AI for Business · 4 min read
Your website probably answers every question a visitor has. It just makes them hunt for it across six pages and a FAQ. Here is what changes when they can ask instead, and the one thing to check before you put a chatbot on your own site.
Every business website I look at has the same problem, and it isn’t a design problem.
The information is all there. Hours, service areas, pricing structure, who to talk to about what, the thing you do that your competitors don’t. It’s spread across six pages, a FAQ nobody scrolls past the third question, and a PDF someone uploaded two years ago.
Then a visitor lands on your homepage with one specific question. Their options are to hunt for the answer, fill out a contact form and wait until tomorrow, or leave.
Most of them leave. You never find out what they wanted to know.
The gap isn’t information, it’s retrieval
Here’s the part that took me a while to say out loud: your website is organized the way you think about your business. Visitors arrive thinking about their own problem.
You have a services page. They have a Tuesday afternoon and a question about whether you work with people like them.
Those aren’t the same shape, and no amount of navigation redesign closes that gap completely. Menus force people to guess which category their question lives in before they can ask it.
A chatbot that actually knows your business skips the guessing. Someone types the question they came with, in the words they’d use out loud, and gets the answer that was already sitting on page four.
Go break one
I built a demo so you can see what I mean instead of taking my word for it.
Riverstone Realty is a fictional real estate agency. The site is small and ordinary on purpose: some listings, some agents, a FAQ. The chatbot in the corner has been given the agency’s actual information, the same content a visitor could find by clicking around.
Try it. A few things worth asking:
- Ask for three-bedroom homes under a certain price. It’ll name the ones that qualify, with the real numbers.
- Ask who you should talk to about selling. It’ll point you at the right person and tell you why.
- Then try to break it. Ask about something the agency doesn’t do. Ask about commercial warehouses in another state. Ask it to make something up.
That last one matters more than the first two.
What separates a useful chatbot from a liability
A general purpose chatbot will answer anything you ask, confidently, whether or not it knows. For a business, that isn’t a feature. That’s your website telling a customer something untrue, in your voice, with your logo above it.
The Riverstone bot will tell you it doesn’t have that information and hand you to a human. It does that because it was built to answer from the agency’s own material and nothing else. When the answer isn’t there, saying so is the correct output.
That’s a design decision, not an accident, and it’s the one I care most about. A chatbot that invents a price is worse than no chatbot. A chatbot that invents a policy can get you sued.
So when you evaluate one of these, for your own business or from any vendor, the question isn’t how smart it sounds. The question is what it does when it doesn’t know.
What this is actually good for
This isn’t a support ticket deflection scheme, and I’m not going to pretend it replaces anyone on your team.
What it does well is the first conversation. The questions people ask before they’re willing to fill out a form. Do you work in my area, do you handle situations like mine, what does the process look like, who would I even be talking to. Those questions are cheap for you to answer and expensive for a visitor to go find.
It also tells you something analytics can’t. Analytics say a visitor spent ninety seconds on your services page. A chat log says they asked three times whether you work with first-time buyers and couldn’t tell from your site. That’s a content problem you now know about.
And for businesses where the information shouldn’t leave the building, this can be built to run entirely on hardware you control, with nothing sent to an outside service. That isn’t the right fit for everyone, but when it matters, it matters a lot.
The same pattern, different content
A real estate site is a useful demo because listings, agents, and neighborhoods are concrete. You can ask specific questions and get specific answers. But the underlying capability, a chatbot that only speaks from your organization’s own information and knows when to say it doesn’t know, travels well.
A few places the same pattern fits:
A community-focused nonprofit with a funder-facing knowledge hub. Program officers at foundations Google around, land on the site, and want to know if a grant idea is a fit before they email anyone. The bot answers from published program descriptions, past awards, and eligibility criteria. When it doesn’t have the answer, it points them at the right person to email.
A school district or education organization with a family-facing Q&A. Enrollment windows, transportation, immunization requirements, calendar. Parents ask in the language they use at the kitchen table, not the language that’s in the district handbook. For anything touching student data, the bot can live on hardware the district controls, not a third-party service.
A small professional services firm. Law, medical, accounting, financial planning. The first question a prospective client has is “do you handle situations like mine.” The bot describes practice areas honestly and hands anything close to case-specific to a human. It doesn’t give advice. It clears the intake filter for both sides.
A local government or civic organization with a “where do I go for X” problem. Permits, services, office hours, forms. The bot walks someone through what applies to their situation without pretending to be a legal or policy authority. When the answer requires human judgment, it says so.
The through-line: each of these has good information that visitors can’t easily find, and each has a real cost when the bot invents something. So the bot is trained to be grounded, honest about limits, and quick to say “you need a human for this.”
If you want one
The build depends on what your business actually needs, which is why I don’t publish a package or a price. A real estate agency with forty listings and a law firm with a compliance review process need different things, and pretending otherwise produces the kind of chatbot that ends up embarrassing you.
Go play with the Riverstone demo first. Ask it hard questions. Notice what it does and doesn’t do.
Then tell me what you’re trying to move, and we’ll figure out whether this is the right tool for it.