Skip to content
AskVault Docs
Try Free →

Q&A pairs in AskVault knowledge management

Written by Joseph Russo, Principal Architect, Networks and Tech, AskVault · Reviewed by Allison Chan
Last updated: · 4 min read

What Q&A pairs are

Most AskVault knowledge comes from crawled documents or uploaded files. Q&A pairs are different. They're hand-curated question-answer entries that the bot prefers over retrieval-based answers.

Three patterns where Q&A pairs fit:

  1. Tribal knowledge that never made it into a doc. "What's our policy on bringing pets to work?" might never have been written down. A Q&A pair captures it.
  2. Overriding ambiguous content. Your docs say two different things about refund timelines. A Q&A pair resolves the canonical answer.
  3. Frequently-misanswered questions. The bot keeps getting a question wrong because retrieval finds a slightly-related chunk. A Q&A pair forces the right answer.

Q&A pairs rank above crawled documents. When a question is close to one of your pairs, that pair gets a ranking boost, so it is usually the first source the bot writes its answer from.

Setup

Two ways to create Q&A pairs.

Manual entry

  1. Open Knowledge > Add Source > Q&A Pairs in AskVault.
  2. Click Add Pair.
  3. Enter the question and answer. Question is what a customer would type; answer is what the bot should say.
  4. Save.

Once it is indexed, test it by asking the question, and a paraphrase of it, in the chat playground. The bot should answer from the pair.

Auto-generation from existing content

Once you have content indexed, AskVault can draft Q&A pairs from it:

  1. Knowledge > Q&A Pairs > Auto-generate.
  2. Pick how many to generate. Between 3 and 30 per run; the default is 10.
  3. Click Generate.

AskVault reads your indexed content and writes pairs for the questions customers are likely to ask. Generated pairs are added and indexed straight away, so read through them afterwards and edit or delete any that miss the mark.

Generation uses your plan's analysis allowance, not your chat message credits.

Q&A pair structure

Each pair has:

  • Question. What the customer asks. Phrase it naturally, the way a real customer would type it.
  • Answer. What the bot says. Markdown supported. Length usually 50 to 300 words.

Q&A pairs have no Audience setting: they are always treated as public. The Audience public/internal switch applies to documents and text snippets only. Anything that must stay off public channels belongs in a document or snippet source, not a Q&A pair.

Phrase the question the way customers really ask it. If people ask the same thing in very different words ("What's your refund policy?" and "Can I get my money back?"), the closer phrasing matches better, so pick the most common one.

Retrieval priority

When a customer's question comes in, AskVault evaluates in this order:

  1. Retrieval. AskVault searches all your content, Q&A pairs included.
  2. Ranking boost. Q&A pairs get a boost over crawled pages, so a close pair usually ranks first.
  3. Answer. The bot writes its reply from the top-ranked sources and cites them.

The bot does not paste a pair's answer word for word; it writes from it. Keep the pair's answer phrased the way you want customers to read it.

When NOT to use Q&A pairs

Three patterns where Q&A pairs are the wrong tool:

  • The answer changes over time. Pricing, contact info, support hours. These should live in your docs and be resynced when they change, not in static Q&A pairs. Resync can miss a small edit such as a new price; see why.
  • The answer is long. Q&A pairs work best for short, factual answers. For multi-paragraph explanations, write a doc and let retrieval handle it.
  • You'd write hundreds of pairs. If you have 200+ Q&A pairs, you're better off writing a FAQ doc and letting retrieval work. Q&A pairs are for the top 20 to 50 questions; beyond that, retrieval scales better.

Limits

  • Count. No fixed number per plan.
  • Per-pair length. Question up to 2,000 characters, answer up to 10,000 characters.
  • Generation. 3 to 30 pairs per run.

Combining with FAQ Builder

The FAQ Builder (under AI Agent > FAQ) is a separate feature that builds a public FAQ page from Q&A pairs. Use it when you want the same Q&A content to surface as both bot answers and a static FAQ page on your site.

Configure under AI Agent > FAQ > Source = Q&A Pairs. Growth+

Common pitfalls

Bot doesn't pick the right Q&A pair. The customer's phrasing is too different from the pair's question. Reword the question to match how people actually ask, or add a second pair for the other phrasing.

Q&A pair answer is outdated. Edit the pair directly under Knowledge > Q&A Pairs > [pair] > Edit. Changes apply within seconds.

Auto-generated pairs are low quality. The auto-generator works from your indexed content. If your content is thin, the generated pairs will be thin. Better to write a few high-quality pairs by hand than to bulk-generate dozens of weak ones.

Conflicting answers between Q&A pair and document. The pair wins on high-similarity match. The retrieval-based answer wins on lower-similarity match. If you have a conflict, update both to agree.

FAQ

Are Q&A pairs better than crawled docs?

Different tools. Q&A pairs are for the high-traffic top-20 questions you want guaranteed-correct answers on. Crawled docs are for the long tail. Use both.

Can I import Q&A pairs from a CSV?

No. There is no CSV import for Q&A pairs. Add them one at a time under Knowledge > Q&A Pairs, or use the auto-generator to draft pairs from content you have already indexed and edit what it produces.

Do Q&A pairs count toward my content storage cap?

Marginally. Each pair is about 1 to 3 KB. 1,000 pairs = 1 to 3 MB.

Can the bot generate a new Q&A pair when a knowledge gap is detected?

Yes, if you turn on auto-train. In Suggest mode AskVault drafts a grounded answer for each recurring gap and waits for your one-click approval; in Auto mode grounded drafts become Q&A pairs automatically. A gap your content can't support is never auto-answered: it waits for you to add content. The knowledge.gap_detected webhook fires once per new gap in either mode.

How do I see which Q&A pairs are matching most often?

There is no per-pair usage report today. Test key pairs in the chat playground and watch AI Insights for questions that still go unanswered.

Was this page helpful?