Skip to content
Panoptix AI
AI for Business

Best AI Chatbots for Customer Support on Your Website (2026)

Intercom Fin, Zendesk, Tidio, Chatbase, Crisp, Freddy and HubSpot compared: 2026 pricing models, free plans, handoff, accuracy risk and bot disclosure rules.

By Panoptix Editorial TeamPublished 13 min read

Disclosure: We may earn a commission when you buy through links on this page, at no extra cost to you. This never affects our ratings. Learn more.

AI customer support chat widget on a website answering a customer question
Table of contents

If you want the short answer: Intercom Fin is the best AI chatbot for customer support in 2026. It's the most complete AI agent in this group, it works on top of several other helpdesks (not just Intercom's), and you pay $0.99 per outcome instead of a big flat fee. For a small budget, Crisp gives you a full shared inbox plus an AI agent for a flat monthly price. For an online store, Tidio Lyro is the easiest fit, with a free plan and Shopify built in.

The bigger decision isn't the brand. It's the pricing model. Some of these tools charge per human seat, some per AI resolution, some per message or "session," and the same bot can cost $40 or $4,000 a month depending on which model you land on and how busy your chat widget is. We'll walk through that, plus the stuff most roundups skip: what happens legally when your bot makes something up, and the disclosure rules that took effect in the EU this August.

Quick picks

  • Best overall: Intercom Fin — strongest agent, pay per outcome, works with Salesforce, HubSpot, Freshdesk and others.
  • Best for small budgets: Crisp — free plan to start, AI included from $45/month with four seats.
  • Best for ecommerce: Tidio Lyro — free plan, Shopify integration, simple setup.
  • Best if you already use Zendesk: Zendesk AI agents.
  • Best if your data lives in HubSpot: HubSpot Customer Agent.
  • Best DIY bot trained on your docs: Chatbase.
  • Best cheap full helpdesk: Freshdesk with Freddy AI.

Our pick

Intercom Fin

4.6

Best for: Support teams that want the most capable AI agent and can budget per resolution · $0.99 per outcome; Intercom seats from $29/mo

Intercom Fin wins overall. Vendor-published numbers aren't independent tests, but Fin pairs a strong agent with fair billing: one charge per conversation, human handoffs don't count as a resolution, and you can bolt it onto your current helpdesk.

On a small budget, pick Crisp. Running a Shopify store? Start with Tidio Lyro.

Try Intercom Fin(opens in a new tab)

How these chatbots charge you (read this first)

Three pricing models dominate, and they reward very different traffic patterns.

  1. Per resolution or outcome. You pay when the AI closes a conversation without a human. Intercom Fin ($0.99 per outcome), HubSpot Customer Agent ($0.50 per resolution since April 14, 2026) and Zendesk (a per-agent allowance, then paid resolutions) work this way. It's fair, since failed answers are free, but your bill grows as the bot gets better.
  2. Per conversation, session or message credit. Tidio counts Lyro conversations, Freshdesk counts Freddy "sessions" (a 72-hour window per customer), and Chatbase counts message credits. You pay whether or not the bot actually helped.
  3. Flat plan with included AI. Crisp bundles AI credits into each plan. Most predictable, lowest ceiling.

A quick illustrative example: a store with 1,000 AI-handled chats a month where the bot fully resolves 500 would pay about $495 in Fin outcomes, or about $250 with HubSpot's per-resolution rate (before seat costs). On a per-conversation model, all 1,000 chats count. Run your own numbers with real volumes before you pick.

