How Much Does an AI Customer Support Chatbot Cost in 2026?

"How much does an AI chatbot cost?" has no single number, because vendors do not sell the same unit. One charges per seat, one per resolved conversation, one per message credit, one a flat monthly fee. Two products that look identical in a demo can differ by a factor of fifteen on the invoice at the same volume, and the cheap one at 300 conversations a month is often the expensive one at 3,000.

This guide is for the person who has to sign off on the bill: a founder, a head of support, or an operations lead at a small or mid-sized business comparing quotes. It explains the four pricing models, puts verified vendor prices side by side, works three volume scenarios so you can find your own row, and lists the costs that never appear on a pricing page. Every price below was checked against the vendor's public pricing page in September 2026, and prices change, so treat the numbers as a snapshot and re-check before you buy.

The four ways AI chatbots are priced

Every AI support product on the market bills through one of four models, or a combination of two. Knowing which one you are looking at tells you how the bill behaves when your volume changes, which matters more than the headline price.

Per-seat pricing charges a monthly fee for each person on your team who logs in to answer customers. It is the traditional helpdesk model. The AI is either included with the seats or added on top, and your bill scales with your team size, not your ticket volume. A seat is a paid login for a human agent; visitors chatting with the bot are never seats.

Per-resolution pricing charges each time the AI closes a conversation without a human. Intercom calls the unit an outcome and Zendesk calls it an automated resolution. The bill scales with how well the bot performs: a better knowledge base means more resolutions and a higher invoice. An automated resolution is a conversation the vendor counts as finished by the AI, usually because the customer confirmed the answer or simply stopped replying.

Message-credit pricing sells a monthly bundle of credits, and each AI reply consumes one or more depending on the model behind it. Chatbase is the clearest example. The bill scales with message count, and the credit cost per message varies with which AI model you pick, so the same conversation can cost one credit or many.

Flat-plan pricing charges a fixed monthly fee for a stated allowance of AI replies, with a published overage rate past it. PepoChat and Crisp's workspace plans work this way. The bill is predictable within the allowance and grows in a known step past it.

The rest of the post puts numbers on each model.

What AI chatbots actually cost in 2026: vendor prices compared

The table below lists public prices as of September 2026. Where a vendor offers annual and monthly billing, the annual price is shown because that is what the pricing page displays by default. All prices are in US dollars.

Sources for the table: the Intercom pricing page, the Zendesk pricing page, the Chatbase pricing page and the Tidio pricing page, all read in September 2026. The Freshdesk, HubSpot and Crisp figures are from the same month and are discussed in more detail in our small business chatbot comparison and the Tidio vs Crisp vs PepoChat write-up.

Two things stand out. First, the helpdesk suites all hide the AI price behind a per-seat price. The seat number is small and prominent; the AI number is either per unit or on request. Second, the standalone tools publish everything, and the difference between them is what the unit is: a credit, a conversation, or a reply.

How much does an AI chatbot cost at 300, 2,000 and 10,000 conversations a month?

Headline prices are useless without volume. The scenarios below apply the same assumptions to one representative vendor per model so you can see how each bill behaves as volume grows.

Assumptions, stated so you can adjust them:

• Your team has two human agents who need a login.
• The AI resolves 50% of conversations without a human. That is the middle of the 30% to 70% range most vendors quote; our ticket deflection benchmarks explain where the range comes from.
• Each conversation uses two AI replies on average, one answer and one follow-up. Chatbase credits are counted at one credit per message, which is the rate for its cheapest models; a premium model consumes more.
• Intercom is priced on the Essential plan at $29 per seat. Zendesk is priced on Suite Team at $55 per agent, because Support Team does not include chat. Chatbase is priced on the smallest plan whose credits cover the volume. PepoChat is priced on the cheapest plan plus overage.

Read across the rows and three patterns appear.

At 300 conversations, everything is cheap and the model barely matters. Intercom is the most expensive at about $207 because the seats and the 50-outcome minimum both apply, but the difference between $40 and $207 a month is not what decides a purchase at that size. Feature fit and setup time matter more.

