Per-token costs are trivial. Per-outcome costs are everything.

AI content is cheap by the token and expensive by the outcome. The bill that matters is the second one, and almost nobody computes it correctly.

Two different bills

A LinkedIn post drafted by a current-generation model costs about a quarter of a cent. The same post written by a freelance copywriter costs $40 to $80 of working time. Three orders of magnitude apart. This is the number every AI-content vendor cites in their pitch deck.

The number that actually matters is the outcome cost: what does it cost to ship a post that performs at parity with a human-written one? That bill includes the iterations needed to fix register, the drafts you throw out, the off-brand posts that ship anyway and quietly erode trust. It is harder to flatter and harder to fake.

What flips the outcome cost

Three things, in order of impact.

Brief schema. The model needs the same input a human writer would need: pillar, voice, hook, key points, CTA, source. Most teams give models a thinner brief than they would give a person and then blame the model for shallow copy. It is an input problem dressed up as a model problem.

Voice contract. A written specification of what your brand sounds like, enforced as a system prompt going in and a scanner running against every output. Drafts that fail the scanner get rewritten before a human ever sees them. This is the layer that prevents the slow brand drift that sinks most AI-content programs by month three.

Observability. Logs structured around outcomes, not around triggers. When a draft is bad you need to know in seconds whether the brief was thin, the voice gate caught it, or the model regressed. Generic LLM logging tells you which API call returned what. It does not tell you which brand decision broke.

The numbers, ours

240 social drafts over 30 days. Total model cost: about $0.60. Operator time: 90 seconds per draft of review, plus 8 iterations on the ones that needed sharpening. Call it $35 of human time per ten posts shipped, plus a dollar of model spend.

A freelance copywriter on the same brief shape would run about $400 per ten posts and a multi-day turnaround. The unit economics flip somewhere around six posts a week of cadence. Below that, a person is fine. Above it, the AI pipeline pays for itself in week one.

What does not flip the math

Bigger models. More expensive providers. Custom fine-tunes. None of those matter if the brief is thin or the voice contract is missing. Most teams over-invest in the token layer and under-invest in the input and output layers. Get the inputs and outputs right and the smallest competent model is more than enough.

Get them wrong and the most expensive model on the market still produces mid copy. The model is rarely the bottleneck.