Skip to content

15. Troubleshooting & FAQ

Most Klariton issues happen at three boundaries: data import, publishing, and embedding. Start with the checks below before changing large parts of your setup.

Check in this order:

  1. The embed snippet uses the current attributes: mode, org-slug, touchpoint-slug, and data-worker-url.
  2. The touchpoint is live.
  3. The page origin is allowed under Settings -> Organization or Settings -> Integrations.
  4. The production domain is correct.
  5. Staging URLs are enabled if you are testing outside the production domain.
  6. The page is served over HTTPS, except explicitly allowed local development hosts.
  7. The worker URL points to the correct Klariton worker or customer worker subdomain.

If you changed an org slug, touchpoint slug, domain, or worker subdomain, copy the snippet again from the Studio.

Possible causes:

  • the BIQ is still a draft,
  • the BIQ is in review but not published,
  • the BIQ was paused or rejected,
  • the answer has insufficient confidence for your publication settings,
  • the touchpoint is not live,
  • the wrong touchpoint slug is embedded,
  • translations are missing for the selected locale.

Publishing is explicit. Generating a BIQ does not automatically make it visible to end customers.

If Chat Advisor or a generated answer uses world knowledge, Klariton did not find a strong enough match in curated BIQs or material.

Fix options:

  • promote a recurring question to a BIQ,
  • add missing source material,
  • regenerate the relevant BIQs,
  • tighten Brand Voice and unsupported-answer behavior,
  • review material gaps in analytics.

See Connecting Data, BIQs, and Settings.

Translations can be missing, current, stale, or waiting for review.

Common causes:

  • a new locale was enabled after BIQs already existed,
  • the source answer changed after translation,
  • the translation engine requires review,
  • organization settings require review for every translation,
  • a SafeGuard correction invalidated older translations.

Use the translation view and filters for missing or stale entries. After bulk corrections, re-run translation for affected locales.

StatusLikely reason
401API key missing, malformed, revoked, or sent with the wrong header.
403API key does not have read_published.
404Touchpoint is not live, slug is wrong, touchpoint was deleted, or it belongs to another organization.

Use API keys server-side only.

Open the Trust Center for the Chat Advisor touchpoint. Activation is blocked until:

  • all mandatory notices are confirmed individually,
  • at least five preview self-test questions were sent,
  • the self-test is confirmed,
  • no required notice version is stale.

See Trust Center.

SafeGuard findings mean a published BIQ may contradict material, lack support, or violate Brand Voice rules.

Recommended flow:

  1. Open the finding detail.
  2. Compare “what the BIQ says” with “what the material supports.”
  3. Decide whether to send the BIQ to re-review, mark false positive, or generate a rewrite suggestion.
  4. After applying corrections, check translations and re-publish only after review.

See SafeGuard.

Test Center failures usually mean the chat answered outside the expected safety or grounding rules.

Common causes:

  • missing source material,
  • prompt injection not refused,
  • product recommendation not grounded in catalog data,
  • private or internal information leaked,
  • answer over-promised legal, price, delivery, or discount commitments.

Fix the underlying BIQs, material, or Brand Voice rules, then run the tests again.

See Test Center.

Check whether the limit is:

  • answer volume,
  • BIQ count,
  • touchpoint count,
  • locale count,
  • team seats,
  • provider budget.

Answer-volume limits and AI provider budgets are different. Changing one does not necessarily change the other.

Contact support when:

  • a live touchpoint with a correct snippet and allowed origin still does not render,
  • a published BIQ with valid confidence still does not appear,
  • a valid API key repeatedly returns the wrong status,
  • Trust Center or Test Center state looks inconsistent,
  • SafeGuard findings cannot be resolved from the Studio.

Include the organization slug, touchpoint slug, affected URL, current locale, and the error status or screenshot.