AI assistant
for business
An AI assistant for business reads questions however customers phrase them and replies using only material you approve: service pages, price lists, policies, class schedules. A generic AI chat will talk about anything; this one has a written brief on what it may discuss, what it must never promise and when a person takes over.
Free: a 30-minute call, an action plan and a price range in writing.
Five guardrails that make
AI safe to show customers
A defined knowledge base
The assistant answers from documents you approve, not from whatever it picked up on the internet.
Topics it declines
Medical, legal or financial advice, competitors, anything off-topic: it politely steers back or hands off.
"I don't know" is allowed
When the answer is not in your material, it says so and offers a person, instead of guessing.
Prices only from your list
No invented quotes or deals. Prices come from the approved list, with the conditions you set.
Conversation logs
Every chat is saved so you can see what was asked and how the assistant answered.
From documents to a live
AI assistant in 5 steps
I deliver AI assistants with a trusted partner; the contract, deadlines and communication stay with me.
Collect the material
Pages, price lists, policies and real customer questions from your inbox, cleaned up where they disagree.
Write the rules
Tone, allowed topics, forbidden promises, hand-off triggers and what details to collect.
Build and connect
The assistant placed on your website or messaging channel and linked to email or CRM.
Stress test
Hundreds of real and tricky questions, including attempts to get it off track, before launch.
Review and improve
Logs reviewed in the first weeks; content and rules tightened where answers fell short.
What an AI
assistant costs
Prices are final and in US dollars. Usage of the AI model is charged to your own account and grows with the number of conversations.
A single assistant on your main channel, built on your approved content.
- Knowledge base prepared
- Rules and boundaries written
- Hand-off to your team
- Leads delivered to email or CRM
- Log reviews after launch
An hour to see whether AI fits your business before you spend more.
- Use case reviewed with you
- A straight answer
What an AI assistant for business can and cannot do
An AI assistant for business is a chatbot that understands natural language. Customers do not have to pick from buttons; they type "can I bring my dog to the patio on Saturday?" or "do you take my insurance for a cleaning?" and get a sensible answer. It shines where questions are varied and the answers already exist somewhere in your material: schools with many courses, clinics with many services, stores with large catalogs, hotels and rentals with long policy pages.
What it should not do: give legal, medical or financial advice, make up prices, promise outcomes, or replace a person in sensitive conversations. A well-built assistant knows its limits and hands off gracefully.
How it learns your business
The assistant does not learn from the open internet or from other businesses. I build a knowledge base from your website pages, price lists, policies, FAQ documents and, most valuable of all, real questions from your inbox with your answers to them. Before that, I clean up contradictions: if the website says one cancellation policy and the booking confirmation says another, the assistant will repeat the confusion. Fixing the source content is often a useful side effect of the project.
When something changes, such as a new price or summer hours, you update the knowledge base and the assistant uses the new information right away. No retraining is needed.
Guardrails I put in place
- Scope. The assistant answers only about your business and declines everything else politely.
- Fixed facts. Prices, hours and policies come from defined fields, not from the model's paraphrase.
- Escalation. Complaints, emergencies, refunds and anything emotional go to a person, with the conversation attached.
- Disclosure. It introduces itself as an automated assistant. Several states, California among them, have rules on bots that interact with consumers, and honesty is simply good service.
- Data minimization. It asks only for the details needed to follow up and warns against sharing sensitive information.
Where it lives
Most often on your website, as a website chatbot that understands free text. It can also run on Messenger, Instagram or WhatsApp. For businesses with lots of post-purchase questions, the customer service chatbot page shows how an assistant connects to order or booking data. Leads it collects go to your CRM with the conversation attached.
Privacy and US rules
Conversations are processed by the AI provider, so your privacy policy should say so, and the assistant should not collect more than it needs. If a state privacy law applies to your business, chat data falls under the same rights to access and delete as any other personal data. Healthcare providers should not let a general-purpose assistant handle patient health information without a HIPAA review and a business associate agreement with the vendor; talk to your compliance advisor first.
There is a separate, growing topic of how AI search tools like ChatGPT describe your business to people who never visit your site. That is not what this assistant does, but if it interests you, see my page on visibility in AI search.
Costs and delivery
The setup is $1,900 for one assistant on one main channel, delivered with a trusted partner. I run the project: content, rules, testing, connections and launch. The contract, the deadlines and the communication are with me. The AI model's usage is billed to your account and, for a typical small business, stays modest. My article on AI chatbot cost breaks down the monthly side in more detail. If you want to talk it through before deciding, a consultation is the easiest start.
Worth reading
next
Questions about
AI assistants
Will the AI assistant make things up?
Any AI model can produce a wrong answer, which is why the setup matters. The assistant is limited to your approved material, instructed to say when it does not know and tested on hundreds of questions before launch. Anything involving prices, availability or commitments follows fixed rules. The risk does not disappear completely, so logs are reviewed regularly.
Am I responsible for what the assistant says?
In practice, customers and regulators are likely to treat the assistant's answers as coming from your business, much like a page on your website. That is why it should never promise what you would not, and why I keep it away from legal, medical and financial advice. For regulated industries, check the setup with your attorney or compliance advisor.
Is customer data sent to the AI provider?
The text of each conversation is processed by the AI model to generate answers. I use business-grade services whose terms say customer data is not used to train their public models, and I tell customers in the chat not to share sensitive details. Your privacy policy should mention the assistant, and I help you word that part.
Which AI model do you use?
It depends on the job, the budget and the channel. I pick a well-established model from a major provider, test it on your questions and can switch later if a better or cheaper option appears. You are not locked into one vendor, and the knowledge base stays yours.
Can it answer in other languages?
Modern AI models can understand and answer in many languages. I write and check the rules and key answers in English, Polish, Russian and Ukrainian. For Spanish, the model can reply, but I do not review or write Spanish content myself, so a fluent person on your side should check those answers before launch.
Try it with questions
your customers really ask
Pick twenty customer questions from last month and send them over. On a free 30-minute call you will see how the assistant handles them and where a person should step in.