Can I restrict my AskVault bot to only answer certain types of questions?

Partly. Limit what it learns with crawl rules, keep strict mode on so it refuses questions your content does not cover, and add a scope line to the system prompt. Prompt wording alone can be talked around; what the agent never indexed it cannot reveal.

Can different visitors see different documents in one AskVault workspace?

Not per visitor today. A source is either public or internal. To give different customers different knowledge, use one workspace per customer.

Do AskVault crawl restrictions apply to WhatsApp and other channels?

Yes. Crawl rules decide what the workspace knows, so they apply to every channel that workspace answers on.

How to restrict an AI chatbot to specific URLs only

Written by Joseph Russo, Principal Architect, Networks and Tech, AskVault · Reviewed by Allison Chan

Last updated: October 20, 2026 · 5 min read

Three controls, three jobs

Each control solves a different problem.

Control	What it does	Where
Crawl rules	Decide which pages are indexed at all	Crawl settings
Audience	Keep a source off public channels	Each source's settings
Workspaces	Give different audiences different knowledge	Workspace switcher

With strict mode on (the default), the agent says it does not know when nothing relevant was indexed, so a page you never crawled cannot leak into an answer.

1. Limit the crawl

Open the crawl settings for your website source:

Include rules. Only URLs matching these patterns are crawled, for example /docs/* and /help/*.
Exclude rules. URLs matching these are skipped even if an include rule matches, for example /admin/* or /blog/*.
Maximum pages. A ceiling on how many pages one crawl takes.
CSS selector. Index only the main content area, such as .main-article, so menus and footers are not learned.
Use sitemap. Turn off if your sitemap lists pages you do not want considered.

Rules apply when pages are discovered, so a page that does not match is not fetched or indexed. One exception: if no page matches at all, AskVault still indexes the starting URL so the source is not empty.

2. Remove pages that slipped in

Rules only affect future crawls. If unwanted pages were indexed earlier, select them in the Knowledge list and delete them. Re-check after the next recrawl.

3. Keep internal sources off public channels

Each knowledge source has an Audience: public (the default) or internal. An internal source is removed from retrieval on the widget, the hosted page and the public API before the model sees anything, so the agent cannot quote it there.

Set it in the source's settings. There are no custom audiences or per-user rules; it is one switch.

4. Different knowledge for different customers

A source is visible to everyone who reaches that workspace's public channels. If each of your customers should only get answers from their own material, give each one a workspace:

One workspace per customer, with crawl rules limited to that customer's pages.
That workspace's widget token on that customer's pages.
Identity verification so the agent knows which signed-in user it is talking to.

Starter includes 2 workspaces, Growth 5 and Business 15. Growth+

Prompt rules are a backstop, not a wall

You can add "Only answer questions about our product and policies" to the system prompt. It helps with tone, but a determined visitor can talk a model round. Rely on what is indexed, not on wording.

Test your restrictions

Crawl test. Ask a question only an excluded page could answer. The agent should say it does not know. If it answers, find and delete that page.
Audience test. Mark a source internal, then ask about it on the widget. The agent should not use it.
Sources test. Ask in the chat playground and open the answer's sources to see which source it came from. The playground has no document picker, so it tests the whole workspace.

Common pitfalls

Relying on the system prompt alone. Prompt instructions can be bypassed. Crawl rules and audience cannot.

Tightening rules without deleting old pages. Pages indexed before the change stay until you delete them.

Expecting per-visitor document filtering. It does not exist today. Use separate workspaces.

FAQ

Can I restrict the bot to certain types of questions?

Partly. Index only what you want answered, keep strict mode on, and add a scope line to your prompt.

What happens when someone asks outside the scope?

With strict mode on, the agent says it does not have that information and can offer a person.

No. They only change what the agent knows.

Does this work on WhatsApp and other channels?

Yes. Crawl rules and audience apply on every channel the workspace serves.

Was this page helpful?