From competitor ads analysis to a variant batch
An AI ads agent should not start by generating. It should start by reading. On Moatt, a Variant factory run opens with a decode: the agent studies the ads your competitors are running — the angles, the hooks that repeat, the formats they keep paying for. What they quietly stopped running matters just as much. An ad a competitor killed after two weeks is a test result you get without spending a cent.
Only then does it draft. The batch is written against the decode, not from a blank prompt: variants that answer a competitor's angle, sharpen a hook they left soft, or take a format they proved into your own offer. Every draft arrives labeled with the reasoning, so you can tell a deliberate counter from a coin flip.
Then everything stops. The batch waits at the approval gate with its credit price on the screen. You cut the weak ones, approve the rest, and only approved variants go anywhere. There is no setting where this step gets skipped.
- Decode — the agent reads competitor ads and maps angles, hooks, and what got killed
- Variants — a batch drafted against the decode, each with its reasoning attached
- Approve — the run pauses at the gate, price shown; only what you approve fires
Media pricing, honestly
Most tools that generate AI ad creative are vague about what an asset costs: pricing hidden behind a trial, or credits sold without a published rate for what one image or one video actually burns. You find out on the invoice.
Moatt meters media per finished asset and publishes the number: a batch of 30 images is 5 credits, a finished 30-second video variant is 2. Media costs more than text runs, by design — rendering images and video burns real compute, and hiding that inside a flat rate means someone else's video habit sets your price. The price shows at the approval gate before the run fires, so the cost is a decision you make, not a surprise you absorb.
Two rules keep it fair: failed runs refund automatically, and talking to the agent costs nothing. Briefing the batch, arguing with drafts, tightening the angle — all free. Only runs bill.
What a month of ad variant generation costs
Concrete cadence, priced from the public rate card: refresh one competitor profile, generate one 30-image batch, and produce two video variants each week. That is 12 credits a week, roughly 48 a month — inside the Starter plan's 60 credits at $99. A heavier program across several competitors fits the Growth plan's 250 credits at $299.
The billing guardrails matter more than the tier. You get alerts at 50, 80, and 100 percent of your credits. Unused credits roll over up to twice your monthly amount, so a quiet month funds a heavy testing month. If you do run over, overage is billed at 1.25x and capped at 3x your plan. A big testing push cannot turn into an invoice you did not see coming.
- Competitor profile, weekly — 3 credits
- Image variant batch, 30 images — 5 credits
- Two video variants — 4 credits
- Total: 12 credits a week, about 48 a month
What stays human
Budget, first. The agent never sets spend, moves spend, or touches a bid. It drafts variants, publishes the ones you approved to a connected account, and pulls results back to your board. Money decisions do not pass through it. If automated spend rebalancing inside one ad platform is the job you need done, tools built for exactly that do it well — keep one.
Strategy stays yours too. The decode hands you facts about what competitors run; deciding what to do about those facts — which market, what offer, what your brand will and will not say — is judgment. An agent is good at producing variants of a direction. It is not good at picking the direction, and this page will not pretend otherwise.
Connecting the ad account works the same as everywhere in Moatt: the agent asks for the account when a run needs it, you approve in an OAuth popup, and the agent never sees a password.
When an ads agent is the wrong buy
If you need one hero film for a brand launch, hire people. Variant generation is built for volume against measured feedback — many attempts, fast reads, keep what earns. A single high-stakes asset that has to be right the first time is the opposite shape of problem.
If your spend is too small to read results per variant, a 30-image batch teaches you nothing; the data cannot separate winners from noise. Start with a handful of variants and grow the batch as spend grows. And if you already have a creative pipeline that ships work you like, keep it — bring the agent in for the competitor decode and the testing tail, where volume is the point.