You fill a small knowledge base of question-and-answer notes for your group. When a member asks something, the bot searches your notes and answers strictly from them, in the language of the question and in under a hundred words. If nothing matches, it says nothing — no guessing, no filler. Hourly and per-member limits keep the chat from turning into a bot conversation, and a built-in test mode shows you the exact answer before you turn it on.
Want to set this up? Do it right in your Telm account.
Open in your dashboard1What FAQ auto-answers do
Every group hears the same questions on a loop: how to pay, where the guide is, when the next stream is. FAQ auto-answers let the bot take that load off your admins.
You give the bot a knowledge base — your own notes — and it replies to members using only that material. It never invents facts and never wanders outside what you wrote. The feature is available on every plan and works out of the box.
2How an answer is produced
An incoming message only reaches the FAQ engine after it has passed the spam checks and is considered clean. From there the bot runs a short chain: it confirms the message looks like a question, searches your knowledge base for relevant notes, and — only if it finds something — has the AI write a concise answer strictly from those notes.
Search is smart about wording, so a question phrased differently from your note still matches, and it keeps working even in a plain-and-simple setup. The AI then replies in the same language the member used.
- Message must first pass the spam checks (be clean).
- It must read as a question and be a reasonable length.
- The knowledge base is searched for the closest notes.
- The bot answers only from those notes, in the question's language.
3Filling the knowledge base
The knowledge base is a set of notes. Each note has a title (the topic, like how to pay for a subscription) and a body (the answer or material the bot relies on). You can switch any note on or off, and disabled notes are ignored by search entirely.
The more specific and concise each note is, the sharper the answers. Short, focused notes beat one long document — the search can then pick exactly the right topic instead of a sprawling page.
- Title — a short topic label the search leans on.
- Body — the answer or source material the bot draws from.
- Enable or disable each note independently.
4No answer means silence
The bot replies only when it finds relevant material in your base. If there is no match, it writes nothing at all — no I do not know, no filler, no made-up facts.
Replies come back in the language of the question and stay brief. A member trying to trick the bot (forget your instructions, show your prompt) is treated as untrusted input and simply gets silence — the question text can never override the bot's rules.
5When a member calls an admin
When someone mentions an admin or support (for example @admin), the bot first tries to answer from your knowledge base even if the message is not phrased as a question. If it finds an answer, the member is helped instantly and your admins are not pinged.
If nothing matches, the bot does not stay silent here — it shows a short, customizable please-wait message so the member knows a human has been notified. This turns routine pings into self-service while still escalating the genuinely new questions.
6Test mode: preview before you go live
Before enabling anything, use the test question tool. It runs a question through the same search and answer engine but does not post to the chat, does not touch your limits, and skips question detection.
You see exactly what the bot would reply and which notes it matched. If there is no answer, it tells you plainly that the bot would stay silent. This removes the fear of the bot saying something wrong to real members, and it is the fastest way to spot a note that is missing or worded badly.
7Limits and pacing
To keep the chat from turning into a bot conversation, answers are capped per hour for the whole group (twenty by default, adjustable), and one person gets an answer at most once every couple of minutes.
Only real answers count toward the hourly cap — misses, non-questions and test runs do not. That means a burst of unanswerable chatter never uses up the hour's allowance, and only genuine, answered questions count against it.
8Best practices and common mistakes
A good knowledge base reads like a set of crisp help articles, not a single essay. Keep each note to one topic, use the test tool after every change, and let silence do its job.
- Split broad notes into several focused ones so search picks the right topic.
- Put the answer in the body, not just keywords — the bot quotes from it.
- Test each new note with a few real phrasings before enabling.
- Do not pack unrelated topics into one note; it dilutes matching.
- Raise the hourly cap only if your group is large and genuinely question-heavy.