Side-by-side comparison

AI customer support chatbots compared (September 2026)
Feature WinnerIntercom FinZendesk AI agentsTidio LyroChatbaseCrispFreshdesk Freddy AIHubSpot Customer Agent
Our rating4.64.34.34.14.24.04.2
Pricing$0.99 per outcome; Intercom seats from $29/moSuite from $55/agent/mo (annual); resolutions over allowance billed extraFree plan; Lyro from $32.50/mo (annual) for 50 AI conversationsFree (50 credits); Hobby $40/mo; Standard $150/moFree (no AI); Mini $45/mo with AI credits; Essentials $95/moFrom $19/agent/mo (annual); 500 AI sessions included, then $49 per 100Service Hub Pro from $100/seat/mo (annual); $0.50 per resolution
Best forSupport teams that want the most capable AI agent and can budget per resolutionTeams already running their helpdesk on ZendeskShopify and small online storesQuick custom bots trained on your docs, with API and actionsSmall teams that want flat, predictable pricingBudget-conscious teams that want a full helpdesk plus AIBusinesses whose customer data already lives in HubSpot CRM
Pricing modelPer outcomePer seat + per resolutionPer AI conversationMessage creditsFlat plan + AI creditsPer seat + per sessionPer seat + per resolution
Free planNoNoYesYesYes, no AINoYes, no Customer Agent
Free trial14 days14 days7 daysFree plan14 days14 days14 days on paid seats
Knowledge sourcesHelp center, docs, data connectorsHelp center, PDFs, Google Drive, BoxWebsite, help center, PDF, CSVFiles, website, Notion, Q&AHelp center, site crawl, files, past chatsKnowledge base, connected appsKB, website, CRM, PDFs, past chats
LanguagesDozensMany12+MultilingualMultilingual60+All HubSpot languages
Vendor-claimed resolution rate76% averageNot publishedUp to 67%Not publishedNot publishedNot published65-70% average
Swipe to see all tools →
Feature checklist
FeatureIntercom FinZendesk AI agentsTidio LyroChatbaseCrispFreshdesk Freddy AIHubSpot Customer Agent
Human handoff with contextYesYesYesYesYesYesYes
WhatsApp channelYesYesYesYesYesYesYes
Shopify integrationNoNoYesYesYesYesNo
Voice channelYesYesNoYesNoNoYes
Built-in helpdesk or inboxYesYesYesYesYesYesYes
Resolution analyticsYesYesYesYesNoYesYes

About the checkmarks: "Shopify integration" means a native, advertised integration; Fin and HubSpot can reach store data through connectors, APIs or third-party apps, but we didn't find a native Shopify listing on their pages. Crisp's analytics exist but we couldn't confirm AI-specific resolution reporting. And every resolution rate above is a vendor claim, measured by the vendor's own definition. Treat them as marketing, not benchmarks. We explain how we weigh claims like these on our how we test page.

The contenders, one by one

Intercom Fin

Pricing
$0.99 per outcome; Intercom seats from $29/mo
Best for
Support teams that want the most capable AI agent and can budget per resolution

Fin is Intercom's AI agent, and it's now sold as a standalone product. The pricing page lists $0.99 per outcome across chat, email, voice, WhatsApp, SMS, Instagram and Messenger, with a 14-day unlimited trial. You can run it inside Intercom (seats start at $29 on Essential, $85 on Advanced and $132 on Expert) or plug it into Salesforce, HubSpot, Freshworks, Front, Zoho, Gorgias and others, with a 50-outcome monthly minimum off Intercom.

Intercom claims Fin averages a 76% resolution rate across 12,000+ customers. That's a vendor number; a customer quoted on Intercom's own pricing page reports 50%, which is probably a more realistic planning figure. The billing mechanics are good: you're charged once per conversation, and it supports procedures (multi-step workflows) and data connectors so it can look things up, not just recite articles.

Downside: cost scales with success. A busy site can spend more on Fin than on seats.

Zendesk AI agents

Pricing
Suite from $55/agent/mo (annual); resolutions over allowance billed extra
Best for
Teams already running their helpdesk on Zendesk

If your team already lives in Zendesk, its AI agents are the path of least resistance. They're included in every Suite and Support plan, with Suite Team at $55 and Suite Professional at $115 per agent per month billed yearly. Each plan includes a resolution allowance: 5 automated resolutions per agent per month on Team, 10 on Professional and Growth, 15 on Enterprise. Beyond that you pay per resolution; third-party breakdowns put it at about $1.50 committed or $2.00 pay-as-you-go.

The agents can pull from your help center plus external sources like PDFs, Google Drive and Box, and work across messaging, email and voice. Copilot for human agents is a separate $50 per agent add-on. Zendesk hasn't published an average resolution rate that we could find.

Tidio Lyro

Pricing
Free plan; Lyro from $32.50/mo (annual) for 50 AI conversations
Best for
Shopify and small online stores

Tidio is built for small online stores, and Lyro is its AI agent. There's a free plan (50 billable human conversations, live chat and ticketing), and Lyro pricing starts at $32.50 a month billed annually for 50 AI conversations. Tidio's pages quote slightly different starting prices in different places, so check the checkout screen. Paid Tidio plans start at $24.17 a month billed annually.

Lyro learns from your website, help center, PDFs and CSVs, supports 12+ languages, and runs on Anthropic's Claude model, per Tidio's Lyro page. Tidio claims Lyro solves "up to 67%" of customer problems; the Premium tier even advertises a guaranteed 50% resolution rate with pay-per-resolution billing. Shopify, WhatsApp, Instagram and Messenger connect out of the box.

