Every chatbot vendor will tell you that a website chatbot is a must-have. That it increases conversion by 30%. That customers love it. That it pays for itself.
And I'll tell you the truth. Because I've tested chatbots for over a dozen clients at kaminski.link. And the results? Mixed. Very mixed.
When a chatbot makes sense
Let's start with the positives. Because chatbots can work. But under specific conditions.
First scenario: high-traffic e-commerce. An online store with 10,000+ monthly visits selling products that need guidance. Electronics, supplements, specialized tools. An AI chatbot that knows the product catalog, can compare models, and suggest choices. For one client (electronics store), the chatbot generated 12% additional conversions. But that store had 40,000 monthly visits. On lower traffic, that 12% would be 3 sales. Not worth it.
Second scenario: companies with lots of repetitive questions. Law firm, medical clinic, insurance company. Customers ask about opening hours, service scope, initial consultation prices. If you get a lot of such questions by email, it's also worth considering AI-powered email automation. The chatbot responds in 3 seconds instead of 3 hours (because the receptionist is busy). Here the chatbot doesn't so much increase conversion as decrease bounce rate. People don't leave the site because they get an answer immediately.
Third scenario: SaaS and digital products. Onboarding, FAQ, troubleshooting. A chatbot that knows the documentation better than support. ROI is highest here, because every support ticket costs $4-7. 100 fewer tickets = $400-700 in monthly savings.
When a chatbot is a waste of money
Now the anti-examples. And there are more of these than I'd like to admit.
Small service website with 500 monthly visits. We deployed a chatbot. In a month, it talked to 8 people. Four of them were testing if it works, typing "hello" and "test." Two asked about something that was on the page two scrolls down. Two asked a reasonable question. Chatbot cost: $50/month. Value of those two conversations? Maybe $12. We shut it down after 3 months.
Restaurant. Client wanted a chatbot for reservations. Sounds reasonable, right? Problem: people booking a table want to do it in 10 seconds. The chatbot carries on a conversation: "When for? How many people? Do you have table preferences?" A reservation form does the same thing in 3 clicks. The chatbot lost to the form. And it always will when the process is linear and simple.
B2B with a long sales cycle. A consulting firm deployed a chatbot to "qualify leads." The chatbot asked about budget, project scope, timeline. Nobody wanted to discuss their $50,000 project budget with a bot. Zero conversions from the chatbot. People in B2B care about relationships, not scripts. A good CRM and systematic follow-up work much better here.
How much it actually costs
Let me break down costs, because chatbot vendors love hiding them behind "contact us":
- Off-the-shelf solutions (Tidio, Intercom, Drift): $25-120/month depending on plan and conversation volume
- Custom AI chatbot: $750-2,000 one-time + $75-200/month (API + maintenance)
- Enterprise (Dialogflow, Watson): $1,200-5,000 implementation + $250-750/month
Important note here. Off-the-shelf solutions like Tidio are fine to start. But their AI is generic. It doesn't know your business. It answers in generalities. If you want a chatbot that actually helps - you need a custom implementation with your knowledge base. And that costs money.
How to do it right
If you decide a chatbot makes sense for you, here's my advice after dozens of implementations:
- Always offer a "talk to a human" option. A chatbot you can't escape from annoys people
- Don't pretend the bot is human. Write clearly: "I'm an AI assistant. I can help with..." People appreciate honesty
- Limit the scope. A chatbot that tries to answer everything answers nothing. Better 20 topics done well than 200 done poorly
- Measure conversions, not conversations. 1,000 chatbot conversations mean nothing if none ended in a sale
- Test on real traffic for at least 30 days before judging. A week isn't enough
One more thing. A chatbot isn't a one-time implementation. The knowledge base needs updates. Conversational flows need optimization. New customer questions need new answers. If you deploy a chatbot and forget about it for six months - it'll give nonsense answers and hurt your brand.
My verdict for 2026
AI chatbots in 2026 are significantly better than those from 2 years ago. GPT-4, Claude - these models truly understand context and can carry on a meaningful conversation. Technology isn't the problem.
The problem is fit. A chatbot is a tool. Like a hammer. Great for driving nails. Terrible for turning screws.
So before you spend $1,200 on implementation, ask yourself one question: do my customers actually want to talk to a bot, or would they prefer a form, phone call, or email? The answer to that question is worth more than any chatbot on the market.
Don't deploy a chatbot because "everyone has one." Deploy it because you have a specific problem that a chatbot solves better than the alternatives. If you can't point to such a problem - you don't need a chatbot. And that's OK.