Sit down to write an FAQ page and the questions rarely come. So the list opens with “what services do you offer?” — something nobody has ever asked. Yet the real list already exists: it is sitting in the enquiry emails and chat messages you have received.
Why imagined FAQs miss
You know your own service too well to remember where customers get stuck. Collect the actual enquiries and the top of the list looks different from what you expected — people ask about timelines more than price, and about whether they can keep what they already use more than features.
An FAQ does not exist to explain you. It exists to remove whatever is blocking a decision, which is why the questions have to come from real enquiries.
The procedure — collect, strip, group
Collecting means pasting recent enquiries into a plain text file in order. Twenty or thirty is enough for patterns to show. Include the questions you keep hearing on the phone, as best you remember them — phone questions are usually the important ones.
Stripping personal data is the rule that matters here. Names, company names, phone numbers, addresses and site URLs contribute nothing to grouping, so remove them before anything goes into an AI tool. An enquiry a customer sent you was sent to you, not to a third-party service. Only the substance of the question is needed, and the results are identical without the rest.
Grouping is the part AI is genuinely good at. “Group these enquiries by topic, tell me how many are in each group, and give one representative question per group” is a sufficient instruction. It is tedious, slow work for a human, and it is pattern recognition rather than judgement — which is exactly the kind of task that is safe to delegate.
You write the answers
An FAQ answer is effectively a public commitment. Write “usually about two weeks” and customers plan around it. So answers in particular must never be invented — write them yourself, within what you can actually deliver. Handing them back to AI for a final tightening pass afterwards is fine.
One useful habit while writing: include the unfavourable answers too. A line like “this work is not possible on shared hosting” looks as though it reduces enquiries, and it does — it reduces the wrong ones and keeps the right ones. Declining early always costs less than declining late.
Where it lives, and keeping it current
A single FAQ page and per-service question blocks are both valid; doing both is easiest when answers are maintained in one place and pulled into several screens. If the same answer is copied into two templates, one of them will inevitably go stale.
An FAQ is also never finished. Add entries when a question recurs, remove entries nobody reads. Running the same procedure once a quarter is plenty.
Other uses of AI for repetitive work are in the AI archive, and if you are not yet certain enquiries reach you at all, the mail deliverability part of the launching on your own series covers it. A whole-site diagnostic is part of our optimization program.