What producing content with AI actually costs

Pricing for automatic generation tools is almost always in credits, minutes or words, and almost never in pieces of content. Buyers want to know "what does a post cost me"; sellers answer "it depends". This article tries to explain what it depends on, with real numbers.

Why nobody sells "ten posts a month"

The provider's cost has two components, and neither is measured in content:

• transcription, which depends on the length of the audio;
• generation, which depends on how much text is produced and which model produces it.

A forty-minute video costs about four times a ten-minute one to transcribe. But the article that comes out costs nearly the same in both cases, because an article's length is decided by the format, not by the length of the video.

That asymmetry is what makes per-content pricing impractical: to offer it, a provider would have to price in the worst case — very long videos, many outputs — and charge everybody for it, including the people uploading ten minutes a week.

The order of magnitude of the real costs

Some numbers measured on our own infrastructure, to give the proportions. They are not prices: they are costs, what we pay the model providers.

The generation cost of one complete job — transcript, social posts, article, chapters, newsletter, summary — is in the order of a few euro cents. The single long-form article is the most expensive line in the package, and on its own it weighs more than all the social posts put together.

⚠️ The most useful thing to understand is that this cost depends very little on the length of the video. Sixty minutes spent on one long video cost the provider far less than the same sixty minutes spread over twelve short ones, because every job starts the generations from scratch. It is why trial plans have not only a cap on minutes but also a maximum number of jobs: without it, the worst case would cost ten times the average one.

What you are actually paying for in a subscription

If generation costs cents and the subscription costs tens of euros, the difference is not pure margin. It is: infrastructure, development, support, compliance, tax — and, in most companies in this sector, the funding of growth.

That leads to a practical consequence worth more than any price comparison: today's aggressive prices are not a commitment for tomorrow. When raising capital gets harder, price lists realign with costs. One of our comparisons became inaccurate within three days: a price-list detail we thought was stable wasn't.

When you choose, consider how sustainable the price you are looking at is, not only how low.

How to tell if it is worth it, in one question

It isn't "how much content do I produce". It is "how often do I upload something".

A subscription is worth it to whoever has a cadence: an episode a week, two lessons a month, a video every ten days. A cadence fills the plan and comes out ahead.

Whoever works in waves — three videos in March, then nothing until September — pays for empty months on a subscription. In that case a small plan plus top-ups makes more sense, or waiting until there is a real cadence.

⚠️ Check one thing before subscribing: what happens to unused minutes. In most price lists in this category they reset every month. It isn't unfair — it is how nearly all subscriptions work — but it changes the sums if you were planning to accumulate them.

The comparison worth making

Not price against price, but price against your own alternative cost.

If producing one episode's package by hand takes half a day, and half a day is worth something, the arithmetic is quick. If instead you don't produce that content at all today, the comparison isn't with time saved: it is with the value of content that does not currently exist — which is harder to estimate, and almost always higher than it looks.

---

If you want a feel for it on your own files, echoflow's trial runs 14 days and asks for no card. For how the work is structured, the place to start is Repurposing video.