Guide
How to give your support bot and AI agents one source of truth for product documentation
Pick one canonical document per product fact — specs, limits, plans, policies — publish it at a permanent link in a text format AI tools read well, and point your support bot, your internal agents and public AI crawlers to that link instead of to copies. Keep it versioned and update it in place, so every bot answers from the same current, checked content.
Why bots give different answers
A support bot answers from what it was given: a help center export, an old PDF spec sheet, a wiki page someone copied last year. Public assistants answer from whatever they found on the web. When the same fact lives in five places, each bot picks a different one.
The result is a bot that quotes a retired limit, a price from last year, or a feature that only exists on one plan — and a customer who trusts it.
Decide what the source of truth is
Every other copy — slides, onboarding emails, sales decks — should link to the canonical document rather than restate it.
| Fact type | Canonical document | Owner |
|---|---|---|
| Specs and limits | Spec sheet per product | Product |
| Plans and pricing | Plans document | Product marketing / finance |
| Policies (refunds, SLA, data) | Policy document | Legal / support |
| How-to and troubleshooting | Help center articles | Support / docs |
| Release changes | Changelog | Product |
Write docs that bots answer from correctly
- One fact, one place, one wording. Do not restate limits in three articles.
- Write limits and numbers with units and scope: “Up to 10 GB per workspace on the Team plan”.
- Say which plan, version or region a statement applies to.
- Date-stamp facts that change, and mark deprecated features clearly.
- Use real headings and tables; avoid screenshots of settings or tables.
- Give each document a short summary at the top that answers the most common question.
Connect the bots to the source
- Point your support bot’s knowledge source at the canonical documents, ideally in Markdown or JSON rather than PDFs.
- Publish a room index (llms.txt) that lists your canonical documents, so external AI tools can discover them.
- Re-sync the bot whenever a canonical document changes, or have it read the live link.
- Review bot transcripts for answers that do not trace back to a canonical document.
How The Papers fits
The Papers publishes each canonical document as an AI Version at a permanent URL: a page for people, Markdown with front matter (canonical URL, version, status, AI Readiness) and JSON with blocks, metrics and page provenance. A Data Room index at /{room}/llms.txt and /{room}/index.md lists every document.
Before publishing, every number, date and period must match the source, and qualifiers such as “up to” stay attached. Updating a document keeps its URL and its version history, so the bot and public assistants always read the current version. Analytics show which AI agents, crawlers and tools read each document and in which format.
Public documents can be listed and indexable, and a custom domain such as ai.yourcompany.com keeps the room on your own address. It suits your key reference documents; it is not built for bulk conversion of an entire archive without per-document fidelity check.
FAQ
- Should the help center itself be the source of truth?
- For how-to content, often yes. For hard facts such as limits, plans and policies, a single canonical document that help articles link to is easier to keep correct.
- Which format should a support bot read?
- Text formats such as Markdown or JSON are easier for retrieval and less error-prone than PDFs. The Papers publishes both for each document.
- How do public AI assistants find our documents?
- Make the canonical documents public and indexable, and list them in an llms.txt index so AI tools and crawlers can discover them.
- What happens to links when a document changes?
- With The Papers, the URL stays the same and the version history is kept, so nothing that links to it breaks.