- Rewrite each FAQ answer as one to three spoken sentences, and keep a separate list of facts that must never change without a review.
- Mine the exact wording visitors use from chat logs, site search and email, and write headings in those words instead of your internal jargon.
- Give every answer a named owner and a review date, because a stale FAQ answer does more damage than a missing one.
Most FAQ pages were written for two readers: the search engine and the lawyer. That’s why they open with “Thank you for your interest in our services” and answer “Do you offer refunds?” with a paragraph that begins “Subject to the conditions set out in clause 4.2.” Visitors don’t read FAQs like that. They type a short, slightly impatient question into a box and expect a sentence back. If you want to automate FAQ answers, the first job isn’t the technology. It’s rewriting the answers so that a machine can say them without sounding like a terms-and-conditions document.
This guide covers how to do that. It explains why existing FAQ pages convert badly into chat, how to write answers that are short and specific without losing accuracy, how to keep one source of truth, and how to make sure a helpful answer doesn’t turn into a confident wrong one a year later.
Why an FAQ page doesn’t become a chat as it stands
A web FAQ is organised for scanning. A reader sees a list of headings, picks the one closest to their question, and reads down until they find what they need. The page can afford to be long, because the reader controls the pace.
An assistant works differently. It takes the visitor’s question, finds the passage in your content that best matches it, and builds a reply from that passage. If the passage is a dense paragraph written to cover every eventuality, the reply will inherit the density. If the heading is “Cancellation policy” but the visitor asked “can I cancel last minute,” the match may be weaker than it should be. And if the answer to a simple question is spread across three FAQ entries, the assistant may return only one of them.
There’s a second problem, which is that FAQ pages often answer questions nobody asks while missing the ones they do. A dental practice might have eight paragraphs on its sedation options and nothing at all on whether it takes children without a parent present. The FAQ reflects what the business wanted to explain. The chat exposes what visitors actually want to know, and the gap is usually large.
Mine the real questions before you automate FAQ answers
Start by collecting the real wording. You’re looking for the phrases visitors use, not the phrases you use internally. Four sources are usually enough:
- Chat transcripts and contact form messages from the last three to six months.
- Site search queries, especially ones that returned no results or returned a page nobody seemed to want.
- Email subject lines from customers, which often read like headlines: “Can I move Friday’s appointment?” or “Parking near the clinic?”
- Phone notes or call logs, if your team writes them, where the same questions appear in a different accent of language.
Once you have a few hundred examples, group them. You’ll see clusters. A physiotherapy clinic that did this discovered that its FAQ heading “Cancellation and rescheduling policy” was almost never searched. Visitors searched “can I move my appointment” and “what if I’m late.” The policy was identical; the heading was not. Rewriting the headings in visitors’ words made the answers easier to find, both for search and for the assistant.
This is also where you find questions you should have been answering all along. A question that appears in two dozen chats but has no FAQ entry isn’t a chat problem. It’s a content gap that’s been costing you time in the inbox.
Write answers that are short, specific and true
A good FAQ answer for an assistant has a clear shape. It answers the question in the first sentence. It adds the one condition that matters in the second. It ends with a next step where that helps. Most answers need no more than three sentences.
Here is a before and after from the clinic’s cancellation answer:
- Before: “We understand that unexpected circumstances can arise. Our cancellation policy is designed to be fair to all patients and to ensure that our practitioners can give each appointment the time it deserves. Cancellations received with less than 24 hours’ notice may be subject to a late-cancellation fee at the discretion of the practice.”
- After: “Yes, you can move or cancel. Give us 24 hours’ notice and there’s no charge. Inside 24 hours, a $35 late-cancellation fee applies. Would you like me to find you another time?”
The rewritten version is not less accurate. It states the fee, which the original hid behind “at the discretion of the practice.” It removes the reassurance paragraph, which visitors didn’t ask for. And it ends with an offer that moves the conversation forward. The information is identical. The experience is completely different.
Some writing habits make the difference:
- Answer first, qualify second. The direct answer comes in the opening sentence, so a visitor who stops reading still got what they came for.
- State the condition in plain words. “Unworn, with tags, within 30 days” is clearer than “subject to the applicable returns conditions.”
- Use numbers rather than adjectives. “Usually a few days” invites the question “how many?” Write “three to five working days.”
- Cut the throat-clearing. Lines like “We understand your concern” and “Thank you for asking” add nothing and make replies sound scripted.
Read your top ten answers aloud: if you wouldn't say a sentence to a customer standing at your counter, rewrite it until you would. Spoken-sounding answers are the fastest route to replies that don't sound robotic.
What makes a reply sound like a robot
Robotic-sounding assistants usually fail for reasons that have little to do with the model’s fluency and a lot to do with the content and rules around it. A short list of patterns to watch for:
- Over-apology. “I’m so sorry for any inconvenience” in response to a simple question about opening hours signals that the assistant is reading from a script.
- Repeating the question. “You asked about our parking. Parking is available” wastes a line and adds nothing.
