Everyone has met the bad one. Seven menu options, none of them your problem. "I didn't quite understand that", three times in a row. And no visible way to reach a human, because the bot's real job was to keep you away from one.
The word chatbot carries baggage because for a decade the products deserved it. So the honest question is not "are chatbots good now?" It is: what actually changed, what did not, and does your business need one at all?
Why chatbots earned the reputation
The old generation were keyword matchers wired to decision trees. They failed in two characteristic ways. First, the dead end: any question outside the tree got a shrug on loop. Second, the human blockade: many were deployed explicitly to reduce contact with staff, and their success metric was "deflection rate", which is a polite way of counting customers who gave up.
Customers noticed. That is the baggage.
What modern AI actually changed
A current AI assistant reads the actual sentence instead of hunting for keywords. It copes with typos, phrasing it has never seen, and questions asked in any language. Grounded in your own content, it answers with your real prices, your real policies, your real opening hours. And when the answer is not in its sources, it can say so and hand over to a person.
That last part deserves emphasis: the dead end is no longer a technical limitation. If a modern bot traps customers, someone chose to configure it that way.
What did not change
- Garbage in, garbage out. An AI grounded in a thin, outdated help center produces confident nonsense or escalates constantly. The unglamorous prerequisite for any bot is a knowledge base worth answering from.
- Hard cases still need people. Angry customers, ambiguous situations, anything with legal or financial stakes: an instant automated reply proves the opposite of what those moments require.
- The escape hatch is non-negotiable. A human must be one message away, always, visibly. Every design that hides the option re-earns the old reputation on your behalf.
When a website chatbot genuinely helps
There is a real sweet spot: a site with steady traffic and repetitive pre-sales questions. Opening hours, shipping costs, plan differences, stock. A grounded bot answers those instantly, at 11 pm, in the visitor's language, and captures the question that would otherwise have left with the visitor. Judge it by resolution and customer satisfaction, never by deflection, and the incentive to trap people disappears.
The question behind the question
Here is what the chatbot debate usually misses: the widget is the least important part. The capability that matters is grounded, instant, multilingual answers. The widget is just one door it can stand behind, and for many small businesses it is the wrong first door, because the questions do not arrive on the website. They arrive in email, WhatsApp, Instagram and Messenger.
That is the premise FirstReply is built on. Instead of a bot guarding one channel, the same AI works across all of them: it categorizes every incoming conversation, drafts a grounded reply in the customer's language, and either a human clicks send or, for well-proven categories, the answer goes out automatically. Whether the customer wrote in a chat widget or an email stops mattering; the quality of the answer is the same.
The checklist before you switch anything on
- Ground it exclusively in your own content. General knowledge is where invented refund policies come from.
- Give it permission to say "I don't know". A handover beats a hallucination, every single time.
- Keep a human one message away. Visibly, unconditionally.
- Measure resolution, reopens and satisfaction. If you measure deflection, you will build a machine for giving up.
- Start with drafts if you are unsure. A bot that proposes answers for human review earns trust category by category before it speaks alone.
The right question was never "chatbot or no chatbot". It is: where should routine questions answer themselves, so that your people are free for the conversations that actually need them?