Customer service chatbot cost: the whole bill

Most small businesses pay somewhere between nothing and about $80 a month for a customer service chatbot. That is the honest range, and if a number was all you came for, you can stop there. But the customer service chatbot cost that lands on your card in month three is rarely the one on the pricing page you read in week one. Three things move it: the unit the vendor counts in, a second subscription nobody mentions, and the hours you spend writing answers before the bot has anything to say. Here is the whole bill.

What a customer service chatbot costs in 2026

Published pricing across the category clusters into four brackets. These were checked against vendor pages and pricing guides in August 2026, and they move often, so treat them as a map rather than a quote.

• Free, $0. Usually 50 to 250 bot replies a month, one chatbot, one channel, and the vendor's badge sitting on your widget. Good for testing against real traffic. Not enough for a shop taking 30 questions a day.
• $19 to $89 a month. Where most small businesses land. Bot answers drawn from your own help content, a chat widget, one or two extra channels, and an allowance somewhere between 500 and 2,000 conversations.
• $100 to $500 a month. More volume, more integrations, more seats, sometimes a person from the vendor to talk to. Worth it above roughly 300 conversations a day. Rarely below.
• $9,000 and up, one time. A custom build. Almost never the right answer for a business under 20 people, and the year-one maintenance is usually another fifth of the build price.

So the software itself is cheap now. That is a real change. Five years ago the same feature list started in the hundreds and came with a sales call.

The catch is that "$29 a month" is a price for a quantity. The quantity is where small businesses get caught.

Chatbot pricing models: what you are actually buying

There are five common ways to count. Same bot, same month, and the invoices are nowhere near each other.

• Per seat. You pay per person on your team. The bill only moves when you hire. Easiest thing in the world to forecast.
• Per conversation. Every chat a customer starts counts against an allowance. Busy month, bigger bill. Read the definition carefully, because a customer who comes back the next day counts as one conversation at some vendors and two at others.
• Per active contact. You pay for every unique person who messages you in a month, whether they asked one question or fifteen. This one hits hardest on a spike. A mention in a newsletter or a sale weekend can double it.
• Per resolution. A fee, commonly $0.60 to $2.00, each time the bot closes a question without a person stepping in. Rates verified August 2026. We went through the fine print in per resolution pricing, including the part where every vendor writes its own definition of "resolved."
• Flat. One price, no counters, no allowance to watch. Uncommon in this category, and the reason we built our plans the way we did.

Four of those five are variable costs. Fine for a business with flat, predictable traffic. Rough on a business with a season.

One month, priced four ways

Picture a small online shop. 400 chats in a month, from about 260 different people. The bot handles 220 of them on its own and passes the rest to a human.

• Billed per conversation on a plan that includes 250, you are into the next tier up by the third week.
• Billed per active contact, you pay for all 260 people, including the 90 who asked one thing and never came back.
• Billed per resolution at $0.75, the bot having a good month costs you $165 on top of your seats.
• Billed flat, you pay exactly what you paid in your quiet month.

Look hard at that third line. The better the bot performs, the more you owe. Which is awkward, because your entire plan is to write better help articles and push that number up.

The costs that never make the pricing page

The subscription is the part everyone compares. These are the parts that decide whether the thing was worth buying.

Writing the answers

This is the big one, and almost nobody budgets for it. A support bot can only answer from what your business has already written down. If your policies live in your head and in three years of sent email, the bot has nothing to work with and will say so.

Plan 10 to 40 hours of somebody's time turning your top 20 or 30 questions into real help articles. Returns, shipping, hours, sizing, cancellations, what happens when something breaks. At a modest internal rate that is a few hundred to a couple of thousand dollars of work that never appears in the software comparison. Skipping it is the single most common reason a chatbot gets switched off in month two.

The second subscription

Plenty of small businesses buy a chatbot from one company and a help desk from another. Now you have two bills, two logins, and a gap where the handoff should be. The customer types a question, the bot cannot answer, and the conversation has to jump systems to reach a person who cannot see what was already said.