- Hedging on facts. “Our prices may vary, and it’s best to check” sounds unhelpful when the price list is right there. If the price is fixed, say so.
- Unnecessary disclaimers. A long legal line at the bottom of every reply trains visitors to skip the reply.
- Lists where a sentence will do. A bulleted reply to “do you take card payments?” is overkill.
The underlying fix is to match the reply’s length and tone to the question. A yes-or-no question gets a yes or no, plus the one detail that matters. A complicated question gets a short answer and an offer to go deeper. Visitors notice when the reply size fits the question, and they relax when it does.
Keep one source of truth
The most common cause of contradictory answers isn’t a bad rewrite. It’s the same fact living in four places. The price appears on the pricing page, in the FAQ, in the email template your team sends, and on a laminated sheet by the phone. Someone updates the page and forgets the sheet. The assistant then gives the new price to one visitor and the old one to another, and both sound certain.
Keep a single master document for facts that matter to customers: hours, prices, policies, service areas, and any figures that change. The website FAQ, the assistant, the email templates and the staff scripts should all be drawn from it. When a fact changes, you change it in one place and then propagate it deliberately.
The table below suggests a review rhythm for each kind of answer. The point isn’t the exact cadence; it’s that every answer has one, and that someone is responsible for it.
| Answer type | Example | How often it changes | Review cadence | Owner |
|---|---|---|---|---|
| Fixed facts | Address, phone, parking | Rarely | Quarterly, and on any move | Office manager |
| Hours and closures | Opening times, bank holidays | Several times a year | Monthly, and before each holiday | Front-desk lead |
| Prices | Service and product prices | Once or twice a year | On every price change | Owner |
| Policies | Cancellation, refunds, returns | Occasionally | Twice a year, and on any policy change | Owner or operations |
| Seasonal information | Summer hours, delivery cut-offs | Each season | Before each season starts | Whoever runs that channel |
A stale answer is worse than no answer: if the assistant confidently quotes last year's fee, the visitor will hold you to it. Give every answer a review date, and treat anything that has passed its date as unverified until someone checks it.
Teach the assistant where its answers end
An FAQ can’t cover everything, and an assistant shouldn’t pretend it can. Some questions require judgment, and the right answer to those is a route to a person. “Can you make an exception to the cancellation fee because my mother was in hospital?” is a good example. An FAQ entry can’t authorise an exception, and an assistant shouldn’t promise one.
Write the boundaries down in the same master document. For each policy, note who can approve an exception and what the visitor should be told when one is requested: something like “I can’t change the fee myself, but I can pass this to our front desk, and they’ll reply within one working day.” The assistant can then be honest about its limits without sounding evasive.
The same applies to questions outside your business. A dental practice’s assistant should not give opinions on a rival clinic’s treatment, and a hardware shop’s assistant should not diagnose a car engine. Scope is a feature. An assistant that knows what it’s for sounds more credible than one that answers everything with the same confidence.
Keep it human without faking it
Automating an FAQ doesn’t require hiding that it’s automated. The most natural-sounding assistants are plain about what they are and warm about how they help. A few conventions do most of the work:
- Use the business’s name and the visitor’s question, not generic filler. “Our cancellation policy” sounds more human than “the policy.”
- Use contractions and short sentences. They’re how people talk, and they make replies easier to follow on a phone.
- Match the brand’s formality. A law firm and a surf shop should not sound the same, even when they answer the same kind of question.
- Say clearly when the assistant can’t help, and offer the next route. “I don’t have that detail, but the team can confirm it by email” is honest and useful.
Avoid two common mistakes. The first is performing friendliness: exclamation marks, “Great question!” and emoji clusters look like a sales script. The second is pretending to be a named person. A visitor who discovers that a warm assistant called “Sam” is a machine often feels misled, even when nothing untrue was said.
Make a maintenance habit out of it
Automating an FAQ is a one-time project followed by an ongoing habit. Set a calendar reminder for the monthly check, and keep it short. Review the answers with a review date that has passed. Read a sample of recent conversations and look for two things: questions the FAQ should answer, and answers that visitors pushed back on. Each pushback is a signal that an answer is unclear, too long, or out of date.
When something changes, update the master document first, then the website, then the assistant’s source, in that order. If the order is reversed, you’ll eventually find a page that says one thing and a chat reply that says another.
Where this leaves your FAQ
The aim of automating an FAQ is not to replace a good page with a chat window. It’s to make the answers short enough to be spoken, specific enough to be trusted, and maintained enough to stay true. Done well, the FAQ becomes the source of truth for the whole business, and the assistant becomes the fastest way to read it.
Voxy builds its answers from your own content, so a clean, current FAQ becomes a clean, current assistant. If you want to see how that feels in practice, try the live demo and ask a question your FAQ covers. For the wider logic of why an assistant should answer only from what you’ve provided, read how grounding keeps answers trustworthy. You can also see how Voxy is set up on your own content and test it free for 14 days, no card required.