A chatbot is not a product, it is a scope
Ask ten sellers what a chatbot costs and you will hear every number from $200 to $20,000. That is not useful, because nobody is hiding the real variable: a chatbot is not a thing you buy, it is a scope you agree. The price is a consequence of five decisions, and you can see every one of them before you pay.
The five decisions that set the price
- What can it read? Only your own documents (product pages, FAQs, policies, price lists), or the open web? Reading only your content is the whole point of a useful assistant, and it is also the harder build, because it needs retrieval and citing. This alone is often a $300 difference.
- What happens when it does not know?The correct answer is “I do not know, here is the link to a human”. Any vendor whose demo never says “I do not know” is showing you a model that invents, and you will pay for that in support tickets.
- Where does it live? A site widget, WhatsApp, Discord, or an internal tool. Each channel has its own API, its own failure modes and its own message limits.
- Does a human approve the answers? Needed for anything touching billing, medical, legal or refunds. It is a feature, not a limitation, and it should be priced as one.
- Who maintains it? Your prices, policies and FAQs change. Somebody has to update the source documents, or the bot starts answering last quarter’s question correctly and confidently.
What that costs, honestly
- $250–$600 — lead-response and customer-support bot. One channel, trained on up to ~10 documents, writes leads into a Sheet or CSV, human-approval mode on. This is the version most businesses actually need.
- $400–$900 — custom GPT/AI assistant trained on ~30 documents with a live demo link and a handover call. The flagship version.
- $650 — retrieval knowledge base (RAG). Answers only from your documents, cites the source, says “I do not know” when it should. The hardest and most useful build.
Anything quoted far below those numbers is usually a bot wired straight to a model with no retrieval, no citations and no testing against real questions. Sometimes that is exactly what you want. It is fine to buy that, as long as you know it is what you bought.
Why we price by scope, not by seat
A per-seat or per-message pricing model punishes you for the chatbot working. We price the build and the maintenance separately, so the running cost is visible: the assistant is built once against a written scope, and monthly care covers document updates and checking real conversations. Our reviewed project history shows what delivery on a written scope looks like in practice — 153 client reviews, most of them small, mid-range web work.
What you get before anything is built
One hard scope line, one price, one delivery date. Anything outside the scope becomes a change-order quoted before we start, never a surprise on the invoice. Our rate is $40/hour on Upwork and $20/hour on PeoplePerHour, and both profiles are public if you want to check the track record first.
Frequently asked
Is a chatbot cheaper than a human?
Cheaper per answered question, yes. But it removes a person from the loop, and for anything consequential you probably want the person kept. The honest split is: let the bot handle the repetitive 60%, keep a human for the rest.
Do you train it on our website?
Only on documents you give us. We do not train on the open web and we do not sell what your customers ask it. If you need it to know something you have not written down, the first task is writing it down.
What if we already have a chatbot that underperforms?
Then the problem is usually the source documents or the fallback, not the model. That is a scoping conversation before it is a build, and it is cheaper to have it first.
Next step
Tell us what the assistant should know and which channel it lives in. You will have a fixed price and a delivery date within 48 hours: What is the one question you would want your customers to stop asking you twice a day?