If the bot lives inside the same tool as your inbox and your tickets, that handoff is just a conversation continuing. If it does not, you are paying twice for a worse experience. Worth checking before you buy the cheap standalone widget.

The upgrades you did not plan on

• Removing the vendor's badge from your widget is often a paid tier on its own.
• Extra seats. Cheap plans are frequently one seat. Adding your part-time helper can cost more than the plan did.
• Integrations. The connector to your store or your email tool is regularly gated a tier above where the price you quoted sits.
• Overage rates. Ask what happens at 101% of your allowance. Some vendors bill per unit, some stop the bot, and one of those is much worse than the other.

Someone has to read what it said

For the first month or two, put half an hour a week aside to skim the transcripts. You are hunting for the questions where the bot said it did not know, because each one of those is a help article you have not written yet.

Teams that do this watch the customer service chatbot cost pay for itself faster every month, because the share of questions handled without a human keeps climbing. Teams that skip it usually decide after six weeks that the bot does not work, when what happened is the help center stopped growing.

Do the comparison at twice your current volume, not at today's. The renewal price is the price you actually pay. We listed more of these in our piece on the hidden costs of help desk software.

Working out whether it is worth the cost

Skip the vendor ROI calculators. Three questions get you close enough.

1. How many questions do you get a week? Under 20 and a good FAQ page will serve you better than any bot. Over 50 and you have a real case.
2. How repetitive are they? If more than half your inbox is the same six questions, that is exactly what a bot is for. If every message is a different account problem, it is not.
3. What does answering them cost you now? The comparison is an hour of a person's time. The Bureau of Labor Statistics publishes median pay for customer service representatives, and once you add payroll tax and tools, the loaded hourly number is meaningfully higher than the wage. If the person answering is you, the number is whatever your hour is worth doing something else.

Rough example. You field 60 questions a week and each one takes about four minutes to answer properly, including reading it and switching back to what you were doing. That is four hours a week, near enough 17 hours a month. A bot that handles half of them buys back eight or nine hours for the price of a couple of pizzas. You do not need a spreadsheet to settle that one.

One thing the vendor calculators leave out, in your favour for once. A bot answering at eleven on a Sunday night is doing work you were never going to pay for, because you were asleep. That saving does not show up anywhere in a customer service chatbot cost comparison, but the customer who got an answer and finished checking out is real money.

What kills the math is a bad deflection rate, and deflection is downstream of your help articles. A bot pointed at a thin help center will handle maybe a fifth of your questions. Pointed at a decent one, it handles a lot more.

What we charge, and how our bot works

SupportifyGPT is $15 a month for one seat, $39 for up to three, and $79 for up to ten. Every feature is on every plan, including the $15 one. The support bot is not an add-on, there is no per-answer fee, and there is no cap on how many questions it answers. Your bill in a bad month is the same as your bill in a quiet one. Full detail is on our pricing page, and the trial is 14 days.

How the bot works, plainly, because this is the part vendors are vague about. Hand-written rules read the question and search your own help articles for the ones that might answer it. Then a model reads those articles and writes a reply in normal sentences, rather than pasting you a list of links.

The important limit: it answers from your articles or it says it does not know. It cannot invent a returns policy you never wrote. When something goes wrong on our side it falls back to quoting the article it found, so a bad day makes it less useful rather than wrong. And when it drafts a new help article from a solved ticket, that draft waits for you to approve it. Nothing it writes goes live unreviewed.

Where we are honestly not the fit: there is no phone system or call queue, no built-in CRM, and no mobile app yet. If you need those, buy elsewhere and we will not be offended. If you want a shared inbox, tickets, a chat widget, a help center and the bot for the price of a couple of coffees, start a free trial and see what your own questions look like going through it.

Five things to ask before you pay

• What exactly do you count, and what is the overage rate past my allowance?
• What is the price at twice my current volume?
• Is the bot included, or is it a line item on top of my seats?
• Does the bot hand off into the same inbox my team already works in?
• How many hours of writing do I need to do before this is useful?

Get the first four in writing. The fifth one nobody can answer for you, and it is the one that decides whether any of the rest mattered. If you are still shortlisting, our guide to choosing help desk software covers the wider decision.