Short answer
An AI chatbot costs anywhere from zero to a few thousand dollars a month. The price is not driven by the language model, which costs a fraction of a cent per conversation. It is driven by how the tool bills you: per seat, per conversation, per resolution or per active contact. The prices below were checked on each vendor's official pricing page in September 2026.
At GUSTA we sit on both sides of this math. We build Nola AI, a SaaS with public pricing, and we also build custom assistants. So this article lists our own prices next to the others, and explains when none of these tools is the best option.
The four pricing models
Before comparing numbers, it helps to know what each tool is actually charging for. Two products with the same starting price can end the month very far apart.
Per seat. You pay a license for each team member who uses the tool. That is how Intercom and Zendesk work. It is cheap for small teams and grows with headcount. The AI is usually billed on top.
Per resolution or outcome. The tool charges when the AI closes a conversation without human help. It sounds fair, but the vendor decides what counts as resolved. Read that definition carefully before signing.
Per conversation or interaction. Each conversation where the visitor engages uses one unit of the plan. Tidio, Gorgias and Nola AI work this way. For an online store it is the most predictable model, because conversation volume follows traffic.
Per active contact. Some platforms, mostly WhatsApp-heavy ones, bill per contact who engaged within a 24-hour window.
Then add the fees that do not show up in the starting price: setup, mandatory onboarding, an official WhatsApp number and, for cross-border billing, currency conversion.
Price table, September 2026
| Tool | Billing | Starting price | AI support |
|---|---|---|---|
| Shopify Inbox | Free | US$ 0 | AI sales associate included, no published usage limit |
| Tidio | Per conversation | US$ 29 a month (Starter) | Lyro from US$ 39 a month for 50 conversations |
| Intercom | Per seat and per outcome | US$ 29 per seat a month (Essential, annual) | Fin at US$ 0.99 per outcome |
| Zendesk | Per agent and per resolution | US$ 19 per agent a month (Support Team, annual) | Monthly resolutions included per agent, extras billed per use |
| Gorgias | Per ticket volume | US$ 10 a month (Starter) | AI Agent from US$ 30 a month, extra interactions at US$ 1.50 |
| Nola AI | Per conversation | R$ 497 a month (Growth, 500 conversations) | AI included, 14-day free trial with no card |
A few notes. Some vendors show regional price lists, so what you see may differ by country. Annual plans are cheaper than monthly ones almost everywhere. Nola AI is billed in Brazilian reais. And the links to every official page are at the end of the article, because software pricing changes without notice.
What the AI behind the chat actually costs
This is the part that surprises people. Let's run the numbers with gpt-4o-mini, a small OpenAI model widely used for support, priced at US$ 0.15 per million input tokens and US$ 0.60 per million output tokens.
Picture a typical store conversation with six customer messages. For each message, the system sends the model about 3,000 tokens (instructions, products found in the catalog and the conversation history) and gets back a reply of around 250 tokens.
| Item | Tokens in the conversation | Cost |
|---|---|---|
| Input | 18,000 | US$ 0.0027 |
| Output | 1,500 | US$ 0.0009 |
| Total per conversation | 19,500 | US$ 0.0036 |
| One thousand conversations | 19.5 million | US$ 3.60 |
Indexing the catalog is cheap too. The text-embedding-3-small embedding model costs US$ 0.02 per million tokens, so turning 10,000 products into search vectors costs a few cents.
In other words, when a tool charges somewhere between ten cents and a couple of dollars per conversation or resolution, nearly all of that pays for something else. It pays for the catalog integration, the search that answers price and stock questions without making things up, the dashboard, the human handoff queue, reporting, support and the maintenance work every time the store platform changes its API. For model choices in your own projects, our breakdown of GPT-5.6 Sol, Terra and Luna shows how prices shift between large and small models.
