How much does an AI chatbot cost? Four pricing models, worked through

The same month of questions can cost $12 or $227 depending on which AI chatbot answers them. Most of that gap comes from what each vendor counts as one unit, and from whether that unit grows with your traffic.

Vendors charge by the seat, by the reply or credit, by the conversation the AI settles, or by the month. The prices are for monthly billing, from each vendor’s own site on 18 September 2026.

Deacon, our own product, bills by the month and answers visitors from your help pages. It also has a free plan of 50 answers a month with no card.

The site we priced

The site has one founder and takes a thousand questions a month from visitors. The bot writes one reply to each question.

Questions arrive in conversations, and nobody’s bill can be worked out without knowing how many. Dante AI tells buyers to size a conversation at about four replies, and Gleap puts its own at about 4.3 messages. Our own pricing page reckons on fewer replies each, which is how it puts Starter’s 1,000 answers at about 420 conversations. We use the vendors’ four, which makes 250 conversations.

Not every conversation ends with the problem solved. Intercom says Fin’s average resolution rate is 76%, and Help Scout, HubSpot and Tidio publish averages between 67% and 73%. We use 76% for every per-resolution sum, which is 190 resolved conversations. A lower rate would make those bills smaller and leave more questions for you.

So the month is 1,000 questions and 1,000 replies, in 250 conversations, 190 of them resolved by the AI, with one person on one seat. Where a vendor lets you pick the model, every sum uses its lightest one.

Per seat

“You pay per user on your team, and each plan comes with a set number of AI Agent resolutions per month,” says ChatBot.com’s pricing page.

The resolutions decide the bill. A seat pays for a person in the inbox, and the AI inside it has an allowance of its own. ChatBot.com counts one resolution for a chat the AI solved, however many questions were in it, so our site needs 190.

Essential is $25 a user with 10 resolutions included, and extra resolutions cost $0.99 each or $49.50 for a pack of 50 that refills automatically. Covering the month on Essential takes 180 more at $0.99, which makes $25 plus $178.20, or $203.20. Growth is $99 a user and includes 200 resolutions, enough on its own, so the per-seat month is $99.

Intercom, Help Scout, Front and Zendesk start from a seat as well, and all four bill the AI separately.

Per message or credit

A SiteGPT message is “one user question plus the AI reply, counted together”, so our month is 1,000 of them. Dante AI’s pricing page says “a credit is one answer from your AI agent”, and its lightest models spend one credit an answer. On Chatling a reply costs “a certain number of AI credits based on the model used”, half a credit on the lightest.

Each Starter plan covers the month with room to spare, as long as a small model writes the replies. A bigger model spends the allowance faster, and on Dante AI an answer can cost anywhere from 1 to 25 credits. At five credits a reply our month is 5,000 credits. Starter holds 2,000, and the other 3,000 are three top-ups at $20 each, so the month becomes $100.

Chatbase prices its credits the same way, at one to six a message depending on the model. Crisp’s credits are a dollar amount, and it says only that some models cost more than others.

Per resolution or conversation

Fin charges $0.99 an outcome, and at most one outcome a conversation. Intercom charges nothing when Fin cannot answer or the visitor asks for a person, so a quiet month bills fewer outcomes. Intercom’s help centre says a customer who leaves after Fin’s answer without asking for more help “is counted as an Assumed Resolution and will be charged.”

Our 190 resolutions come to $188.10. On Intercom’s own helpdesk Fin needs at least one seat, and Essential is $39 a seat on monthly billing, so the month is $227.10. Fin can also run on a helpdesk you already pay for, with no seat and a minimum of 50 outcomes a month.

Tidio’s Lyro charges for the conversation instead, resolved or not. In Tidio’s words, a Lyro conversation is one “that has at least one reply from our AI Agent”, so all 250 of ours count. Lyro is sold in fixed packages, and the smallest that holds 250 is 300 conversations at $210 a month, busy or quiet. Tidio’s help centre adds that a thread closes after fifteen idle minutes, and a visitor who writes again after that starts a new one.

Flat

“Flat pricing, AI included,” says Chattering’s pricing page. A flat plan charges one price for the month, counts every message or reply the same whatever model writes it, and sells nothing past the allowance except the next plan up. When the allowance is spent, the AI stops until the month turns over.

Chattering needs its Pro plan here, because Starter holds 500 replies, half our month.

Deacon’s Starter holds exactly 1,000 answers, and our site uses all of it. Follow-ups count, and so do the replies where Deacon says it does not know. A slightly busier month means either Growth at $150 or the rest of the month spent taking visitors’ email addresses instead of answering them.

Elfsight says its chatbot pauses at the limit. Chattering and Deacon both stop answering and ask the visitor for an email address instead.

Crisp calls its price “a flat pricing per workspace”, and for seats it is. The AI runs on a pool of credits inside the plan, $5 on Mini and $25 on Essentials, which Crisp reckons at around 90 and 450 conversations. Our 250 need Essentials at $95. When the pool runs dry, Hugo hands new conversations to your team unless you turn on pay-as-you-go, at a rate Crisp does not publish.

The same month, every way

At 2,000 questions there are 500 conversations and 380 resolutions. ChatBot.com’s Growth covers 200 of them, and the other 180 at $0.99 each add $178.20. Fin is 380 outcomes at $0.99 plus the seat. Tidio moves to its 500-conversation package, and Dante AI lands exactly on its Starter allowance. Crisp reckons Essentials at around 450 conversations, so 500 means Plus, or pay-as-you-go at the rate it does not publish.

Doubling the traffic adds $200 to Crisp, $188.10 to Fin, $178.20 to ChatBot.com, $140 to Lyro, $100 to Deacon and $12 to Elfsight. It adds nothing to Chattering, or to the credit and message plans on their lightest models.

None of this says whether the answers are any good, and we have written about how to test that separately.

Intercom, Tidio, Crisp and Chattering all come with an inbox for your team to work in. If two or three people will reply from one every day, part of those bills buys something you would pay for anyway.

What changes the numbers for your site

For the tools that bill by the resolution or the conversation, questions per conversation matter more than anything else. At two a conversation instead of four, the same thousand questions become 500 conversations and 380 resolutions, and those bills rise exactly as they do when traffic doubles. The message and credit plans stay put, since the number of replies has not changed.

Each vendor’s small print moves the count as well. Tidio’s fifteen-minute rule, Intercom’s assumed resolutions and SiteGPT’s question-and-reply pairs all change what the same traffic costs.

At the limit, ChatBot.com’s packs refill on their own, and Botpress adds packs of conversations automatically, which its pricing FAQ says you cannot turn off.

You can price your own site from last month’s inbox or chat history. Count the questions people asked, which is also your count of replies. Divide that by the number of conversations they arrived in, and you have your own questions per conversation in place of our four. The conversations a person did not have to finish are roughly what a per-resolution vendor would bill.