AI Chatbot for Your Website: 5 Services, Prices and How to Choose

An AI chatbot for a website is a widget that answers visitors in plain language, drawing on your own content: website pages, a knowledge base, answers to frequent questions. Unlike old button-based bots, it understands a question however it’s phrased, and when it doesn’t know the answer, it hands the conversation to a person.

Below are five services you can add to a website without building your own bot, with prices checked on the vendors’ websites on 27 September 2026. Plus how to set up a bot so it doesn’t make answers up, what to know about data security, and what to measure after launch.

Rule-based bot or AI agent: what’s the difference

A rule-based bot follows a script: “Choose a topic, then Delivery, then London”. It’s predictable and cheap, but it gets lost as soon as someone writes something outside the script. An AI agent is built on a large language model: it reads the question, looks for the answer in your materials and puts it in its own words.

In practice most services combine the two: a script collects a contact or books a call, and the AI agent answers open questions.

5 AI chatbot services compared

Prices are from the vendors’ pricing pages, billed annually, excluding tax. Services change their plans often, so check the current page before deciding.

Intercom (Fin): for high-volume support

Intercom is a full support platform: chat, email, knowledge base, reports. Its AI agent, Fin, answers based on your content and can take actions, for example checking an order status through an integration.

The pricing model: from $29 a month per agent seat (Essential plan, billed annually) plus $0.99 for every conversation Fin resolves on its own. If the bot can’t cope and hands over to a person, you don’t pay for that one. It’s a fair model, but worth calculating in advance: 2,000 resolved conversations a month is already about $2,000 for AI alone. There’s a 14-day free trial with no card required.

Who it’s for: companies with hundreds of requests a day, where support is a separate department. For a site with a dozen questions a week, Intercom is overkill.

Tidio (Lyro): for online stores and small businesses

Tidio is easy to set up: there are ready-made plugins for WordPress, WooCommerce and Shopify, and a knowledge base for its AI agent, Lyro, can be built from your site pages and FAQ in an hour.

The free plan includes 50 conversations a month and a one-off pack of 50 Lyro conversations, enough to see how the bot handles your real questions. Paid plans start at $24 a month, and Lyro is billed separately, from $32.50 a month for 50 AI conversations. For a store with seasonal peaks it’s convenient that the Lyro pack can be increased for the sales month and reduced afterwards.

Who it’s for: online stores and small companies that need a chat with AI without a separate support department.

HelpCrunch: a Ukrainian-built service with AI agents

HelpCrunch is a support platform built by a Ukrainian team, often considered a more affordable alternative to Intercom: website chat, messengers, email, knowledge base, pop-ups.

The Basic plan at $12 a month doesn’t include AI. AI agents come with the Pro plan, from $20 a month per agent. It includes 50 AI conversations a month; more are sold in packs, from $29 a month for an extra 100 conversations. You pay per conversation rather than per resolution, so costs are easier to predict.

Who it’s for: small and mid-sized businesses that want chat, messengers and a knowledge base in one place for a reasonable price.

SendPulse: when customers message you on WhatsApp and Telegram

Many customers don’t type into a website widget at all; they switch to a messenger. SendPulse, a service founded in Ukraine, builds bots for WhatsApp, Telegram, Instagram, Facebook Messenger and Viber, and also has a website chat. All conversations are collected in one inbox.

The free plan allows up to 3 bots, 500 subscribers and 10,000 messages a month. The paid Pro plan is priced by the number of subscribers and shown in local currency for your region; it includes unlimited bots and chats, payments, an API and a ChatGPT integration for AI-generated answers.

Who it’s for: businesses that sell through messengers, such as services, education and local businesses. How to collect enquiries from forms, messengers and calls in one system is covered in our article on website CRM integration.

HubSpot (Customer Agent): if your CRM is already HubSpot

HubSpot‘s AI agent works inside its CRM: it sees the customer’s history, deals and tickets and answers with that in mind. It’s fully available in Service Hub Professional, from $90 a month per seat. AI conversations are paid for with HubSpot credits: some credits are included in the plan each month, and more can be bought beyond the limit.

