What Counts as a Chatbot Resolution? 8 Vendors' Definitions Compared

Per-resolution pricing sounds simple: you pay when the AI solves a problem. But each vendor's help center defines "solved" its own way, with its own silence window and its own rule for human replies, so two similar quotes can count the same month of chats very differently.

This comparison is for founders and support leads weighing per-resolution quotes or resolution-rate claims. It quotes eight vendors' definitions, sets out their windows and exceptions, and runs one hypothetical month through every rule. For the Fin bill worked at three rates, see Intercom Fin pricing explained; for metric formulas, see AI chatbot metrics that matter.

Every definition was read on the vendor's own pages on 22 October 2026. PepoChat is our product; its billing and inbox behaviour was checked in its code and documentation the same day.

What is a chatbot resolution?

A chatbot resolution is a conversation that a vendor counts as solved by its AI without a person, under a rule the vendor writes itself, and in 2026 it is also the unit that Intercom, Zendesk, HubSpot, Gorgias, Help Scout and Ada bill. No standards body defines the term; even Ada's own blog says "The term 'resolution' isn't standardized."

The rules fall into three families:

• A confirmed resolution is a chat in which the customer says the answer helped, for example by typing "that worked".
• An assumed resolution is a chat in which the customer stops replying after an answer, and the vendor treats the silence as success once a window has passed.
• A verified resolution is Zendesk's term for a chat that a language model, reading the finished transcript, judges to be satisfactorily resolved.

The resolution rate is resolutions divided by the conversations the AI joined, so it inherits whichever rule produced the count.

How do eight vendors define a chatbot resolution?

Eight vendors publish eight rules: Intercom, HubSpot, Gorgias, Help Scout and Ada count an AI-answered chat that did not end in a handoff to a person, Zendesk bills only chats its language model verifies, and Tidio and Freshdesk bill every AI conversation or session. Each definition below is quoted as published.

Quoted from each vendor's own page, all opened on 22 October 2026; the dates are each page's own update date where it shows one.

Two details have changed recently. Help Scout's AI product is now Envoy, whose built-in agent Scout evolved from AI Answers, and HubSpot's customer agent switched on 14 April 2026 from $1.00 per conversation to $0.50 per resolved conversation.

When does each vendor count a resolution, and what cancels it?

The silence window runs from none stated at Help Scout to 24 hours at Intercom and 72 hours at HubSpot and Gorgias, and each vendor treats a human's reply differently, which decides whether a chat a teammate finished still costs money.

Rules and list prices in US dollars from each vendor's pricing page, help center or terms, checked 22 October 2026, before seats, plan fees and included allowances.

Three rows deserve a second look. Intercom bills a chat that a teammate joined uninvited: "If the customer has already received an answer from Fin and they haven't explicitly asked to get additional help or speak to a person, the conversation is considered 'resolved' and you would be charged for this." Gorgias bills that chat only as a helpdesk ticket, and Zendesk files it as an unbilled assisted escalation.

HubSpot's customer agent article fixes the status after 72 hours: "Subsequent actions, such as messages from a live agent, a transfer request, or negative feedback, will not change the resolution status."

Tidio's pages disagree: its pricing FAQ says a customer who "leaves the chat window and returns to it at a later time" still counts once, while its help article on the Lyro limit, updated 15 September 2026, says "a thread closes after 15 minutes of inactivity" and counts the next message again.

What is the difference between a confirmed and an assumed resolution?

A confirmed resolution needs the customer to say the answer worked; an assumed resolution needs only silence, and Intercom counts a customer who disengages for 24 hours after Fin's last answer as resolved and bills it at $0.99.

Intercom's Fin pricing: Outcomes article, dated 17 September 2026, is explicit: "If a customer disengages from the conversation for 24 hours after Fin's last answer, it is considered an assumed resolution." Its FAQ covers a frustrated customer who just leaves: "A customer leaving after receiving Fin's answer without requesting further help is counted as an Assumed Resolution and billed at $0.99."

Counting silence has a fair rationale, since many satisfied people close the window without typing thanks to a bot. But a wrong answer that makes someone give up looks exactly like a right one.

Customers have argued the point in public. In a May 2025 thread on Intercom's community forum titled "Fin's flawed resolution assumption", one customer found that of 16 tickets marked resolved by Fin in four weeks, "9 out of 16 are not solved by Fin at all". An Intercom employee replied on 4 June 2025: "A conversation is billed as a resolution when the end user gets an answer and doesn't ask for further help."

Verified resolutions are Zendesk's answer to that problem. Since 18 May 2026, its automated resolution tiers run a check when a conversation ends: "a verification process is performed by an LLM that evaluates the text of the conversation to confirm that the customer's request was satisfactorily resolved." A chat that fails becomes a contained resolution, a tier that "does not count against your resolution allowance." The judge is a model rather than the customer, and Zendesk publishes no pass rate.

