Live chat vendors bill their AI chatbots in at least six different units, and the unit decides what you pay for a chat that goes nowhere. Some charge only when the AI resolves a question, but count a customer going quiet as a resolution. Others charge for every session whether it helped or not. This guide sets out each vendor's written definition and the catches in it.
The billable units at a glance
| Unit | What is charged | Products | Published price |
|---|---|---|---|
| Resolution | A conversation the AI is judged to have resolved | HubSpot customer agent; Zendesk AI agents; LiveChat AI Agent | US$0.50; US$1.50 committed or US$2 pay-as-you-go; US$0.99 |
| Outcome | A resolution, a configured handoff, or a sales result | Intercom Fin AI Agent | US$0.99; sales qualification US$9.99 |
| Session | All AI or bot activity with one visitor in a fixed window | Freshchat Freddy AI Agent | US$49 per 100 sessions |
| Conversation | A chat the AI replied in, resolved or not | Tidio Lyro AI Agent; Smartsupp Mira AI Agent; Lime Connect AI Agent | US$39 for 50; €20 per 100; €99 incl. 100 |
| Message credit | Each AI message or action | Chatwoot Captain AI; tawk.to AI Assist | US$20 per 1,000; US$30 per 1,000 |
| Token credit | A dollar value consumed by model tokens | Crisp Hugo | No pay-as-you-go rate published |
Resolutions and outcomes: silence can count
Intercom bills Fin AI Agent per outcome, at most one per conversation. Intercom's help centre counts a resolution when the customer confirms the answer, or when the customer "exits the conversation without requesting further assistance", which Intercom calls an assumed resolution. A Procedure you configured to end in a handoff to a human is also a billable outcome. A customer asking for a human is not charged, and Intercom's pricing page says an outcome is refunded if the conversation is reopened.
HubSpot bills its customer agent 50 HubSpot Credits per resolution, at US$0.01 a credit. A conversation counts as resolved when the agent shared a content source or performed an action and there was no qualifying handoff within 72 hours of the visitor's last message. Only a visitor-initiated handoff qualifies: HubSpot's knowledge base says manual assignment by your team or a workflow does not. A customer who simply goes quiet is therefore billed as resolved. The agent needs Service Hub Professional, which includes 3,000 credits a month, enough for 60 resolutions.
Zendesk bills automated resolutions: conversations the AI agent handled without escalation that end by inactivity and then pass a check by a large language model. For messaging the inactivity window is 2 hours by default, adjustable up to 72. The customer never has to confirm. Suite Team includes 5 automated resolutions per agent a month, capped at 10,000 a year, and the allowance expires annually.
LiveChat's AI Agent is sold in Text and ChatBot.com plans, not classic LiveChat plans. Text's pricing documentation counts a resolution when the AI answer directly solves at least one question, once per conversation; a handoff without an answer is not counted. Over the allowance, a pack of 50 resolutions at US$49.50 is added automatically.
Sessions: billed whether or not anything was solved
Freshchat defines a session as all interactions between one visitor and the Freddy AI Agent or a rule-based chatbot within a 24-hour window. Freshworks' FAQ is explicit that a session is billed whether or not the issue is resolved. Previewing a bot while building it also consumes sessions. Growth and higher plans get 500 complimentary sessions once per account, and they are removed as soon as you buy a paid pack. Auto-recharge adds a pack when the balance falls to 50 sessions, and unused sessions do not carry over.
Conversations: one unit per chat, with edge cases
Tidio counts a Lyro conversation when Lyro replies at least once, resolved or not; an immediate handoff is not deducted. Tidio's help centre says a thread closes after 15 minutes of inactivity, so a visitor who comes back is counted again, while Tidio's pricing FAQ says a returning customer "still counts as a single conversation". The free plan's 50 Lyro conversations are a lifetime allowance that never resets.
Smartsupp counts each unique chat session in which Mira answers as one conversation, whatever the message count, sold in blocks of 100 a month. Whether unused conversations roll over is not documented. Lime Connect bills per "complete chatbot interaction" and says unused conversations carry over within the contract term, but prices for packages above 100 conversations and any overage rate are not published.
Credits: per message or per token
Chatwoot charges 1 Captain credit per AI message or action, so a five-reply conversation costs at least five credits. Startups includes 300 credits a month; purchased credits expire after 6 months. At zero, Chatwoot hands the conversation to a human.
tawk.to also charges one credit per AI message, including Smart Reply and AI Commands used by your own agents. Its help centre says that when credits run out, AI Assist keeps answering and the account "may result in a negative balance, which you can settle later". Purchased credits do not expire.
Crisp bills Hugo in dollar-denominated credits consumed per token: US$5 a month on Mini, US$25 on Essentials and US$75 on Plus. Crisp estimates a conversation at US$0.05–US$0.10 and calls that "only an estimate". Tests and Playground use burn credits too. When credits run out, Hugo stops and conversations go to humans unless pay-as-you-go is switched on.
The catches to look for
| Catch | Where it is documented |
|---|---|
| Customer silence counted as a resolution | Intercom (assumed resolution), HubSpot (72-hour rule), Zendesk (inactivity plus LLM check) |
| Billed regardless of outcome | Freshchat (24-hour session), Tidio (any Lyro reply) |
| One-off allowance presented as free AI | Tidio (lifetime Lyro quota), Freshchat (complimentary sessions removed on first purchase), Smartsupp (test conversations) |
| Automatic top-ups | Freshchat auto-recharge, LiveChat overage packs, Zendesk optional automatic pay-as-you-go |
| Negative balance | tawk.to AI Assist |
| Tests and previews consume units | Freshchat, Crisp |
| No overage price published | Lime Connect, Crisp |
To see how these products rank on AI pricing transparency, read the best live chat with an AI chatbot. For the difference between an AI agent and a rule-based bot, see live chat vs chatbot vs AI agent.
Frequently asked questions
Which is cheaper: per-resolution or per-conversation AI billing?
It depends on how often your AI fails. Per-conversation and per-session billing charges for chats that go nowhere; per-resolution billing does not, but can count a silent customer as resolved. Estimate your volume, then apply each vendor's written definition.
Does a handoff to a human cost anything?
It depends on the vendor. Intercom does not charge when the customer asks for a human, but does charge for a Procedure you configured to end in a handoff. HubSpot only waives the charge for a visitor-initiated handoff, not one made by your team. Tidio does not deduct an immediate handoff.
What happens when AI credits run out?
Crisp and Chatwoot hand conversations to humans. Freshchat's agent stops and hands over. tawk.to keeps answering and lets the balance go negative. Zendesk lets the admin choose between pausing and automatic pay-as-you-go at the higher rate.
Are free AI allowances monthly?
Not always. Tidio's 50 free Lyro conversations are lifetime, and Freshchat's 500 complimentary sessions are one-off. tawk.to's 100 credits and Chatwoot's plan credits reset monthly.
Did you test these AI chatbots?
No. This guide reads each vendor's billing documentation. We have not run conversations through these products. See How we review.
Keep reading
- Is a free live chat plan enough for your team?guide
- Per-seat vs per-workspace pricing: what your team size does to the billguide
- Live chat vs chatbot vs AI agentguide
- Does a chat widget slow your site?guide
- Best live chat software in 2026best list
- Free-plan audit: does your team fit a free live chat plan?
This page was researched and drafted with AI assistance from the sources listed on it. We have not installed or tested these products ourselves. Method: How we review