Should your business put an AI assistant on its website?
We have built five of them for five different businesses: a transit company, a custom home builder, a landscaper, a painting company and a financial advisor. Each one has a name, answers from the company’s own site, and files leads into that company’s CRM. Here is what actually matters.
A website assistant pays for itself when it does three things: answers questions from your own content instead of making things up, qualifies the visitor with a few taps, and books time or files the lead. It fails when it is a generic chatbot with no connection to your business.
Give it a name and a job
A box that says "Chat with us" gets ignored. An assistant called Drew that opens with "Looking at a renovation or a new build?" starts conversations. The name matters less than the opener, which should match the page the visitor is on. Someone reading a kitchen page should be asked about kitchens.
It should only know what you know
The single biggest risk with an AI assistant is that it invents an answer. Ours are trained only on the live website and a short set of facts the owner approves, and they are retrained every time the site is published. If the answer is not on the site, the assistant says so and offers to have someone call. It never quotes a price unless the owner has published one.
Qualify with taps, not a questionnaire
People will tap a button far more readily than they will type. "Is this for a campus, a hotel or a community?" with three buttons gets an answer almost every time. Two or three taps give the owner enough to know whether the lead is worth a call.
Turn the price question into a callback
Every assistant gets asked "how much". The right move is not to dodge it and not to guess. It is to say "that depends on a couple of things, want someone to call you with a real number?" and collect the phone number. The price question is the best lead signal on the site.
Book the time, file the lead, tell the owner
A good assistant ends a conversation by booking a slot on the owner’s calendar or handing off a complete lead to the CRM: name, contact, what they asked, what they tapped. The owner gets a recap every morning of every conversation from the day before, including the ones that did not convert, which are often the most useful because they show what people are asking that the site does not answer.
Where it does not belong
If your business depends on a careful first conversation (some legal matters, some medical practices), the assistant should collect facts and book the call, not advise. We also keep assistants off pages meant for existing customers who need a human, and we never let one pretend to be a person if someone asks.
The honest math
An assistant costs a fraction of a part time receptionist and works at 11pm. On a site with any real traffic it only has to produce one extra job a month to pay for itself. The ones we run do more than that, mostly because they catch the visitor at the exact moment of interest instead of the next morning.
Questions people ask
Will the AI assistant make things up about my business?
Not if it is built correctly. Ours answer only from your live website and an approved fact list, say when they do not know, and never quote prices the owner has not published.
Does it replace my contact form?
No, it sits beside it. Some visitors prefer a form, some prefer a conversation. Both land in the same CRM on the same timeline.
Can I read the conversations?
Yes. Every conversation is logged in an admin page inside your CRM, and you get a morning recap email. Unanswered questions are the best source of new content for the site.
Want this done for your business?
Two minute intake. A real person reads every one and replies within a business day.
Keep reading
What an AI assistant on a website actually does all day
What a website AI assistant does all day for a Florida service business: page matched openers, tap qualifying, price callbacks, CRM leads and a morning recap.
AI at workHow we pick an AI model for a client project
How a Delray Beach studio picks an AI model for a client project: the job, speed versus depth, cost, privacy, and why one product often uses several models.
AI at workOpenAI, Anthropic, Google and the rest: what each is for when you are building
A builder's map of OpenAI, Anthropic, Google, Meta, xAI, DeepSeek and Mistral: what each is known for, hosted versus open models, and how we choose per job.