TL;DR: The hard part of a website chatbot isn’t the model. It’s that the bot’s knowledge and your website are two things, and only one of them gets updated. Nearly every option below asks you to maintain a second copy of what your business knows. Fin (formerly Intercom) is the strongest AI agent here and the only one with a published per-unit AI price: $49/mo standalone including 50 resolutions, then $0.99 each, with no seat fee. Zendesk (Suite Team from $55/agent/mo annual) bills only resolutions its own verification confirms, the stricter definition, but it won’t publish a rate for anything beyond your allowance. Tidio is the cheapest entry at $29/mo, with a catch worth knowing before you buy. HubSpot has a genuinely free chat widget that is rule-based, not AI. And Surmado Pro+ ($199/mo annual) includes a chatbot that reads the site we already maintain, which removes the second-copy problem entirely and is a much bigger commitment than installing a widget. Pick by whether you need a support desk, a website, or just answers.
Most people don’t shop for a chatbot because they want one. Something specific happened.
The same four questions keep arriving by email and you’ve started answering them with copy and paste. You added a chat widget two years ago, it told a customer something that stopped being true in 2024, and now you’re not sure what else it’s saying. You looked at a quote for a support platform and could not work out what it would actually cost you per month. Or you’re staring at a pricing page that charges per conversation and doing uncomfortable arithmetic about what happens if the thing works.
Those lead to different products. Here’s the map.
The three levels
“Chatbot” covers three genuinely different purchases, and buying the wrong level is the usual way this money gets wasted.
Level 1: a widget that answers from content you give it. You point it at your site, a PDF, or a help doc, it builds an index, and it answers from that. No inbox, no tickets, no team. This is most of the cheap AI chatbot market. It’s a good fit when your real problem is “the same questions over and over” and nobody is going to staff a chat queue.
Level 2: a customer-support platform. A shared inbox, ticketing, routing, human handoff, agent seats, reporting, and an AI agent sitting in front of it all. Fin and Zendesk live here, and so does most of Crisp and the upper end of Tidio. You’re buying a support operation, and the AI is one component of it. This is right when conversations need to become tickets that someone owns.
Level 3: a managed website that includes a chatbot. The bot reads the website that somebody else keeps current, so there’s no separate knowledge base to maintain. Surmado Pro+ is this. So is a full-service agency retainer with chat bolted on.
Level 1 and Level 2 are both products you operate. Level 3 is a service where operating it is the point. That distinction matters more than any feature list, for a reason worth its own section.
The actual failure mode is stale knowledge, not a bad model
Here’s the thing the vendor comparisons skip.
When a small-business chatbot gives a wrong answer, the cause is usually not the language model. It’s that the content it was trained on stopped being true. The industry’s own term for this is knowledge base rot, and practitioners are fairly blunt about it: outdated documentation, not model failure, is the common root cause. If a policy changed last quarter and the source article didn’t, the bot repeats the old policy confidently and forever.
This is a much bigger deal for a five-person business than for a company with a support team. The standard advice for keeping a bot accurate is a maintenance cadence: weekly transcript review, monthly content audits, owners assigned to articles, expiration dates on content. That’s a real job. If you’re the owner, the salesperson, and the person who answers the phone, it is the job you will silently stop doing in month three.
So the question underneath “which chatbot” is: how many copies of your business’s knowledge will exist, and who updates them?
- Point a widget at a PDF: two copies. The PDF and the website.
- Build a help center for a support platform: two copies, and now the help center is its own project.
- Point a bot at your live website: one copy, but only if someone keeps the website current.
That last condition is doing all the work, and it’s where the honest version of our own pitch lives. More on that below.
What a small business actually needs
Strip the enterprise features and the list is short:
- Answer the handful of questions that actually repeat. Hours, service area, pricing ranges, whether you do the specific thing they need. For most local businesses that’s under twenty questions.
- Say “I don’t know” instead of inventing. A bot that confidently invents a return policy is worse than no bot.
- Capture the lead when it can’t answer. The valuable outcome usually isn’t deflection, it’s a name and a phone number.
- Not go stale. See above.
- Not cost more when it works. Which brings us to the pricing shift.
Things you probably don’t need: omnichannel routing across SMS and WhatsApp and social, SLA management, workforce scheduling, multi-step agentic workflows that call your order system, or a per-agent seat model. Those are real features that real companies need. If you have three employees and no support queue, they’re overhead.
The pricing shift nobody warns you about
The chatbot market moved from flat monthly fees to usage-based pricing, and the unit is roughly “a conversation the AI handled.”
That inverts the economics. Under a flat fee, a bot that answers more questions is better value. Under per-resolution pricing, a bot that answers more questions costs more money. Success is a line item.
It gets sharper depending on what counts. Fin bills an “assumed resolution” when the customer exits the conversation without requesting further assistance, which its docs also trigger after 24 hours of disengagement, so a visitor who wanders off still bills. If they come back asking for help, the charge reverses. Zendesk splits resolutions into three tiers and bills only the LLM-verified one, which is the stricter definition, though it’s worth being precise about what it isn’t: an answer the customer never replied to can still pass verification and bill. Neither vendor requires the customer to confirm anything. Know which definition you’re buying before you sign.
