Two website chatbot plans both say "2,000 per month". On one, that's 2,000 answers to your customers. On the other, it could be a few hundred — because each reply from a stronger AI model uses several credits at once.
Nothing about that is hidden, exactly. It's in the pricing table, a tooltip or the FAQ. But most small business owners don't have time to decode it, and they shouldn't have to.
The short answer: AI chatbots count usage in one of four ways — per reply (message), per credit, per conversation, or per human-answered conversation with AI sold separately. Credits are the tricky one: many tools charge more credits per reply for more advanced AI models, so the number on the plan isn't the number of answers you get. To compare plans, estimate your monthly chats, multiply by replies per chat, then apply each tool's counting rules.
The four ways AI chatbots count usage
| How it's counted | What one unit means | Watch out for |
|---|---|---|
| Per message (reply) | One AI answer to a customer | Whether customer questions or your own replies also count |
| Per credit | A share of a reply — the cost depends on the AI model | Advanced models can use many credits for one reply |
| Per conversation | One whole chat, however many replies | What starts a "new" conversation; long chats are cheaper, short chats aren't |
| Per human conversation + AI add-on | Chats your team answers; AI bought separately | Paying twice — for the helpdesk and for the AI |
None of these is wrong. They just behave very differently once real customers start chatting.
What a "model multiplier" is
Chatbot tools run on large language models (LLMs) from companies like OpenAI, Anthropic and Google. Newer, larger models cost the tool more to run, so many chatbot platforms pass that on: pick a stronger model, and each reply costs more credits.
Here's how some popular tools describe it on their own pricing pages (as of September 2026):
- SiteGPT says GPT-4.1 messages are 10 times the cost of GPT-4.1-mini messages, and counts both the customer's question and the AI's reply as messages. Verdesk vs SiteGPT
- FastBots charges 1 credit per reply on standard models and 5–10 credits on advanced models. Verdesk vs FastBots
- Chatling charges between 0.5 and 25 AI credits per reply, depending on the model. Verdesk vs Chatling
- Chatbase says a message costs a number of credits depending on the model you choose. Verdesk vs Chatbase
Other tools count differently: Wonderchat estimates about 10 credits per conversation on average, Yamidoo counts AI-answered conversations, and Tidio sells its Lyro AI agent as a separate add-on. Prices change, so always check the provider's own page before you decide.
Why more choice isn't better for a small business
Being able to pick your AI model sounds like a feature. For most small businesses, it's a burden:
- You're not an AI engineer. Few owners can say whether GPT-4.1 or Claude Sonnet is the right model for a flower shop's delivery questions — and they shouldn't need to.
- The "better" option drains your allowance quietly. Switch to a stronger model and the same monthly credits cover a fraction of the replies.
- Bills stop being predictable. When cost per reply varies, so does the month you run out.
- Every extra decision slows you down. Some research on choice overload suggests that too many options can make people less likely to decide at all. Setup should take minutes, not an evening of reading model comparisons.
What a business owner actually wants is simpler: good answers, and a number on the plan that means what it says.
How to work out what a plan really costs
Use this five-step check on any chatbot plan:
- Estimate your monthly chats. Start with your monthly website visitors and assume a small share will ask something. If you get 30 enquiries a week by email today, expect more once answers are instant.
- Estimate replies per chat. Most website questions take one to three replies. Use 3 to be safe.
- Check what counts. Do customer questions count? Your team's replies? Test messages while you set it up?
- Apply the model rule. If replies cost more on the model you'll use, multiply.
- Check what happens at the limit. Does the bot stop answering, pause, or charge overage? Can you top up?
A worked example
Say you expect 300 chats a month, averaging 3 AI replies each — 900 replies.
| Counting method | Units used |
|---|---|
| 1 message per reply | 900 messages |
| Credits, standard model (1 credit per reply) | 900 credits |
| Credits, advanced model (10 credits per reply) | 9,000 credits |
| Per conversation | 300 conversations |
| Message counting that includes the customer's questions | 1,800 messages |
Same customers, same questions — and a tenfold difference in what you need to buy. Conversation-based pricing looks efficient here; it gets less so when most chats are a single question.
Questions to ask before you buy
- Is the AI included, or is it an add-on?
- Does one reply always cost the same, whichever model answers?
- Do customer questions, my own replies or test messages count?
- What happens when I hit the limit, and can I buy more?
- Which plan removes the vendor's branding, and what does that cost?
- Is there a free plan to test on my real website?
How Verdesk counts: one reply, one message
Verdesk keeps it deliberately simple:
- Every AI reply to a customer counts as one message. Always.
- Customer questions are free, and so are your own replies.
- There's no model to pick. A fast, current AI model is included in every plan.
- Testing doesn't touch your live messages — each plan has a separate test allowance.
- Plans: Free (300 messages a month), Starter ($29/month for 2,000), Growth ($99/month for 7,500) and Business ($299/month for 25,000).
- Top-ups on paid plans: 1,000 extra messages for $25, and they never expire.
So when a plan says 2,000 messages, that's 2,000 answers to your customers. See the full pricing.
Frequently asked questions
What is a message credit in an AI chatbot?
A credit is a unit a chatbot platform uses to measure usage. On many platforms, one AI reply uses one or more credits depending on the AI model, so a plan's credit count isn't always the number of replies you'll get. Check how many credits a reply costs on the model you plan to use.
Why do some chatbots charge more for advanced AI models?
Larger, newer AI models cost more to run, so some chatbot tools charge several credits per reply when you choose them. It lets you pick a stronger model, but it makes your monthly allowance harder to predict.
Is per-conversation pricing better than per-message pricing?
It depends on your chats. If conversations are long, per-conversation pricing can work out cheaper. If most visitors ask one or two quick questions, per-message pricing is usually easier to predict and compare.
What happens when a chatbot runs out of messages?
It depends on the tool. Some stop answering until the next billing period, some pause AI replies, and some charge overage. With Verdesk, the chatbot stops answering until you top up, upgrade or the new period starts, and you're emailed before you reach the limit.
The bottom line
The cheapest-looking chatbot plan isn't always the cheapest in use. Before you compare prices, compare what a unit actually means. And if you'd rather not do the maths at all, pick a tool where one reply is one message.
Start free with Verdesk — no card, and no model to choose.