On this page
The chatbot you remember from 2018 — a decision tree that offered four buttons and could not answer any of them — deserved its reputation. What is deployed now is a different thing: a system that reads your actual documentation and answers questions in your customer's own words.
That change matters commercially, because the failure mode has flipped. The old question was "will it be able to answer anything". The new question is "will it answer confidently when it should not" — and that is a design problem you can solve.
The short version
A modern chatbot is worth deploying if you have (a) real content it can learn from, and (b) a repeating set of questions. Without content it will guess. Without repetition it will not save you anything.
What a modern AI chatbot actually does
- Answers from your own material — pricing pages, policies, product docs, past support replies — rather than from a generic model's guesswork.
- Holds context across a conversation, so "what about the larger one?" makes sense as a third message.
- Captures leads naturally, asking for an email at the moment it is useful rather than blocking the conversation behind a form.
- Escalates to a human with the full conversation attached, so the customer does not have to start again.
- Works in more than one place — website, WhatsApp, Messenger, in-app — from one knowledge base.
- Tells you what people are asking, which is frequently the most valuable output. Your chat logs are the cheapest customer research you will ever get.
Where a chatbot beats the alternatives
The honest framing is that a chatbot does not replace live chat; it removes the 60–70% of conversations that never needed a person, so the people you do employ spend their time on the ones that did.
What it costs
Chatbots are the cheapest entry point into AI for most businesses — meaningfully cheaper than voice, because there is no telephony, no real-time latency budget, and text is inherently more forgiving than speech.
As with voice agents, the price has three parts: a one-time build (ingesting your content, designing the escalation rules, styling and installing it), a monthly plan, and usage. TensoraAI's chatbot tiers are published on the pricing page. Self-serve platforms exist and are genuinely usable if you are willing to do the content work yourself. For a full breakdown of setup fees, subscriptions and AI usage charges — including current pricing from the major platforms — see our AI chatbot pricing guide.
Where the real cost hides
Not in the software. In getting your own content into a state the bot can learn from. If your pricing lives in three contradictory places and your refund policy is in someone's head, that is the project — and it is worth doing whether or not you ever deploy a bot.
Five deployments that consistently work
Pre-sales questions on a pricing page
The highest-intent moment on most websites. Someone is looking at your prices and has one blocking question. Answering it in ten seconds instead of never is the whole game.
Tier-one support deflection
Password resets, order status, how-do-I questions, policy lookups. High volume, low variance, and entirely documented already.
Booking and availability
Combined with a calendar integration, chat becomes a booking channel — often preferred by people who do not want to make a phone call. See AI appointment booking.
WhatsApp for service businesses
In many markets WhatsApp is where customers already are. The same knowledge base, a different surface.
Internal helpdesk
Overlooked and often the easiest win: a bot that answers your own team's questions about process, policy and where-do-I-find. No customer risk, immediate time saved.
How to deploy one without annoying people
Do not auto-open on arrival
A bubble that expands over the content two seconds after landing is the single most common complaint. Let people open it.
Say what it is
"AI assistant — I can answer questions about pricing, delivery and returns." Setting the scope up front prevents most bad interactions.
Make the human handoff obvious
One click, always visible, no interrogation first. A customer who wants a person and cannot find one will leave.
Ground it in your content
The bot should answer from your material and say "I don't know, let me get someone" when it is not there. Confident wrong answers are worse than no bot.
Read the transcripts weekly
For the first month, all of them. The gaps in your own documentation will be embarrassingly obvious and cheap to fix.
A chatbot is a mirror held up to your documentation. If the answers are not there, it will make them up — and that is a content problem, not a model problem.
Limitations worth knowing
- It can be confidently wrong. Grounding it in your content reduces this substantially but does not eliminate it. Never let it be the final word on money, contracts or safety.
- It will not fix a bad product. Faster answers to "why is this broken" do not make it less broken.
- Multilingual support needs testing per language, not assuming.
- Accessibility is easy to get wrong. Chat widgets are frequently keyboard traps. Test with a keyboard only before launch.
- Consent and data. Conversations are personal data. Retention, storage location and what you do with them all need an answer before launch, not after.
Chatbot or voice agent — which first?
Follow your customers. If most enquiries arrive by phone, start with a voice agent; if they arrive through your website, start with chat. Businesses that get both tend to build the knowledge base once and point both channels at it, which is the right architecture regardless of which you start with.
Frequently asked questions
How long does it take to build an AI chatbot?
A bot grounded in existing website content can be live in days. The timeline extends when it needs to look things up in live systems — order status, account details, availability — because that is integration work rather than content work.
Will it hallucinate answers?
The risk is real and is managed rather than eliminated. Grounding responses in your own documents, constraining the bot's scope explicitly, and instructing it to escalate rather than guess reduce it substantially. Testing against your hardest real questions before launch is the part people skip.
Can it hand over to a real person?
Yes, and it should. The handover must carry the full conversation so the customer never repeats themselves, and it must be available from the first message rather than only after the bot has failed twice.
Does a chatbot help SEO?
Not directly — chat content is not indexed. Indirectly it can help, because chat logs reveal the exact questions your customers ask, which is excellent raw material for pages that do rank.
What content do I need to provide?
Anything a new employee would need on their first week: pricing, services, policies, delivery, returns, hours, common troubleshooting. If it is already written down somewhere, it can usually be ingested as-is.
You can see how we approach this on the AI chatbots service page, or start with a free AI audit to work out whether chat, voice or back-office automation is the right first move.
