Verified against ChatGPT · 2026-08-10
Build an FAQ page from your actual ticket history instead of guessing what people ask
Turns a batch of real support tickets into an FAQ page organized by the questions customers genuinely repeat, with answers phrased the way people actually ask, not how the product team assumes they ask.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are building an FAQ page from a batch of real support tickets, not from a blank-page guess at what people probably ask. RAW TICKET SAMPLE 40 tickets from the last 30 days: 11 asking about the free trial ending early, 8 about export formats not matching Excel, 6 about billing dates shifting after a plan change, plus 15 miscellaneous. PRODUCT/SERVICE CONTEXT A project management SaaS tool for small agencies, billed monthly per seat. EXISTING FAQ (if any, to check for gaps or outdated answers) Current FAQ says trials last 14 days, but recent tickets suggest it's now cutting off at 10 for some users due to a timezone bug. AUDIENCE Mostly non-technical small agency owners, many on their first project management tool ever. HOW TO BUILD THIS First, cluster the raw tickets into genuinely repeated questions — a question only earns an FAQ entry if it shows up more than once in the sample in some form; a single one-off ticket is not an FAQ, it's an edge case, and padding the page with edge cases buries the questions people actually search for. Phrase each FAQ question the way a customer would actually type or say it, not how the product team would formally title it internally — "why did my card get charged twice" not "understanding duplicate transaction holds." Write each answer to actually resolve the question in the first two sentences, then add necessary detail after — a customer scanning an FAQ page decides in the first line whether to keep reading. If the existing FAQ has an entry that's now outdated or contradicted by the ticket sample (e.g., a policy that changed), flag it explicitly rather than silently leaving it as-is. Group the finished questions into no more than five logical categories based on what the tickets actually cluster into, not a generic template of categories that may not match this business. WHAT NOT TO DO Do not invent a plausible-sounding question that wasn't actually represented in the ticket sample just to round out a category. Do not write an answer that links out to "contact support" as the entire answer — if the FAQ doesn't actually resolve the question, that's a sign it needed a real answer synthesized from the tickets' resolutions, not a redirect. OUTPUT FORMAT 1. Categories (up to 5), each with its clustered questions and answers. 2. A short note flagging any existing FAQ entries that are now outdated, with why. 3. A one-line note on any ticket in the sample that didn't fit any cluster and was excluded as a one-off.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Given only a product description and no real ticket data, GPT-5.1 will confidently generate a plausible-looking FAQ built from generic patterns common to that product category — which produces a page that answers the questions a product manager expects rather than the ones customers actually file, and the gap between the two is exactly what makes so many company FAQ pages useless in practice. Requiring the model to cluster from a real ticket sample and explicitly exclude anything that appears only once forces frequency to be the deciding factor rather than plausibility, which is the actual criterion that should determine FAQ placement but is invisible to a model working from imagination alone. Phrasing questions the way customers actually type them rather than how a product team would formally title them matters because FAQ pages are frequently searched via on-page or site search, and a mismatch between how a customer phrases a query and how the FAQ entry is titled means the answer might as well not exist — GPT-5.1 defaults toward the more formal, internal-sounding phrasing unless explicitly told to mirror the customer's actual words, since that formal register is more common in the kind of documentation text it's likely drawing stylistic patterns from. The instruction to flag outdated existing entries against the ticket evidence catches a specific real failure: policies and product behavior change, but FAQ pages are rarely audited against current ticket volume, so an entry that was accurate a year ago can now be actively generating more tickets by giving customers wrong information with confidence.
What you get back
Category: Trials & Billing Q: Why did my free trial end early? A: If your trial ended before day 14, it's likely a timezone calculation bug affecting accounts outside US time zones — we're aware of it and it's being fixed. Contact support and we'll extend your trial manually in the meantime. Flag: existing FAQ states "trials last 14 days" as an unconditional fact — this is now inaccurate for affected accounts and should note the known exception until the bug is resolved.
Verified against
ChatGPT GPT-5.1 · 2026-08-10
Changelog
- 2026-08-10 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