Which chatbot billing units are not resolutions?

Three common units are not resolutions at all: Tidio bills Lyro conversations, Freshdesk bills 24-hour sessions and PepoChat bills AI replies, so each charges for a chat whether it ended well, badly or in a handoff.

A Lyro conversation is Tidio's unit: any customer interaction "that has at least one reply from our AI Agent", counted once however often Lyro answers, at $39 for 50 up to $700 for 1,000 a month. A chat Lyro answered badly costs the same as one it solved. Tidio's quote-only Premium plan adds pay-per-resolution billing and a "Guaranteed 50% Lyro AI resolution rate".

A session is Freshdesk's unit for its Freddy AI Agent: every interaction with the AI in the 24 hours from the start of a chat, or the 72 hours from a first email. Each account gets 500 sessions once on Growth, Pro or Enterprise, then pays $49 per 100, and the chat agent runs only in Freshdesk Omni.

Credits are a currency rather than a unit. HubSpot charges "$0.010 per credit" and 50 credits per resolution, so the 72-hour rule still decides the count. Gorgias runs two meters on one chat: its pricing FAQ says "Each AI Agent interaction also counts as a helpdesk ticket and is billed accordingly."

An AI reply is PepoChat's unit: one answer the agent generates in a visitor conversation, whether or not the conversation resolves. Replies your team types in the inbox, the greeting and system messages do not count. Our comparison of Tidio Lyro, Intercom Fin and Zendesk AI agents sets three of these rules against PepoChat's.

What does the same month cost under eight different rules?

One hypothetical month of 1,000 AI chats, run through all eight rules, produces between 400 and 1,000 billable units and AI bills from $325 at HubSpot to $800 at Zendesk, before any seats, plan fees or included allowances.

The month, in five groups:

1. 200 confirmed. The AI answered and the visitor said it helped.
2. 400 silent. The AI answered and the visitor left without replying and never came back.
3. 250 asked for a person after at least one AI answer.
4. 50 joined by a teammate after the AI's answer, before the visitor asked for anyone.
5. 100 with no real answer. The visitor only said hello, or left when the AI asked a clarifying question.

The other assumptions: list prices on monthly billing; no seats, plan fees, included allowances or minimum commitments; no chat reopened; every AI answer cites a help article, which HubSpot's rule requires; Zendesk's check passes all 200 confirmed chats and half of the 400 silent ones, a guess because Zendesk publishes no pass rate; and two AI replies per conversation for PepoChat.

Unit prices from each vendor's pricing page or help center, checked 22 October 2026. Hypothetical month; swap in your own counts.

Two cells are judgement calls. Ada's service terms count any conversation not escalated, so the 100 unanswered chats count unless they fail the "Relevant, Accurate and Safe" standard, and the 50 teammate chats count unless that was a handoff "through the Services". Help Scout requires a resolution "without help from a human", so we left out its 50 teammate chats; with them, the bill is $487.50.

Read the table two ways. The identical month scores 40% at Zendesk, 60% at Gorgias and Help Scout and 65% at Intercom and HubSpot, so a resolution rate describes the vendor's rule as much as its AI. And the bills do not follow the rates: HubSpot counts as many resolutions as Intercom at half the unit price, while Zendesk counts the fewest and bills the most.

PepoChat's row is not like for like: its $49 is the whole Starter plan, while the other rows leave out the seats or helpdesk plan each vendor requires, with email, social channels and ticketing that PepoChat lacks. Its own inbox would show 200 resolved, 300 escalated and 500 unresolved, a strict 20%, because nothing turns silence into a resolution. The AI customer support pricing index prices 30 tools, seats included, at three volumes.

What is a good resolution rate for an AI chatbot?

There is no neutral benchmark for a good AI resolution rate, because every published average is measured under its own vendor's rule; on 22 October 2026, four vendors' own averages ranged from 65% to 76%.

Intercom's Fin homepage claims an average of "76% across 12,000+ customers". Tidio's pricing page says "Our customers average a 75% resolution rate with Lyro". Help Scout's pricing page says "The average resolution rate is 73%." HubSpot's April 2026 announcement says its customer agent "already resolves 65% of conversations". None of the four pages explains how its average was measured.

In the hypothetical month, the same chats scored 20% under a strict status count and 65% under Intercom's rule. A strict rate far below the vendor averages may simply mean you count strictly. Judge your rate against your own baseline under a rule you fixed before launch; the ticket deflection benchmarks post explains why published figures drift so far apart.

How do you measure your own chatbot resolution rate honestly?

Measure your own resolution rate by counting every conversation the AI joined, keeping silent chats in a separate "no reply" column, reporting a strict and a generous rate side by side, and checking a two-week sample by hand before you trust either.

