What it actually costs
Change any assumption and every number below recalculates. The defaults are 50 customer sites, each with 50 blog posts and 4 images per post.
Read this first. These are engineering estimates built from token counts
and published rate cards, not quotes, and pricing changes. Instrument your first five real
builds and replace these numbers with measured ones before anything here reaches a
pricing page. The shape of the model is far more reliable than the digits.
Your assumptions
1 · Building the content
Paid once per site. Nothing here recurs.
2 · Storing it all
Per month, for the whole fleet.
3 · Serving the traffic
4 · Changing a site after it is live
What each kind of edit costs us. The free rows are free because the model is choosing from a fixed vocabulary rather than rewriting markup.
5 · Monthly running cost
6 · Unit economics
What could break this model
- Image retries re-bill. A failed generation that retries charges again. We have already been burned by exactly this, and at fleet scale a retry storm is real money that nobody is watching.
- Citation tracking has no ceiling. It is the largest recurring line and the one I am least confident about. Some engines have no API, so it becomes SERP scraping priced per call. Halving the question count or moving to fortnightly nearly halves your biggest variable cost.
- Demand data is unmetered today. Demand-checking a large page-set matrix on onboarding can spike hard, and nothing currently caps it.
- My token estimates could be 2× out. They are reasoned, not measured. Log real input and output tokens per pass on the first five builds.