Calculator

Knowledge Base Sizing

Almost everyone overestimates how many sources they need. More pages means more chances to retrieve the wrong one, so past a point every extra page makes answers worse. This works out where that point sits for your site.

Ask about CreobotSee all tools

YOU GIVE ITpage count and page typereadcomputereturnPAGES ON YOUR SITE5 of 36 worth indexingYOU GET BACKwhich pages help and which ones hurtRuns in your browser. Nothing is sent anywhere.
  • Fewer sources, better answersThe counterintuitive part
  • Names what to excludeNot just what to include
  • Maps to plan limitsSo the tier makes sense

Work it out

Your numbers

Result

—pages worth indexing
  • Answer-bearing pages —
  • Removed as duplicate or stale —
  • Pages that would hurt answers —
  • Plan that covers it —
  • Share of the site indexed —
  • Verdict —

A page earns its place if a visitor could ask a question that only that page answers. If two pages answer the same question differently, indexing both means the assistant will sometimes pick the worse one.

What the numbers mean

Two pages answering one question is worse than one

Retrieval picks what looks closest, not what is most correct. If an old pricing page and a current one both exist, some visitors get the old one, and you will not know.

Marketing pages rarely earn a place

They are written to persuade, not to answer. Indexing them teaches the assistant to answer operational questions in brochure language, which reads as evasive.

Start with twenty pages, not four hundred

Index the pages that answer the questions you already know people ask. Add more only when a real conversation shows something missing. Every addition should be earned.

Stale content is a correctness risk, not a tidiness one

An assistant will quote a superseded refund policy with total confidence. That is a worse outcome than having no assistant at all.

Questions

Up to a point, yes. The goal is coverage of real questions without duplication. Adding pages that answer nothing new only adds ways to retrieve the wrong thing.

Then start with the most-asked areas and expand deliberately, checking answer quality as you go. Large corpora work; large corpora added all at once do not.

They count the same for retrieval. A PDF handbook is one source, and it is usually worth splitting into the sections people actually ask about.

Want this run against your actual site?

Tell us the page your assistant lives on and the questions you care about. A person runs them and sends back the raw answers.