Model your bill at the volume where the thing is working, not at today’s volume.
Quick comparison
| Entry price | AI pricing model | Best for | |
|---|---|---|---|
| Fin | $49/mo standalone, no seats; or $29/seat/mo annual | $0.99 per resolution, 50 included | The strongest AI agent, priced per outcome |
| Zendesk | $55/agent/mo annual (Suite Team) | Verified resolutions only, rate not published | Teams that need real ticketing |
| Tidio | $29/mo (Starter) | Lyro from $39/mo per 50 conversations | Cheapest credible entry |
| HubSpot | Free | None, rule-based only | A free widget and a CRM |
| Surmado Pro+ | $199/mo annual | Included, no metering | Not running a chatbot vendor at all |
Prices are as of August 2026 and this category re-prices often. Check the vendor’s page before budgeting.
Fin: the strongest AI agent
Fin is what Intercom is now called. The company renamed itself Fin in May 2026, and on June 15, 2026 Salesforce signed a definitive agreement to acquire it for approximately $3.6 billion. The deal is expected to close in Salesforce’s fourth fiscal quarter of 2027, so Fin operates independently for now. Factor that into a multi-year decision; product direction after a Salesforce acquisition is not something either company can promise you today.
It does four things, and nothing else here does all four: retrieves from your connected content, runs multi-step procedures against live systems, applies plain-language policy rules before answering, and escalates to a human on conditions you set. If you want a bot that does something rather than just says something, this is the one.
Pricing is outcome-based and unusually transparent: $0.99 per resolution, and only one outcome bills per conversation regardless of how many steps Fin took. Conversations simply handed to a human with no outcome aren’t billed. There is a pricier $9.99 lead-qualification outcome, but it’s currently only available through Fin for Sales on Intercom, so it isn’t something a standalone deployment stumbles into.
The genuinely useful detail: Fin can run standalone without Intercom seats, on top of Zendesk, Salesforce, HubSpot, or Freshdesk. That’s $49 a month including 50 AI resolutions, then $0.99 for each one after, with unlimited teammates and no seat fee. If you already have a helpdesk you like and just want a better AI layer, that is the option we would recommend without qualification for that buyer.
If you do want Intercom’s helpdesk underneath it, seats are $29, $85, and $132 per seat per month on annual billing ($39/$99/$139 monthly). Those seats are not required for the standalone product, which is the part people miss.
The catch is the “assumed resolution” trigger. A visitor who reads Fin’s answer and closes the tab bills the same as one who says “thanks, that solved it.” On a marketing site with high bounce, that gap matters.
Choose Fin if: you need actions and policy rules rather than just answers, you can model per-outcome costs, or you want to add AI to a helpdesk you already run.
Zendesk: if you need real ticketing
Zendesk is a support platform first. Suite Team is $55 per agent per month on annual billing, Suite Professional $115, with a support-only tier at $19. Enterprise moved to contact-sales.
Two things changed in 2026 and both favor buyers. Zendesk removed the Essential/Advanced AI split in May, so agentic reasoning, multi-step procedures, and external API calls are now included across Suite and Support plans rather than sold as an upgrade. And the resolution-billing change described above means you’re only charged for resolutions its verification step actually confirms.
The real catch is budgetability. Zendesk does not publish a universal per-resolution rate. Its own docs describe pay-as-you-go and added capacity, then point you at Sales or your account terms for the number. That isn’t the same as “no overage.” You can absolutely spend more than your plan. It means you can’t work out in advance what more will cost, and a small account has no leverage in the call where that number gets set.
The allowance is also denominated in dollars per agent per month rather than a resolution count: $2 on Suite Team and Support plans, $5 on Growth and Professional, $10 on Enterprise. How many resolutions that buys depends on the rate you negotiated, which is the circular part. Copilot, the agent-assist layer, is $50 per agent per month on annual billing and is available across Suite plans rather than gated at Professional.
Choose Zendesk if: conversations need to become tickets somebody owns, and you have enough volume to negotiate.
Tidio: cheapest credible entry
Tidio is the SMB-shaped option. Starter is $29/mo and Growth starts at $59, with annual billing bringing those to about $24 and $49. Above that it climbs steeply: Plus starts around $300/mo plus usage, and Premium is quote-only.
Its AI agent, Lyro, is priced separately from the base plan, starting at $39/mo for 50 Lyro conversations ($32.50 on annual billing). Higher volumes are configured dynamically rather than published as a fixed ladder, so you’ll need a quote to model anything larger. The real entry price for AI is Starter plus Lyro, not the $29 headline.
The detail to read carefully: on the Free and Starter plans, the 50 included Lyro conversations are a lifetime total, not a monthly allowance. Once spent, Lyro stops answering until you buy paid capacity. That is an unusual structure and it catches people who assumed the number refilled. Tidio also meters three things separately: billable human-agent conversations, Lyro conversations, and unique visitors reached with Flows. Three quotas is harder to forecast than one.