Chatbase

Pricing
Free (50 credits); Hobby $40/mo; Standard $150/mo
Best for
Quick custom bots trained on your docs, with API and actions

Chatbase is the "upload your docs and get a bot" option, and it's good at it. You can train on files, websites, Notion and Q&A pairs, add actions (order lookups via Shopify, invoices via Stripe, custom API calls), and deploy to your site, WhatsApp, Slack, email or voice.

Pricing is credit-based: Free gives 50 message credits and one agent (deleted after 14 days of inactivity), Hobby is $40 a month for 700 credits, Standard $150 for 4,000 credits with a helpdesk and API access, Pro $500 for 15,000. Removing the Chatbase branding costs $99 a month extra. It's great for a fast pilot, but at scale credit pricing can surprise you.

Crisp

Pricing
Free (no AI); Mini $45/mo with AI credits; Essentials $95/mo
Best for
Small teams that want flat, predictable pricing

Crisp is our budget pick because the price doesn't jump every time a customer opens the chat. The plans are Free (two seats, no AI), Mini at $45 a month for four seats with $5 of AI credits (about 90 automated conversations), Essentials at $95 for 10 seats with $25 of credits (about 450), and Plus at $295.

The AI agent trains on answer snippets, your knowledge base, a website crawl, past conversations and uploaded files, and it has confidence settings that push shaky answers to a human. WhatsApp, Messenger, Instagram and a Shopify integration are available. Crisp doesn't publish a resolution rate, and its reporting is lighter than Intercom's or Zendesk's.

Freshdesk Freddy AI

Pricing
From $19/agent/mo (annual); 500 AI sessions included, then $49 per 100
Best for
Budget-conscious teams that want a full helpdesk plus AI

Freshdesk is the cheapest full helpdesk here at $19 per agent per month on Growth (billed annually), and every plan includes 500 Freddy AI Agent sessions. After that it's $49 per 100 sessions, which works out to $0.49 per session. There's no free plan, just a 14-day trial.

The Freddy AI Agent claims 60+ languages, works across web chat, email, WhatsApp and social, and connects to Shopify, Stripe, PayPal and FedEx for order-type questions. Note that sessions bill whether or not Freddy resolves anything. Freddy Copilot for agents is $29 per agent per month on Pro and Enterprise.

HubSpot Customer Agent

Pricing
Service Hub Pro from $100/seat/mo (annual); $0.50 per resolution
Best for
Businesses whose customer data already lives in HubSpot CRM

HubSpot's Customer Agent makes the most sense when your contacts, deals and ticket history are already in HubSpot. It trains on your knowledge base, website, CRM data, past conversations and PDFs, and works over chat, email, WhatsApp, Messenger and voice.

It costs $0.50 per resolution (50 HubSpot Credits) for messaging and $0.50 per minute for voice. The catch is entry cost: HubSpot says it requires Service Hub Professional or Enterprise, and Professional runs $100 per seat per month on an annual commitment plus a $1,500 onboarding fee. HubSpot claims a 65–70% average resolution rate; again, a vendor number.

Accuracy, hallucinations and who's liable

Every bot here uses a large language model, and every one of them can state something false with total confidence. The legal position is not "the AI said it, not us."

In Moffatt v. Air Canada, 2024 BCCRT 149, decided February 19, 2024, a British Columbia tribunal ruled that Air Canada was responsible for its chatbot telling a customer he could claim a bereavement fare after booking, which contradicted the airline's actual policy. Air Canada argued the chatbot was effectively a separate legal entity. The tribunal rejected that, saying the airline is responsible for all information on its website, whether it comes from a static page or a chatbot. Damages were small, about C$650, but the principle travels: treat bot answers as company statements.

What that means in practice:

  • Keep policies out of improvisation. Refunds, pricing, warranties, eligibility and anything medical or legal should come from fixed, reviewed answers or trigger a handoff.
  • Use grounding settings. Most tools let you restrict answers to your own content. Turn that on and turn "general knowledge" off.
  • Fix the source, not the bot. When an answer is wrong, the cause is usually a stale or contradictory help article. Air Canada's bot contradicted another page on the same site.
  • Review transcripts weekly for the first couple of months.

Telling users they're talking to a bot

This is now a legal requirement in places, not just good manners.

