Our AI Ad Creative Workflow: 40 Ads in an Afternoon (Without Looking Like AI)

Every ad platform now runs on the same basic rule: the algorithm needs a steady supply of fresh, varied creative to find your customers. Meta, TikTok and Google’s Performance Max all reward advertisers who can test more concepts, more often. For most small teams, that is the single biggest bottleneck. There is no designer on staff, no video crew, and nobody has twelve spare hours a week to write ad copy.

AI changes that math, but only if you use it properly. Most people open an image generator, type “ad for my product,” and get something glossy, generic and instantly recognizable as AI. Audiences have learned to scroll past that look. This post walks through the workflow we actually use with students to produce around 40 usable ads in an afternoon, and the guardrails that keep them looking like they came from a sharp human creative team.

Stage 1: Mine Real Customer Language (30 Minutes)

The quality of AI output is capped by the quality of your input. The fastest way to make AI ads sound human is to start with words humans actually used. Before we write a single prompt, we collect raw material:

• Product reviews (yours and your competitors’, especially the 3-star ones, which are full of honest nuance)
• Sales call notes or transcripts
• Comments on your past ads and organic posts
• Support tickets and frequently asked questions
• Reddit threads and forum posts where your customers talk about the problem

Paste that material into your AI assistant of choice and run an extraction prompt:

Below are customer reviews, comments and call notes for [product].
Extract and organize:
1. The top 7 pains or frustrations, in the customer's exact words
2. The top 5 desired outcomes, in their exact words
3. Objections or hesitations before buying
4. Surprising or specific phrases that would make a great ad hook
5. Any "before vs after" descriptions

Quote directly wherever possible. Do not paraphrase into marketing language.

[PASTE RAW MATERIAL]

The output of this step becomes your “voice of customer” document. Save it. You will reference it in every prompt for the rest of the afternoon, and you will reuse it for months.

Stage 2: Build the Concept Matrix (30 Minutes)

Generating 40 ads that are all secretly the same ad is the most common waste of time we see. The algorithm treats near-duplicates as one idea. So before producing anything, we map out deliberately different concepts.

We use a simple grid: angles down one side, formats across the top.

Angle	UGC-style video	Static image	Carousel	Founder video
Pain agitation	Yes	Yes
Social proof / reviews		Yes	Yes
Us vs. them comparison		Yes	Yes
Behind the scenes / origin story				Yes
Objection crusher	Yes			Yes
Offer / urgency		Yes

Pick around 8 cells. Then use AI to flesh out each one:

Using the voice-of-customer document below, write a creative brief for each of
these 8 concepts: [list angle + format pairs].

For each brief include:
- Target persona (one sentence)
- Core message (one sentence)
- 3 hook options (under 10 words, using real customer phrases)
- Visual direction (what is on screen, setting, who appears)
- Proof element (review, stat from OUR data, demo, guarantee)
- Call to action

[PASTE VOICE OF CUSTOMER DOC]

A human now reviews the briefs and cuts or rewrites anything that feels off. This 10-minute edit is the highest-leverage moment of the day. Strategy errors here multiply across every downstream asset.

Stage 3: Produce Copy at Scale (45 Minutes)

With approved briefs, copy is the easiest part to scale. For each concept we generate primary text in three lengths, five headlines and the on-screen text or script.

Write ad copy for concept: [paste brief].
Platform: [Meta feed / Reels / TikTok].
Deliver:
- 3 primary text versions: short (under 40 words), medium (60-90), long story (150-200)
- 5 headlines under 40 characters
- If video: a 20-30 second script with timestamps, on-screen text, and a hook
  in the first 2 seconds

Rules: write at a 6th-grade reading level. Use the customer's words from the brief.
No exclamation marks in headlines. Avoid these words: revolutionary, game-changer,
unlock, elevate, seamless, cutting-edge, delve.

That banned-word list matters more than people think. Certain words have become strongly associated with AI writing, and readers notice them even when they cannot articulate why. Maintain your own list and add to it whenever something sounds robotic.

The human pass on copy

Do not publish raw output. For each concept, a person should spend two or three minutes doing three things: tighten the hook until it is impossible to ignore, add one specific detail only your business would know, and read it out loud. If it sounds like a brochure, rewrite it until it sounds like a person talking.

Stage 4: Generate and Assemble Visuals (90 Minutes)

This is where most AI ads go wrong, and where your workflow choices matter most. Our core principle: use AI to extend real assets, not replace them.

Static images

• Start with real product photos. Use AI tools to change backgrounds, generate lifestyle scenes around the real product, extend the canvas for different aspect ratios, or remove clutter. The product itself stays authentic.
• Design text-forward statics. Some of the best-performing static ads are simple: a bold customer quote, a screenshot of a review, a plain-text “notes app” style message. These are fast to make in any design tool and look nothing like AI.
• Use generative images for concepts, not products. AI-generated scenes work well for illustrating a problem state (a messy desk, a stressed parent) where photorealism of a specific product is not required.