Building your own chatbot: where the money goes
With tokens this cheap, the obvious question is why not build a chatbot straight on the API. You can, and sometimes it is the right call. But the list of what has to exist beyond the model call is long:
- a connector to the store platform that reads variants, sale prices and stock;
- a sync that keeps up with catalog changes during the day;
- search that treats "under $400" as a price filter, not as similar-looking text;
- a widget that works well on phones and does not weigh down the site;
- a conversation dashboard, human queue, business hours and lead capture;
- funnel metrics, data export and team access control;
- monitoring, usage limits and someone to keep all of it running.
Each item looks small. Together they add up to months of development and a fixed maintenance cost. Building pays off when the process is too specific for an off-the-shelf tool, or when volume makes the subscription more expensive than the team. Our article on custom AI solutions for business goes through those signals.
How to estimate the cost for your store
- Count conversations, not visits. The best starting point is your current support volume on WhatsApp, chat and email. A visitor who only browses costs nothing on a per-conversation plan.
- Split what the AI handles from what goes to a person. On per-seat tools, the bill depends on how many people keep answering.
- Check what happens at the plan limit. Some tools bill overage automatically. Others pause automated replies until the next cycle.
- Add the hidden fees: setup, onboarding, official WhatsApp and currency conversion.
- Do the math per unit. Divide the monthly price by the number of conversations you actually expect. It is the only fair way to compare tools with different billing models.
Nola AI plans
Nola bills per conversation and only counts conversations where the visitor actually engaged. Every plan includes the AI, catalog sync, the in-chat cart, the human support queue and the dashboard with funnel and export.
| Plan | Monthly price | Conversations per month | Catalog products |
|---|---|---|---|
| Growth | R$ 497 | 500 | up to 5,000 |
| Pro | R$ 997 | 2,000 | up to 20,000 |
| Business | Custom | Custom | Custom |
When the plan limit runs out, automated replies pause until the cycle resets, unless you upgrade or buy an extra pack.
| Extra pack | Conversations | Price |
|---|---|---|
| Growth | 250 | R$ 149 |
| Pro | 1,000 | R$ 349 |
| Business | 2,500 | R$ 599 |
The trial lasts 14 days, needs no card and includes up to 500 conversations. Without a subscription, the assistant leaves the site on day 15. On Shopify, billing goes through the Shopify invoice.
If you are still deciding on the kind of assistant, the article on AI chatbots versus AI personal shoppers explains what each one delivers. Once you have picked a tool, the guide on how to add an AI chatbot to your website shows where the snippet goes on each platform.
Frequently asked questions
Is there a free AI chatbot?
Yes, with limits. Shopify Inbox is free for Shopify stores, and Tidio's free plan includes 50 Lyro conversations once. For a store with steady traffic, those limits run out fast.
Why does the price per resolution vary so much between tools?
Because each vendor defines what a resolution is and what the price covers. Some charge only for the AI. Others bundle the license, the dashboard and the integration. Always compare total cost divided by the number of conversations.
Is per-seat or per-conversation pricing better?
For an online store with a small team and lots of repeated questions, per-conversation pricing is usually more predictable. For companies with a large support team and few automated conversations, per-seat pricing can make more sense.
Is it cheaper to use the OpenAI API directly?
The model is cheaper, at a fraction of a cent per conversation. The total rarely is, because integration, dashboard, catalog search and maintenance become your team's job.
Does the price include WhatsApp?
It depends on the tool. The official WhatsApp Business Platform has its own fees charged by Meta, separate from the subscription. In Nola AI, WhatsApp is an option for human support and for closing the sale.
Sources
Prices checked on official pages in September 2026: Intercom, Zendesk, Tidio, Gorgias, OpenAI and Shopify Inbox.
CTA
Want to know what Nola AI would cost at your store's volume? Try Nola AI free for 14 days, no card, and watch the real conversation count in the dashboard. If your case calls for a custom assistant, talk to GUSTA.
Comments
No comments yet. Be the first to comment.