EU AI Act. Article 50 applies from August 2, 2026. AI systems that interact directly with people must be designed so people are told they're dealing with an AI, "from the start of the first interaction," clearly and accessibly, unless it's obvious. The formal duty sits with the provider (the vendor), but if you rename the bot "Sarah" with a human photo, you're working against it. Reporting on the EU's 2026 Digital Omnibus deal says high-risk deadlines moved but Article 50 kept its August 2026 date.

California. The state's bot disclosure law, Business and Professions Code section 17941 (from SB 1001, in force since July 1, 2019), makes it unlawful to use a bot to mislead people about its artificial identity to encourage a purchase. A clear, conspicuous disclosure is the safe harbor. A sales-leaning support bot in California should say what it is.

The simple fix works everywhere: a greeting like "Hi, I'm the AI assistant for [Store]. I can help with orders and returns, or connect you with a person." Not legal advice; ask counsel if you're in a regulated industry.

Designing the handoff to a human

The difference between a bot customers tolerate and one they hate is almost always the escalation path.

  • Always offer an exit. "Talk to a person" should work at any point, and the bot shouldn't argue.
  • Escalate on triggers: repeated rephrasing, angry language, keywords like "cancel," "refund" or "complaint," high-value customers, and low-confidence answers.
  • Pass the full transcript so the human doesn't restart with "How can I help?"
  • Be honest about hours. Outside business hours, collect an email and say when someone will reply.
  • Remember the billing angle: on per-resolution tools, a handoff isn't billed as a resolution, so there's no financial reason to trap people with the bot.

Setup checklist

  1. Audit your knowledge base first. Delete outdated articles, resolve contradictions, and write clear pages for your top 20 questions.
  2. Pick the pricing model that fits your volume and run the math on last quarter's chat numbers.
  3. Restrict the bot to your content and write explicit instructions for tone and forbidden topics. The principles in our prompt writing guide apply to bot instructions too.
  4. Add the AI disclosure to the greeting and widget header.
  5. Set escalation rules and test them with angry, vague and off-topic messages.
  6. Connect order data (Shopify, Stripe) only with read access at first.
  7. Launch on a slice of traffic, for example one page or off-hours only.
  8. Review transcripts weekly and track resolution rate, handoff rate and CSAT against your pre-bot baseline.
  9. Automate the follow-ups like tagging, CRM updates and Slack alerts; our n8n automation guide shows how.

Which one should you choose?

  • Growing support team, chat is a core channel: Intercom Fin. Plug it into your existing helpdesk if you don't want to migrate.
  • Already on Zendesk, Freshdesk or HubSpot: use the native agent first. Switching helpdesks for a chatbot rarely pays off.
  • Small team, tight budget: Crisp Mini at $45 a month, or Freshdesk Growth if you need a proper ticketing system.
  • Shopify store: Tidio Lyro. For the rest of your store's AI stack, see our AI tools for ecommerce and Shopify roundup.
  • You want a custom bot fast, or need an API: Chatbase.

For the wider picture of where a support bot fits among your other tools, see our pillar guide to the best AI tools for small businesses.

Frequently asked questions

What is the best free AI chatbot for customer support?

Tidio has the most useful free option for a small site: a free plan plus a small allowance of Lyro AI conversations. Chatbase's free tier (50 message credits) is fine for testing but not for real traffic, and Crisp's free plan doesn't include AI.

How much does an AI customer support chatbot cost?

As of September 2026, anywhere from about $40 a month (Chatbase Hobby, Crisp Mini at $45) to thousands for busy sites. Per-resolution pricing runs $0.50 (HubSpot) to $0.99 (Intercom Fin) per resolved conversation, on top of any seat costs.

Is a company liable if its chatbot gives wrong information?

It can be. In Moffatt v. Air Canada (February 2024), a Canadian tribunal held the airline responsible for its chatbot's incorrect refund advice and rejected the argument that the bot was a separate entity. Treat bot answers as your own statements.

Do I have to tell customers they're talking to an AI?

In the EU, yes: AI Act Article 50 requires people be informed they're interacting with an AI from August 2, 2026, unless it's obvious. California's bot law requires disclosure when a bot is used to encourage sales. Disclosing everywhere is the simplest policy.

What resolution rate should I expect?

Vendors claim 65–76% averages, but those use their own definitions. Plan for less at first, and expect the number to track the quality of your knowledge base more than the brand of bot.

Share:

Written by

Panoptix Editorial Team

Our editors test AI tools hands-on for weeks before we publish a word. We pay for our own subscriptions and never accept payment for rankings.

How we test AI tools →