AI Chatbot Pricing in 2026: Per-Resolution vs Flat-Rate, Compared

Most AI chatbot pricing pages are designed to be hard to compare. One charges per resolved conversation, one charges per message credit, one charges per seat, and one charges per "AI action" without defining what an action is. The monthly number on the pricing page is rarely the number you pay.

This post compares the three billing models actually in use in 2026, works out where each one becomes expensive, and shows the volume at which they cross over. Pricing was checked in September 2026 and will drift, so treat the shapes as durable and the exact figures as a snapshot.

The three billing models

1. Per resolution

You pay when the bot successfully handles a conversation without a human. Intercom's Fin is the best known example at $0.99 per resolved conversation.

The appeal is obvious: you pay for outcomes. The problem is that your bill scales with your success. If the bot gets better, you pay more.

There is no volume at which this gets cheaper per unit. A support team that automates well is punished for it.

2. Flat rate with a usage ceiling

You pay a fixed monthly fee for a bucket of messages or documents. Chatbase runs $40/month (Hobby) up to $500/month (Pro), with custom enterprise pricing above that. SiteGPT runs $39/month, with a Growth tier at $79/month.

Predictable, and the per-unit cost falls as you use more of your bucket. The catch is the overage policy, which is where these plans differ most and where the pricing page is usually vaguest. Ask two questions before you sign:

• What happens at 101% of the quota? Hard stop, or silent overage billing?
• Does an unanswered question count against the quota?

3. Per seat

Inherited from live chat tools, where seats meant humans. It makes little sense for an AI that has no seat, and it mostly survives in suites where the AI is bolted onto an existing helpdesk. If you are being quoted per seat for an AI feature, you are paying for the helpdesk, not the AI.

Where the models cross over

The interesting question is not "which is cheapest" but "at what volume does the answer change".

Take a flat plan at $79/month against per-resolution at $0.99:

• At 80 resolutions/month, they cost about the same.
• At 200 resolutions, per-resolution costs $198 and flat costs $79.
• At 1,000 resolutions, per-resolution costs $990 and flat costs $79.

So the crossover sits at roughly 80 conversations a month, which is under three a day. Almost any site with real traffic clears that bar. Per-resolution pricing is cheapest only for sites that barely use the product.

Where per-resolution genuinely wins: if your volume is spiky and low, or if you are running a pilot and want costs to start at literally zero, paying $0.99 a handful of times beats committing $79/month. It is also easier to justify internally, because the line item maps to a ticket a human did not have to answer. That is a real advantage and not just a billing gimmick.

The costs that are not on the pricing page

Content preparation. Every one of these tools is only as good as what you feed it. If your help docs are three years stale, budget the week it takes to fix them before you budget the subscription. This is the single biggest hidden cost and nobody sells it to you.

The escalation path. A bot that cannot hand off cleanly to a human creates angry customers, and angry customers cost more than the subscription. Check that handoff exists and that it carries the conversation across.

Failed answers. Some plans count a message the bot could not answer against your quota. You pay for the failure.

Branding removal. "Powered by" badges are commonly a paid upgrade. If you are an agency reselling this to clients, white-label is not optional, and it is often the tier above the one you were planning to buy.

What actually drives the bill

Underneath every one of these products, the model cost per message is small. Modern small models answer a grounded support question for a fraction of a cent, and a full 20-message conversation typically costs well under a cent in inference.

That gap between inference cost and sticker price is not necessarily unfair. You are paying for retrieval quality, the dashboard, the integrations, uptime, and someone else's on-call rotation. But it does mean the pricing is a business decision rather than a cost pass-through, which is exactly why the models vary so wildly for products that do nearly the same thing.

It also means you should be sceptical of any vendor who tells you per-resolution pricing is forced on them by model costs. It is not.

How to choose without a spreadsheet

1. Estimate conversations per month, not messages. Take your current support ticket volume and assume the bot will attempt all of them.
2. If that number is above roughly 100, start with flat-rate. The crossover maths above is not close.
3. Read the overage clause before the feature list. It matters more.
4. Check white-label pricing if you are an agency. It changes which tier you are actually buying.
5. Run your real questions through a trial. Retrieval quality varies more between products than pricing does, and it is the thing that determines whether any of this works.

A note on where this fits

We build Askably, which is a flat-rate, document-grounded chat widget, so we are not a neutral party here. Our own pricing is flat for the same reason we would recommend it to you: per-resolution billing means our incentive and your incentive point in opposite directions.

If you are earlier than that and just want to understand the machinery, RAG explained covers how these products find the right passage before answering, which is the part that actually decides whether the answer is any good.

Sources

• AI Chatbot Pricing Comparison 2026, SiteGPT
• Intercom Fin AI Review 2026
• Chatbase Review 2026

Pricing verified September 2026. All figures are list prices before negotiation, and enterprise deals routinely land well below them.

What model are you on, and did your bill land where the pricing page implied it would?