Product docs

Troubleshooting

Use this troubleshooting guide when indexing fails, the bot answers poorly, website discovery misses pages, uploaded documents parse incorrectly, or the embedded widget does not appear as expected.

Last updated: 2026-05-31

This guide covers the most common setup, indexing, and runtime issues in AskAnyDocs. Most problems come from one of three areas: source content quality, indexing configuration, or widget installation.

The bot answers poorly

First, check whether the answer exists in the indexed knowledge sources. Weak answers usually mean the content is missing, outdated, too broad, or contradicted by another source.

Use this checklist:

  • ask the same question in the bot preview
  • find the source page or document that should contain the answer
  • confirm the answer is written clearly in that source
  • remove outdated or duplicate versions of the same information
  • reindex the updated source
  • test the same question again

If users ask a question that your website never answers directly, add a new FAQ entry or documentation section. The assistant works best with explicit answers.

Website discovery finds too few pages

If you add a scoped website path, only pages under that path are considered. For example, a /help path will not include pages under /blog or /pricing.

Check:

  • the entered website URL and path
  • whether the site has a valid sitemap
  • whether the sitemap includes the missing pages
  • whether important pages are blocked by redirects or authentication
  • whether the pages are public and reachable

If discovery still misses pages, use a direct sitemap URL or a URL list for the exact pages you want to index.

Website discovery finds too many pages

Sometimes a sitemap includes blog archives, tags, author pages, legal pages, or old campaigns. Do not index every discovered page automatically.

Exclude low-value pages and focus on content that helps users answer product, pricing, setup, and troubleshooting questions. Too much unrelated content can make answers less precise.

A document is parsed incorrectly

Use the resource type that matches the content. PDF or DOCX should be treated as documents, not HTML pages. For image-heavy PDFs, scans, or screenshots, text extraction may be weaker than with normal digital documents.

If a document produces poor answers:

  • check whether the text can be selected in the original file
  • remove unnecessary cover pages or decorative content
  • split very large documents into smaller files by topic
  • convert complex tables into clearer text if needed
  • upload a cleaner version and reindex

The widget does not appear

If the widget is missing from your website, check the installation first:

  • the embed snippet is present on the page
  • the script URL contains the correct bot ID
  • the website domain is listed in allowed domains
  • the browser console does not show a 403 or script loading error
  • the snippet is included in the production template, not only a staging template

Allowed domains are a common cause. If the current origin is not approved, the widget can stay hidden even when the script tag exists.

The widget does not match the site

Review the widget settings page and verify color, avatar, starter question, and button text. Use the live preview to catch branding issues before launch.

If the widget looks cramped on mobile, shorten the bot name or starter question. Compact text usually performs better in a chat interface.

Escalation or fallback does not help users

If users ask questions that require a human, configure a useful fallback URL such as a support email, contact form, help desk, or live chat link. A fallback that points to a generic homepage may frustrate users.

Review escalated conversations regularly. Some should remain human-handled, but repeated escalations about the same topic usually mean the documentation needs improvement.

When to contact support

If a source repeatedly fails indexing or the widget behaves unexpectedly after saving settings, contact support and include the bot, resource, page URL, and browser details involved. For answer-quality issues, include the question asked and the source document that should contain the answer.

Escalation integrations do not receive user details

If webhook, helpdesk, CRM, or notification integrations are missing user context, check whether the widget receives window.AskAnyDocsUser before the embed script loads. The supported fields are id, email, name, and phone.

If the visitor is anonymous, the widget can ask for an email during escalation. Logged-in products should pass the known user email directly. See Pass user data to the widget and Integrations and escalation.

Related articles