AI chatbot pricing in 2026 — four billing models, decoded
"$0.99 per resolution" looks like a cheap chatbot. So does a flat monthly subscription. They both can be — and they both can be ruinous — depending on which usage curve your team lands on. The four billing models in market in 2026 produce wildly different bills at the same conversation volume, and reading them wrong is how teams end up with a four-figure invoice they didn't budget for.
This is the unvarnished breakdown. No vendor names — the patterns are the patterns.
Model 1 — Per AI-resolved conversation
You pay a unit price each time the AI fully resolves a customer issue without human intervention. Headline rates in market today range $0.45 – $2.50 per resolution, often on top of a base platform / seat fee.
The math (typical $0.99/resolution + $39/seat):
- 100 resolutions/month: ~$140
- 1,000 resolutions/month: ~$1,030
- 10,000 resolutions/month: ~$9,930
Watch out for: the definition of "resolution" varies — some vendors count each AI response, some require the customer to confirm satisfaction, some count escalations as failed resolutions (free) and non-escalations as resolved (charged). Read the fine print.
Best fit: low conversation volume (under ~500/month) or teams whose finance prefers usage-based cost-of-goods accounting.
Model 2 — Per seat
You pay a monthly per-user fee for each human agent or admin who logs in. AI features are bundled or sold as an add-on per seat.
The math (typical $59/seat × team of 8): $472/month before an AI feature is enabled. AI add-ons commonly add $25–50/seat → $672–872/month for the same team.
Watch out for: the price scales with team size, not with how much value you're getting from the AI. A 10-person team that uses the AI for 80% of conversations pays the same as a 10-person team that uses it for 5%.
Best fit: teams whose support workflow centers on a shared inbox with multiple human operators — the per-seat model fits the staffing model.
Model 3 — Per conversation (gross)
You pay per conversation, full stop — whether the AI handles it, escalates it, or the visitor abandons after one message. Rates in market range $0.20 – $1.50 per conversation.
The math (typical $0.30/conversation):
- 1,000 conversations/month: $300
- 10,000 conversations/month: $3,000
Watch out for: "conversation" definitions vary on session timeouts (some count two minutes idle = new conversation, doubling the bill). Bots, scrapers, and 1-message bounces count too.
Best fit: teams with consistent, predictable traffic where the per-conversation rate × expected volume fits the budget cleanly.
Model 4 — Flat plan per website
You pay a fixed monthly fee per website (or per workspace), with a monthly message ceiling per plan tier. Above the ceiling, you upgrade to the next tier or hit a soft cap.
The math (NovixAgent AI plans — $29–$399/mo flat):
- Starter $29/mo — small site, 1,000 messages a month
- Pro $99/mo — adds voice (150k chars / mo of TTS), Cal.com / Calendly booking, expanded cap
- Business $399/mo — white-label, persona images, larger cap, expanded voice quota (500k chars / mo)
- Custom — sales-led contracts, terms scoped per agreement; on the roadmap: SSO and multi-tenancy (see the roadmap section of /pricing for current status)
Watch out for: hit the monthly message cap and you're either declined-until-upgrade or charged for an overage pack. The cap is the trade-off for the flat bill.
Best fit: teams that want a predictable line item and conversation volume that fits inside a tier.
How to actually compare two vendors
Sticker pricing comparison is a trap. Run these three numbers instead:
- Expected conversation volume. Pull last month's analytics — visitors × conversation-engagement rate. If you don't have it, assume your site's monthly unique visitors × 1-2% as a starting estimate.
- Expected AI-handle rate. What share of questions do you expect the AI to resolve without a human? A well-documented site hands off far more than one with thin docs — so model it conservatively at first, watch your real numbers, and expect the rate to climb as you close the content gaps the agent surfaces.
- Effective $/month at your volume. Run the math on each model with those two numbers. The cheapest sticker often becomes the most expensive at your real volume.
One more thing — what the price doesn't tell you
The pricing model captures cost. It says little about quality:
- Does each answer cite a source page, or does the agent invent answers?
- Does the AI know when to escalate, or does it confidently mislead the visitor?
- Does it speak the visitor's language, or only English?
- Does it auto-re-crawl when your prices change, or stay frozen at the date you trained it?
A cheap chatbot that hallucinates prices on an e-commerce store costs you refunds. Compare on price and on grounding architecture. The first is on the pricing page; the second is on the company's engineering blog (or visibly absent from it).
For our own answer to "which model are you?" — flat plan per website, no per-resolution surprises. See /pricing.