back to Meta, and creative-level attribution shows how to tie each ad to revenue.
Where does Admira fit?
Admira's Growth plan includes AI Ads, an AI ad creative generator, next to attribution that ties Meta, Google Ads, TikTok and Click-to-WhatsApp leads to CRM revenue. The practical benefit is step 7: you can see which creative produced sales, not just clicks, and feed that into the next round. Admira is a young brand with few public reviews, and if you only need high-volume video avatars, a dedicated tool may suit you better. Our comparison of the best AI ad generators covers the options, and pricing lists what each plan includes.
Who this is for, and who it isn't
- For: teams running Meta, TikTok or Google ads that need more creative variety than their designers can produce, and that can measure results past the click.
- For: agencies producing variations for several clients with clear brand rules.
- Not for: regulated claims (health, finance, legal) without a compliance review step.
- Not for: brands whose main asset is authentic founder or customer content; AI can support it, not replace it. See our guide to AI UGC video ads for where synthetic presenters fit.
FAQ
Can I create ads with AI for free?
Yes, to a point. Meta's generative AI features in Ads Manager and Google's Performance Max asset generation do not charge extra, and many standalone tools have free tiers with watermarks or limits.
Will Meta penalize AI-generated ads?
Meta labels qualifying ads with "AI info"; its public pages describe this as transparency, not a ranking penalty. Ads still have to follow Meta's advertising standards.
How many AI variations should I test at once?
As many as your budget can feed to a decision. Divide weekly budget by the cost of about 20 target results to find the number of cells.
Can I use AI to write fake customer reviews in ads?
No. In the US the FTC rule effective October 21, 2024 bans fake reviews and testimonials, including AI-generated ones. Use real customers with permission.
Should I generate text inside images?
Usually not. Image models still misspell words. Generate the visual, then add headlines and prices in a design tool.
What is the best metric to judge AI ads?
Cost per qualified lead or cost per sale from your CRM. Click-through rate and cost per lead can reward hooks that attract the wrong people.
.svg)
%20(2).png)