At 2,000 conversations, per-resolution pricing separates from the pack. Intercom's Fin line alone is $990, nearly 17 times its seat line. The credit plan and the flat plan are within $81 of each other. If your resolution rate rises from 50% to 70%, the Fin line goes to $1,386 and the other two do not move; the Intercom Fin pricing breakdown walks through that table.

At 10,000 conversations, the unit price is the whole story. Chatbase's $40 per extra 1,000 credits is four times PepoChat's $10 per extra 1,000 replies, and that multiplier is what separates $700 from $249. Intercom's $5,008 is what "pay for value" looks like at scale: a good bot on a per-outcome plan is a large fixed cost that you cannot reduce without making the bot worse.

Cost per conversation, which is the number to put in a spreadsheet, comes out of the middle row as about $0.52 for Intercom, $0.075 for Chatbase and $0.035 for PepoChat. Zendesk's per-resolution rate is not published, which is why its column reads "plus AI"; ask sales for the rate and the overage policy before you sign, because the AI can be set to pause mid-month when the allowance runs out.

The scenarios deliberately exclude Tidio and Crisp. Tidio's Lyro price above the 50-conversation tier is set in a calculator rather than a list, and Crisp's AI credits are described as an approximate number of conversations, so a clean per-volume row would be a guess. Both are in the vendor table with their entry prices.

What does a free AI chatbot plan actually include?

Free plans are where most small businesses start, and they differ far more than the paid tiers do. The question is not whether a free plan exists but what it withholds.

Chatbase Free gives 50 message credits a month and 1 MB of training data, with no AI actions, and its pricing page states that agents are deleted after 14 days of inactivity. Fifty credits is roughly 25 short conversations, enough to test the product and not enough to run on. The deletion rule means a side project that goes quiet for two weeks has to be rebuilt.

Tidio Free includes 50 Lyro AI conversations as a one-off lifetime allowance, not a monthly one. Once they are used, the AI stops until you buy the add-on. The live chat itself stays free at 50 conversations a month.

Crisp Free includes two seats and no AI credits at all. The free plan is a live-chat inbox for two people; the AI agent starts at Mini.

Intercom, Zendesk and Freshdesk have no free plan, only a 14-day trial. HubSpot's free tier includes live chat with HubSpot branding and limited rule-based bots, but the Breeze customer agent needs Professional or above.

PepoChat Free includes every feature: the AI agent, the team inbox, all thirteen integrations and custom actions, appointment booking, voice, meeting memory and analytics. The limits are 200 AI replies a month, 5 knowledge sources, one team member and 20 MB of knowledge-base text. When the 200 replies run out, new conversations are handed to the team inbox with a friendly message instead of failing, and the agent resumes on the first of the next month. A whole-site crawl counts as one source, so five sources covers a website, a PDF, a Notion connection and two more. The pricing page lists the full comparison.

The honest reading of the free tiers is that they are sized to prove the product works, not to run your support on. The exception is a business with fewer than about 100 conversations a month, which fits inside PepoChat's free allowance with room to spare, and inside Chatbase's only if each conversation stays very short.

The hidden costs of an AI support chatbot

The pricing page shows the subscription. The total cost of running a bot includes at least six lines that never appear there.

Content preparation time

An AI agent answers from your content, and most businesses discover that their content is thinner than they thought. A support agent's first week is spent finding that the returns policy lives only in a PDF, the pricing page is out of date and the shipping FAQ contradicts the checkout page. Budget a day or two of someone's time to import the site, upload the documents and fix the wrong answers with Q&A pairs, and an hour a week after that to review conversations. The training guide shows what that work looks like in practice. At any pricing model, this is the largest cost in the first month.

Overage on your busiest month

Every usage-based plan is priced on your average and billed on your peak. A retailer at 800 conversations a month in June and 3,000 in December pays the December bill in December. On a per-outcome plan that is nearly four times the June invoice. On a flat plan with a published overage rate the step is smaller and known in advance: on PepoChat's Starter, 3,000 conversations is 6,000 replies, which is $49 plus $40 of overage. Ask every vendor for the overage rate in writing, and whether the AI keeps answering or stops when the allowance runs out.

