"What does an AI chatbot cost?" is one of the most common questions — and one of the worst-answered. Most providers dodge it, because the honest answer is: it depends on what the bot does. This post makes the cost factors transparent so you don't walk into a pricing trap.
In short: A chatbot doesn't cost "per seat," it costs per use. Answer volume, answer length and model choice are the real drivers — think in usage rather than licenses and you only pay for actual value.
What costs really depend on
The biggest mistake is to tie chatbot costs to headcount. A chatbot doesn't cost "per seat" — it costs per use. The three real drivers are:
- Answer volume: How many questions does the bot answer per month?
- Answer length and complexity: A short FAQ reply is cheaper than detailed, multi-step advice.
- Model choice: A fast standard model costs a fraction per answer of a high-end one.
Understand these three and you can assess any offer realistically — instead of being dazzled by a low entry price.
Credits instead of guesswork
That's exactly why Zeptix bills in credits: a usage unit that reflects the actual answer volume. A short answer consumes few credits, a long, complex one more. So you pay for what the bot delivers — not for a flat "license."
The plans include a monthly credit allowance:
| Plan | Early-bird price | Credits / month |
|---|---|---|
| Starter | €29 | 5,000 |
| Pro | €69 | 15,000 |
| Business | €249 | 50,000 |
The current terms and included features are always on the pricing page — where you'll also see the regular prices after the early-bird phase.
The hidden charges to watch for
The advertised monthly price is rarely the whole truth. These items often surface only in the fine print:
| Hidden charge | What to check |
|---|---|
| Surcharge per extra bot | Are multiple bots included or does each cost extra? |
| Setup or onboarding fee | One-off costs not shown in the monthly price |
| Overage rates | What happens when the allowance runs out? |
| Data-location surcharge | Does EU hosting cost extra? |
| White-label / custom domain | Often only in higher plans |
A fair provider makes these visible before you buy. When in doubt, ask directly — the answer reveals a lot about the pricing culture.
Build it yourself vs. a platform — the math
"We'll build it ourselves, so it's free" is a common fallacy. The model API is only one part. On top come: hosting, a vector database for the knowledge base, embedding costs, a frontend widget, monitoring, maintenance — and the working hours to keep it all running. For most, that math quickly tips in favor of a platform that bundles these pieces.
A platform pays off especially when you want predictable costs instead of a fluctuating cloud bill at month's end.
How to keep your costs down
You have more influence than you think:
- Pick the right model. Not every FAQ needs the most expensive model. A fast, cheap one handles standard questions.
- Keep the knowledge base clean. A well-structured bot finds the answer directly instead of taking expensive detours.
- Constrain answer behavior. Clear boundaries prevent sprawling answers — and save credits.
These levers are no secret. How to set the bot up efficiently from the start is shown in the guide How to build an AI chatbot for your own website.
Conclusion
An AI chatbot doesn't cost "a flat X," it costs as much as it actually works. Think in usage rather than seats, check the fine print for hidden charges, and match the model to the task — and you'll pay exactly for the value, not for air. Transparent credits make that math clear from day one.
Want to see your own numbers? Build a chatbot for free and test how much volume you really need.
Build your own chatbot
Upload your knowledge, pick your model and go live in minutes — no code.