Choose Tidio if: you want low commitment and predictable low volume, and you’ve checked which of the three quotas you’ll hit first.
HubSpot: the honest free option
HubSpot gives away live chat and a chatbot builder on its free tier, with no conversation limit, alongside a genuinely useful free CRM.
Be clear about what it is: on Free and Starter, the chatbot is rule-based, not AI, and on Free it’s narrower than “rule-based” suggests. HubSpot’s own catalog limits the free tier to predefined ticket, qualification, and booking bots with no custom branching logic. It collects information and hands off to a human; it does not read your content and reason about a question. The widget carries HubSpot branding, and Free Tools cap the whole account at two users. “Unlimited conversations” is real, but it describes volume, not seats.
The actual AI agent, Breeze Customer Agent, requires a Professional or Enterprise Hub subscription and consumes 50 HubSpot Credits per resolved text conversation, with extra credits at $0.010 each.
That’s not a criticism. A bot that routes people correctly and captures a lead beats a bad AI bot, and it costs nothing. For a business whose repeat questions are genuinely simple, this is a real answer.
Choose HubSpot if: your budget is zero, your questions branch predictably, and you’d use the CRM anyway.
Surmado Pro+: when you don’t want a chatbot vendor
Every option above is a vendor you add and a knowledge source you maintain. That’s the difference here, and it cuts both ways.
Surmado Pro+ is $199 a month billed annually, or $250 month to month. It’s a managed website: we migrate your existing site, host it, maintain it, publish weekly posts, handle tracking and consent, and include a custom chatbot. Scout, our AI agent, answers from the indexed content of the site it already maintains, cites the source page, and when the site doesn’t have the answer it says so and offers to collect contact details for follow-up. Chat can answer in any language the site is published in, so a Spanish visitor gets Spanish. It’s included with Pro+ and isn’t sold separately.
The structural argument is the one from earlier: there’s no second copy. The bot’s knowledge is the website, and the website is maintained as part of the service. Knowledge base rot is a problem created by having two systems and one busy owner. Removing the second system removes the failure mode rather than scheduling a monthly audit to catch it.
Here is exactly where we are the wrong choice. Surmado is not a customer-support platform. There’s no shared inbox, no ticketing, no agent seats, no SLA management, no routing rules, and no omnichannel presence across SMS, WhatsApp, and social. Our chat also answers strictly from what’s published on the site: it will not know your internal return policy unless that policy is on a page, and extending what it knows means adding content to the site, not uploading a separate document.
So if you’re fielding enough conversations that they need to become tickets with an owner, or your bot has to look up an order and issue a refund, or you need it answering on WhatsApp, buy Fin or Zendesk. Those are real requirements and we don’t meet them. What we won’t do is pretend that list describes a plumber in Phoenix whose site gets forty questions a month about hours and service area.
One clarification, since “agent” is doing double duty in this article: Scout is the agent that runs the website, and chat is one surface of it. The products above are support tools. Nothing here compares Scout against Fin at managing a website, because managing websites isn’t something Fin does.
And the commitment is genuinely larger. Every other product here is a widget you install in an afternoon on the website you already have. This one requires moving your website to us. That’s the right trade only if the website itself is something you’d rather hand off, which is a much bigger question than “should I add chat.”
Choose Surmado if: the chatbot is one item on a longer list of website responsibilities you no longer want to own.
How to pick
- “I already have a helpdesk and want better AI on it.” Fin standalone. Genuinely the best answer for that buyer.
- “Conversations need to become tickets someone owns.” Zendesk.
- “I want AI chat, cheaply, at low volume.” Tidio, after checking the quota structure.
- “My budget is zero and my questions are simple.” HubSpot free.
- “My website is also a problem I’m tired of owning.” Surmado.
Two things worth doing before any of them.
Write down the questions you actually get. If the list is twelve items, a well-structured FAQ page might solve most of it for free, and it’ll help your search and AI visibility at the same time. Not every problem needs software.
Model the bill at success volume. Under usage pricing, the scenario where the bot works well is the expensive scenario. Run the arithmetic there, not at today’s traffic.
The bottom line
These are good products, and they win in specific places. Fin resolves more when the job is a high-volume support queue and the bot needs to take actions in your other systems. Zendesk wins when conversations have to become tickets somebody owns. HubSpot wins when the budget is zero. Those are the conditions, and if you’re in one of them, go buy that product.
What none of them wins is the thing this article is actually about: whether you wanted a second copy of your business’s knowledge. Every option except the last one leaves you owning one, and the reason chatbots go stale is that the second copy has an owner who is busy. No model improvement fixes that, because it isn’t a model problem.
Surmado’s answer is to not have the second copy. It costs more than a widget and it only makes sense if you were ready to hand off the website too. If you were, Surmado Sites rebuilds your existing site and sends you a preview before you pay anything. If you’re weighing the broader question of what a managed site includes, website as a service and what an AI webmaster does go deeper, and what a small-business website actually costs covers the budget side.
Pricing and features are as of August 2026 and change without notice; verify on each vendor’s own page before deciding.