Seats for people who only look

Per-seat products charge for everyone who needs to see conversations, not only those who reply. The founder who wants to read transcripts, the marketing lead who checks what customers ask about, and the developer who debugs an integration all need logins. Intercom offers free lite seats for lookers; Zendesk and Freshdesk charge full price. Count the people who will ever open the inbox before you compare seat prices.

Annual commitment

Most of the per-seat prices in the vendor table are the annual rate. Monthly billing is typically 20% higher, and annual billing means paying for a year of a product you have used for two weeks. A month-to-month plan costs more per month and less if the product is wrong for you. PepoChat's yearly billing is ten months' price for twelve, with no annual requirement for the monthly plans.

Removing the vendor's branding

"Powered by" badges on the widget are removed on higher tiers almost everywhere. On PepoChat the badge comes off on Growth. If a white-label widget matters to you, price the tier that removes it rather than the entry tier.

The cost of a wrong answer

This is the line nobody budgets and everybody pays. A bot that answers confidently and wrongly generates a second conversation, an email or a refund, and on a per-resolution plan the wrong answer was billed as a success because the customer stopped replying. A bot that says "I don't know" and hands off to a person costs nothing under per-outcome pricing and one reply under a flat plan. When you compare vendors, ask how each one behaves when the answer is not in your content; the guide to stopping hallucinations lists the questions to ask.

Which chatbot pricing model is cheapest for your volume?

The scenarios point to a rule of thumb that holds across vendors.

Below about 200 conversations a month, per-resolution pricing can be the cheapest paid option, because you pay only for the handful of conversations the bot closes and nothing for the ones it hands off. Intercom's 50-outcome minimum, about $49, sets the floor; below that volume a free plan is cheaper still.

Between about 300 and 5,000 conversations a month, a flat plan or a credit plan wins, and the gap widens as volume rises. This is the range most small and mid-sized businesses live in, and it is where per-resolution pricing is at its most expensive relative to the alternatives.

Above 5,000 conversations a month, the overage rate decides everything. Compare the published per-1,000 price rather than the plan price, and ask whether volume discounts exist.

There is one case where the price is not the point. If each support conversation is attached to a five-figure account, the difference between $0.99 and $0.035 per conversation is noise, and the only question is which product answers best. Per-outcome pricing is painful when conversations are numerous and low-value, which describes most ecommerce and consumer software support and very little enterprise sales support.

How to calculate your own AI chatbot cost

Work your own row of the scenario table in five steps.

1. Count last quarter's monthly conversations, and note the busiest month separately. If you have no chat yet, count support emails and tickets; a widget typically adds volume rather than replacing it, because questions that were never emailed get asked.
2. Estimate a resolution rate between 30% and 70%. Use 50% if you have no data. The bot's share rises as your content improves, so run the calculation at both ends.
3. Multiply for the AI line. Per resolution: conversations times rate times the per-unit price. Credits: conversations times messages per conversation, matched to the plan that covers it. Flat: conversations times replies per conversation, matched to the allowance plus overage.
4. Add seats for everyone who will open the inbox, at the monthly rate if you are not ready to commit for a year.
5. Add the hidden lines: content preparation in month one, the branding tier if you need it, and a margin for your peak month.

Put the results side by side with the free plan you would otherwise use. For many businesses the honest answer is that the first three months should run on a free plan while the knowledge base is built, and the paid decision should be made with real conversation counts rather than estimates.

What to do next

If your volume sits in the few-hundred-to-few-thousand range where flat plans win, start on PepoChat's free plan, import your website as one knowledge source, and watch the reply meter for a month before choosing between Starter and Growth on the pricing page. Every feature is on the free plan, including the integrations listed on the integrations page, so the test is a real one. For a fuller comparison of the products in the table, read the small business chatbot guide, and if Intercom is on your shortlist, the Fin pricing breakdown works the per-outcome maths in detail. You can create a free workspace without a credit card.