Video

• Repurpose existing footage. Most businesses have more usable video than they realize: phone clips, old webinars, product demos, customer testimonials. AI editing tools can find highlights, auto-caption, resize and add B-roll quickly.
• Record simple selfie video. A founder or team member filming three 20-second hook variations on a phone often outperforms anything polished. AI writes the script; a human delivers it.
• Use AI avatars and voices cautiously. They can work for explainer-style content, but disclose where platforms require it, and test them against human-delivered versions rather than assuming they will perform.

We maintain an up-to-date list of the tools we use for each of these jobs on our AI ad tools page, since the landscape shifts every few months.

Stage 5: The “Doesn’t Look Like AI” Quality Check (30 Minutes)

Before anything goes live, every asset goes through a checklist. These are the most common giveaways we look for:

Visual tells

• Hands, fingers, teeth or text inside the image that are subtly wrong
• Overly smooth, plastic-looking skin or impossibly perfect lighting
• Product details that differ from the real item (logos, labels, colors, proportions)
• The same glossy, symmetrical “stock AI” composition repeated across ads
• Backgrounds with nonsensical objects when you zoom in

Copy tells

• Words from your banned list sneaking back in
• Lists of exactly three adjectives in every sentence
• Generic claims with no specifics (“high-quality,” “trusted by many”)
• Em-dash-heavy rhythm and rhetorical questions in every line

Compliance and accuracy

• Every claim, number and testimonial is real and verifiable
• No AI-generated “customers” presented as real people giving testimonials
• Platform policies for your category (health, finance, housing, employment) are respected
• AI-generated content is labeled where the platform requires it

That last group is non-negotiable. AI makes it easy to fabricate reviews or before-and-after results. Doing so can get your ad account restricted, damage your brand, and in many jurisdictions create legal exposure. Use AI to present real proof better, never to invent it.

Adapting Each Ad for Every Platform

A single concept rarely works identically everywhere. The same idea needs different packaging for Meta, TikTok and Google, and AI makes that adaptation fast. Before the quality check, we run every approved concept through a short platform pass.

Platform	What to adjust	Common mistake
Meta feed	Square or 4:5 crops, a clear headline, primary text that front-loads the hook	Burying the key benefit below the “See more” cut-off
Reels and TikTok	9:16 vertical, captions inside the safe zone, hook in the first second, native creator tone	Reusing a polished horizontal video with black bars
Google Performance Max	Multiple aspect ratios, short headlines, logo variants, at least one real video	Letting the platform auto-generate video from statics
YouTube Shorts	Vertical, slightly longer setup tolerated, strong verbal hook	Assuming TikTok trends translate directly

We use one simple prompt to handle the copy side of this:

Here is an approved ad concept and its copy: [paste].
Adapt it for: (1) Meta feed, (2) TikTok, (3) Google Performance Max.
For each platform, rewrite the hook, the on-screen text and the CTA to fit
the platform's native style and character limits. Keep the core angle and
proof point identical. Flag anything that might conflict with that
platform's ad policies.

Keeping the core angle identical across platforms has a side benefit: when you analyze results, you can see whether an idea is strong everywhere or only on one platform. That insight tells you where to invest more production effort next time.

Resizing without losing the message

AI outpainting and smart-crop tools make it easy to produce every aspect ratio from one master image. Check each version manually. Automated crops regularly cut off the product, a face or the most important line of text. Thirty seconds of review per asset prevents a lot of wasted spend.

A Sample Afternoon Schedule

Here is how the time actually breaks down when a trained operator runs the workflow:

Time	Task	Output
1:00–1:30	Mine customer language	Voice-of-customer doc
1:30–2:00	Concept matrix and briefs	8 approved briefs
2:00–2:45	Copy generation and human pass	Primary text, headlines, scripts for all 8
2:45–3:45	Static images	~20 statics across sizes and hooks
3:45–4:30	Video edits and hook variations	~20 short videos
4:30–5:00	Quality check and naming	40 launch-ready ads

Your first run will take longer. That is normal. By the third or fourth afternoon, your prompts, templates and brand assets are saved, and the same workflow usually takes noticeably less time.

Name everything for learning

Use a consistent naming convention so you can analyze results later. We use a format like Concept_Angle_Format_Hook_Version, for example “C3_Comparison_Static_PriceHook_v2.” When results come in, you can instantly see which angles and formats are winning across the whole batch, not just which individual ad got lucky.

Turning the Workflow Into a Weekly System

Forty ads in one afternoon is impressive. Forty ads every two weeks, informed by what won last time, is what actually grows an account. The loop looks like this:

1. Launch the batch into a structured creative test (see our creative testing framework).
2. Read results by angle and format after enough spend to be meaningful.
3. Double down: feed the winning angles back into Stage 2 and produce new executions of them.
4. Explore: dedicate a portion of each batch, perhaps a quarter, to completely new angles so you keep discovering.
5. Update your voice-of-customer doc with new comments and reviews monthly.

Over time, your saved prompts and briefs become a creative operating system. That is far more valuable than any single winning ad.

If you want the full template library, including our brief templates, banned-word list, naming conventions and platform-specific prompt packs, it is all inside the AI ad creative module of the program. To get feedback on your current creative first, request a free ad audit. And when you are ready to have coaches review your actual ads every week and help you build this system for your business, apply to the AI Ads Coaching Program.