Who it’s for: companies that already run sales in HubSpot. Buying HubSpot just for a chatbot is expensive; it’s simpler to pick one of the services above and pass enquiries into your own CRM.

A ready-made service or your own bot

You can build your own bot on the OpenAI or Anthropic API and embed it in your site. That makes sense when:

• the bot has to work with your internal systems (stock, schedules, customer accounts) more deeply than ready-made integrations allow;
• conversation volume is very high, and per-conversation SaaS pricing becomes more expensive than building and maintaining your own;
• your data can’t be passed to a third-party support service.

In other cases a ready-made service wins: launch in days rather than months, with an agent inbox, reports and human handover already built. Your own bot also means ongoing maintenance: models get updated, APIs change, and someone has to keep up.

How to set up an AI bot so it doesn’t do harm

The most common problem with an AI bot isn’t that it doesn’t know something, but that it confidently answers wrong: invents a discount, promises delivery you don’t offer, quotes an old price. To prevent that:

1. Build the knowledge base from real questions. Ask your sales team to write down the 30–50 questions customers ask most often, with answers. That’s a better foundation than just “read the website”.
2. Restrict the bot to its materials. Most services let you forbid answers on topics outside the knowledge base. “Let me check with a colleague” beats an invented answer.
3. Spell out what the bot never does: no discounts, no delivery promises, no medical or legal advice, depending on your niche.
4. Set up handover to a person. People should get a human on the first request, not after three rounds of “I didn’t understand you”. Out of hours, the bot collects a contact and promises a reply at a specific time.
5. Check the language. During the trial, ask the bot 20 questions in each language your customers use, with typos and informal phrasing, and see how it answers. Answer quality varies between services and languages.
6. Update the knowledge base along with the site: if prices or delivery terms change, update the bot too. Once a week, review conversations where the bot failed and add answers.

Security: what the bot knows about your customers

A chatbot connected to your CRM is another door to your data. In August 2025, attackers stole the access tokens through which Salesloft’s Drift chatbot connected to customers’ CRMs and exported data from the Salesforce instances of hundreds of companies. Unit 42 (Palo Alto Networks) published an analysis of the incident.

Lessons for any business:

• give the bot the minimum access it needs: if it only has to create leads in the CRM, it shouldn’t be able to read your whole customer base;
• keep a list of services connected to your CRM and disconnect the ones you don’t use;
• check where the service stores conversations and whether it signs a data processing agreement (DPA), which matters if you have customers in the EU;
• tell people in the chat that they’re talking to AI, and don’t ask for passwords, card numbers or documents through the bot.

The rest of the basic website security rules are in our article on website security.

What to measure after launch

The number of chat conversations on its own says nothing. Look at what affects revenue and your team’s workload:

• The share of requests resolved by the bot: how many conversations ended without an agent and without a repeat request.
• Handovers to people and their reasons: they show which answers to add to the knowledge base.
• Leads from chat in the CRM: how many contacts the bot collected and how many became customers.
• First response time outside working hours, before and after launch.
• Post-chat ratings and complaints about the bot.

Response time really matters. A study published in Harvard Business Review (an audit of 2,241 US companies) found that those who responded to an online enquiry within an hour were nearly seven times as likely to have a meaningful conversation with a decision maker as those who responded just an hour later, and more than 60 times as likely as those who waited a day (HBR, 2011). A bot that replies instantly and collects a contact at night addresses exactly that problem.

When you don’t need a chatbot

An AI bot doesn’t always justify the setup. It’s unnecessary if:

• your site gets a few dozen visitors a day and there aren’t many questions; it’s simpler to reply quickly yourself;
• all the answers are already on the site and people just struggle to find them; then the problem is your page structure, not the lack of a bot;
• your customers want to talk to a person: high-ticket services, complex B2B projects, healthcare.

In those cases a clear form, messenger buttons and a fast reply from your team work better. How to choose a channel for enquiries is in our article on forms, messengers and calls on a website. And if you’re planning a new site and want to budget for chat and integrations from the start, see what goes into the cost of a website.