1. Fix the denominator. Count every conversation the AI replied in, greetings and dead ends included, minus test chats.
2. Sort outcomes into four columns: confirmed, escalated, no reply after the AI's last answer, and no real answer.
3. Report two rates. The strict rate is confirmed chats divided by all chats; the generous rate adds the no-reply column. In the hypothetical month they are 20% and 60%.
4. Read a two-week sample of the no-reply column. Mark 20 silent chats answered correctly, answered wrongly or unclear. If 12 were right, credit 60% of the column: 200 plus 240 is 440 of 1,000, or 44%.
5. Apply the thumbs-down adjustment. If your tool records a vote per conversation, remove resolved chats with a negative vote from the numerator.

In PepoChat the counting comes from the team inbox, because Analytics has no resolution-rate chart: it shows only the Sentiments card, thumbs-up and thumbs-down votes over 90 days, and the Topics card, the most asked topics over six weeks. A conversation is resolved when the agent closes it after the visitor signals they are done, when the visitor picks End chat in the widget menu, or when your team resolves it. It is escalated at a handoff or when an operator replies. Nothing is resolved on silence, so a chat the visitor walked away from stays unresolved.

Export CSV gives each conversation's status, timestamps, message count and transcript, with lines marked user:, assistant: or operator:, as the team inbox docs describe. Resolved rows with no operator: line form your confirmed column, after a glance for visitors who ended the chat unhappy; unresolved rows ending on an assistant: line form your no-reply column. Votes are not shown per conversation, so treat the Sentiments card's negative share as a warning light, and do not bulk-resolve stale chats before you export.

What should you ask before signing a per-resolution contract?

Ask five questions before you sign a per-resolution contract: how long the silence window is, what a returning customer does to the count, whether a human reply cancels the charge, what happens at the cap, and how you audit each billed chat.

1. How long is the silence window? Intercom waits 24 hours, HubSpot and Gorgias 72, and Help Scout states none. Zendesk checks a messaging chat when it ends, 2 hours after the first message by default.
2. What happens when the customer comes back? Intercom deducts the resolution when the customer returns to the same conversation "(even across billing periods)". HubSpot reopens an email thread answered after more than 72 hours, and "If the customer agent delivers another resolution, additional credits are consumed."
3. Does a human reply cancel the charge? At Gorgias and Zendesk it does. At Intercom it does not unless the customer asked for help, and at HubSpot only a handoff the visitor starts counts. Help Scout and Ada do not say what happens when a teammate joins uninvited, so get it in writing.
4. What happens at the cap? Zendesk's default is "Allow overage"; its "Don't allow overage" setting pauses the AI. Intercom offers hard limits. At HubSpot, when credits run out "the agent will temporarily stop being assigned to new conversations", and Help Scout's spending limit means Scout "will be disabled in your Beacon until the next cycle or until you raise the limit."
5. Can you audit every billed chat? Zendesk's allowance dashboard "includes links to all the tickets that contributed to the resolution count", and Help Scout's billing article points to billable conversations in its Scout reports. Under Gorgias's billing docs one chat can use a ticket and an automated interaction, so ask for both counts. Ada calculates from a sampled rate; ask to see the sample.

Does flat or per-reply pricing change the incentive?

Flat and per-reply pricing take away the vendor's stake in calling a chat resolved, because the bill is the same whatever the outcome; they do not take away the cost of wrong answers, which you pay for under every model.

Per-resolution pricing charges nothing for a chat that ends in a handoff, a real advantage while your AI is new, but the vendor writes the rule that turns silence into revenue. Per-conversation, per-session and per-reply pricing charge for every chat, good or bad, so you carry the cost of failures while the vendor gains nothing from a generous count.

PepoChat, an AI support agent for websites, bills per AI reply, not per resolution: 200 replies a month on Free, 2,000 on Starter at $49 and 10,000 on Growth at $149, then $10 per 1,000 on paid plans, with every feature on the free plan. Nothing is counted on a visitor's silence. Analytics has no resolution-rate chart, only votes and topics, so a workspace computes its own rate from inbox statuses or the CSV export, as above. PepoChat publishes no resolution rate of its own.

What to do next

• Sort last month's chats into the five groups above and work out your strict and generous rates before you compare quotes.
• Send the five contract questions to each per-resolution vendor on your shortlist.
• Set the per-resolution bills from the table against the flat plans on the pricing page.
• To measure a strict rate on your own traffic, create a free workspace; the team inbox, its statuses and the CSV export are on every plan, including Free.
• Related guides: Gorgias pricing, HubSpot Breeze Customer Agent pricing, helpdesk AI add-on costs and Help Scout vs Zendesk.