A website AI chatbot that answers from your own content.
A website AI chatbot answers visitors’ questions using only the pages and documents you approve, collects contact details when someone is ready to talk, and passes the conversation to a person when it cannot help. Its quality depends on your content, its limits and its handoff rules.

What can a website AI chatbot handle?
Visitors arrive with specific questions: do you cover my area, how long does it take, what does it include, can I talk to someone. Your website probably answers most of them already, spread across a dozen pages. A chatbot finds the right passage, answers in a sentence or two and links to the page it used.
- Product and service questions, answered from your published pages, brochures, policies and FAQs.
- Pre-sales questions that decide whether someone gets in touch: coverage, lead times, what is and is not included.
- Enquiry capture, asking for the few details your team needs and sending them to your CRM or inbox.
- Signposting, sending people to the right page, form or booking link rather than repeating it.
- Handoff, to a person in real time where you staff the chat, or to a call-back with the conversation attached.
What it should not do is improvise. Quotes for custom work, complaints, account-specific questions and anything that depends on a person’s judgement belong with your team.
Is AI chatbot a job for an agent?
Tick what is true of your ai chatbot work.
Tick the statements that apply and the suggestion appears here: an AI agent, ordinary automation, or a person with software support.
How does it answer from your own content?
The usual approach is retrieval: the chatbot searches an index of your approved content, passes the most relevant passages to a language model and instructs it to answer from those passages only. Nothing is “trained” in the sense of changing the model; your content is looked up at the moment of each question.
- Choose the sources: website pages, PDFs, policy documents, a product table. Leave out drafts, internal notes and anything you would not publish.
- Split them into passages and index them, keeping the page address so every answer can link back.
- For each question, retrieve the closest passages and ask the model to answer from them, citing the source.
- If nothing relevant is found, the chatbot says so and offers a person, instead of filling the gap with a plausible guess.
- Re-index when content changes, so a price or policy update reaches the chatbot the same day.
This is why a chatbot is often a content project in disguise. If two pages contradict each other, the chatbot will find both. Fixing the source is better than adding special rules.
Do you need AI, a rules-based bot or a better FAQ page?
| Option | When it fits |
|---|---|
| A clear FAQ page and contact form | You get a modest number of enquiries and most questions repeat. Often the right first step, and it helps search engines too. |
| A rules-based chat widget | A few fixed paths: book, get a quote, talk to sales. Predictable, cheap to run and easy to check. |
| An AI chatbot on your content | Visitors ask varied questions in their own words, your content is large or detailed, and answering quickly affects whether they get in touch. |
| An AI agent | The chat also needs to act: check availability, look up an order, create a booking. See [chatbot vs AI agent](/guides/chatbot-vs-ai-agent/). |
Many good builds combine them: fixed buttons for the common paths and an AI answer for everything typed freely.
How much time could AI chatbot give back?
Put your own numbers in.
The starting values are an illustrative example, not a benchmark or a neu figure. Use your own volume, the time each one takes today and the share you expect the system to handle without a person.
Time saved is only part of the case: response times, missed work and errors matter too. The ROI guide covers how to set a baseline, and the cost guide what the build and running costs include.
How should it capture leads and hand off?
Ask for details when the visitor is ready, not as a gate before the first answer. Collect only what your team needs to respond: usually a name, a way to reach them and a sentence about the job. The chatbot can then create a contact and a note in HubSpot or Pipedrive, or pass it into a lead qualification flow.
Decide how handoff works before launch. If someone staffs the chat during office hours, the bot should hand over the live conversation with its history. Out of hours, it should say when a person will reply and mean it. A handoff that quietly lands in an unmonitored inbox does more damage than no chatbot.
Chat transcripts are personal data once they contain a name or contact details. Update your privacy notice, set a retention period and check where the chatbot provider processes data. The GDPR and AI agents guide covers the questions to ask.
What can go wrong with a website chatbot?
- Confident wrong answers, when the model fills a gap in your content. Instruct it to answer only from retrieved passages and test it on questions your content does not cover.
- Commitments you did not make: a discount, a delivery date, a refund. State in its instructions what it may never promise, and test that too.
- Manipulation: visitors typing instructions to make it say something embarrassing. Keep it away from internal data it does not need and review odd conversations. See AI agent security.
- Stale content, when a price changes on the website but not in the index.
How do you know it is working?
Read the conversations, especially in the first month. Then track a small set of numbers: questions answered without handoff (checked on a sample, not taken on trust), enquiries captured, handoffs and how quickly they were picked up, and questions the chatbot could not answer. That last list is your content backlog.
Running costs depend mostly on conversation volume and the model used; the AI agent cost guide separates the build from monthly usage. When the chat needs to book, look things up or update records, it becomes an AI agent project, the same step our tutoring operations agent took from answering enquiries to booking trial lessons.
Common questions
Will an AI chatbot replace my contact form?
Usually not. Some visitors prefer a form, and a form is still the most reliable way to collect a structured enquiry. Keep both, and let the chatbot offer the form or collect the same fields itself. Removing the form forces everyone through a chat, which suits some people and puts others off.
Can the chatbot answer in Irish or other languages?
Current language models handle many languages, and a chatbot can reply in the visitor’s language while drawing on English source content. Quality varies by language, so test with native speakers before relying on it. If you serve customers in a language regularly, consider translating the key source pages rather than relying on translation on the fly.
How much content does it need to be useful?
Less than people expect, as long as it is accurate and specific. A dozen well-written pages covering services, pricing approach, coverage and policies will answer most pre-sales questions. Hundreds of loosely related pages can make answers worse, because the chatbot retrieves passages that look relevant but are not.
Does the chatbot learn from conversations automatically?
It should not, at least not without review. Letting it absorb conversations directly risks repeating a visitor’s mistake or an unapproved answer. A safer loop is to review unanswered and corrected questions each week, update the source content, and re-index. The chatbot then improves because your content did.
Bring us the task.
Tell us what happens today, which tools are involved and where it goes wrong. On a free 30-minute review, we’ll work out whether an agent, an automation or neither is the right fix, and what a sensible first release would be.
Start a project.

Book a call or send us a message. We read everything.
tash@neu.ie
Working with teams in Ireland and worldwide.
Free 30-minute AI review.