For scaling brands that are able to run at least $10k+/month in paid media.
The AI stack we actually use to run DTC ad accounts. Tools, costs, and where each one fails.
A prospect asked us: "Why shouldn't I just do this with ChatGPT?" Fair question. Here is the honest answer: the full stack, what each tool is good at, what it quietly gets wrong, and the one layer no tool covers.
"Just use ChatGPT" is not a strategy. It is one tool doing six jobs badly.
Running paid acquisition has six distinct jobs: research, copy, statics, video, automation, and measurement. Each job has a different best tool, a different failure mode, and a different amount of human review it needs.
Below is the working stack. Steal it. Most of it costs less per month than one bad ad test.
Research: mining customer language.
Claude
Strong: long-context synthesis. Paste hundreds of reviews, Reddit threads, and support logs into one conversation and extract pain language, trigger moments, and objections in their exact wording. Best-in-class at holding voice without flattening it into marketing speak.
Weak: it will find twelve plausible angles and argue convincingly for all of them. It cannot tell you which one your market believes this quarter.
Apify
Strong: the collection layer. Scrapers for Amazon reviews, Reddit, YouTube comments, and the Meta Ad Library, so the raw language arrives as structured data instead of a copy-paste marathon.
Weak: pay-per-run pricing punishes sloppy configs, and scrapers break when platforms change. Budget review time, not just credits.
The process: scrape with Apify, synthesize with Claude, extract five things only: first-person pain, trigger moments, objections, identity phrases, failed alternatives. Prompt with these targets. Prompt with a product description and you get product ads with zero customer psychology.
Copy: variants, not concepts.
Claude or ChatGPT
Strong: turning one proven winner into a dozen variants: new hooks on the same belief, platform cuts, length versions. An afternoon of work becomes twenty minutes.
Weak: net-new concepts regress to category cliches, and both models invent claims your product never made. Every output needs a human who knows what the brand can legally and honestly say.
Rule that saves accounts: AI iterates winners. Humans originate angles. The moment you flip that, output volume goes up and learning goes to zero.
Statics: production at volume.
Canva + an image model (GPT Image, Midjourney)
Strong: backgrounds, product-scene composites, and fast layout iteration on a proven format. Ten variants of a winning static in an hour.
Weak: text rendering inside generated images is still unreliable, and hands, labels, and product details drift. Generated product shots that misrepresent the product are a returns problem waiting to happen.
Video: the newest lever, the sharpest edge.
Seedance / Higgsfield-class video models
Strong: UGC-style hooks, product demos, and B-roll from a prompt plus product images. Testing ten hook variations no longer needs ten creator shipments.
Weak: physics glitches, uncanny hands, and lip-sync drift on longer clips. Fine for hooks and cutaways. Risky for full testimonials, and platforms are tightening disclosure rules on synthetic people.
CapCut
Strong: assembly, captions, and platform-specific cuts. The unglamorous 80% of video work.
Weak: template-driven editing makes your ads look like everyone else's. Templates are a starting point, not a style.
Automation and measurement.
Make or Zapier + Airtable
Strong: the pipeline around the ads: research ingestion, creative tracking, performance writebacks, alerts. This is where AI leverage compounds, because it runs while you sleep.
Weak: automations fail silently. A dead webhook can eat leads for weeks. Build a heartbeat check before you trust anything.
Meta Advantage+
Strong: delivery optimization. Let it find the buyer once the creative and the offer are right.
Weak: it optimizes toward whatever you feed it. Feed it blended ROAS and it will happily scale margin erosion. It does not know your CAC ceiling. Nothing in the stack does.
The optimal starter stack.
- Claude Pro or ChatGPT Plus for research synthesis and copy variants (~$20/mo)
- Apify for review and ad-library collection (usage-based, modest at DIY scale)
- Canva Pro + one image model for statics (~$15-30/mo)
- One video model subscription once statics are converting (~$30-70/mo)
- Make + Airtable free tiers for tracking and automation ($0 to start)
Under $150/month replaces most of what production-only agencies charge four figures for. That is the honest part most agencies skip.
What no tool on this list does: know which angle your market believes right now, remember why last quarter's winner was killed on purpose, or feel your cash conversion cycle before a scaling call. Every tool multiplies direction. None of them set it.
So when does a partner make sense?
Honestly: later than most agencies admit.
- Someone internal owns paid full-time and can be the human who rejects mediocre output
- Spend is under $10k/month, where structural leaks are usually smaller than a retainer
- You have a working angle and mainly need production volume behind it
- Fresh creatives keep shipping and keep dying at similar CTRs, and nobody can say why
- Scaling decisions run on blended ROAS with no CAC ceiling in sight
- Spend is $10k+/month, where one structural leak costs more monthly than the fix
That judgment layer is the part we sell. Human-led direction from research, with this exact stack multiplying the execution. If the left column describes you, take the stack and run. It is why we published it.
Before you buy a single tool, know your numbers.
Step 1: run your account through the calculator. Two minutes for your aMER, your CAC ceiling, and the monthly amount slipping through the gap. It tells you whether your problem is production speed or economics.
Step 2: if the gap points at the judgment layer, request the audit and we will read your actual account.
Run your numbers Request the full audit