Skip to main content
Loading...

Ask AI FAQ

Ask AI FAQ

Answers to common setup and operating questions about Ask AI.
For what the app is and how to set it up from scratch, see the
Ask AI Overview.


Setup

Do I need to do anything beyond enabling Ask AI before employees can use it?

No. Ask AI requires a licence, and an admin turns it on from Admin → Apps → All Apps. Once it is enabled and your organisation is on the Platform Standard edition or above, employees see the chat sidebar immediately. The default settings — action execution on, voice on, shared knowledge on — are ready to use. You can refine settings and Agent Guidelines at any time, but nothing else blocks employees from asking their first question.

Who can manage Agent Guidelines, Verified Knowledge, the Prompt Library and the other admin tabs?

The Settings tab is open to admins and Ask AI app admins, and an Ask AI app admin can also manage Embedded assistants. Managers can open the Test Suite. Every other admin tab — Agent Guidelines, Questions, Verified Knowledge, Knowledge Map and Analytics — requires the full admin role (admin or above).

Any employee can browse the Prompt Library and use an existing prompt. Only admins (admin or above) can create, edit, delete, or save tenant-private prompts.


Permissions and access

Why can some employees not see Ask AI, or get redirected when they open it?

Three things to check. First, the app must be enabled for the organisation (Admin → Apps → All Apps → Ask AI). Second, if you have restricted the app’s audience to a specific group in Admin → Apps → All Apps audience settings, only members of that group will see the Ask AI entry point — everyone else is redirected. Third, Ask AI requires the Platform Standard edition; it will not load on Platform Lite.


Day-to-day

Where does Ask AI pull its answers from?

Ask AI draws on four layers. First, shared company knowledge — Knowledge Base, HR Policies, SOPs, Sites, Wikis, Drive, contracts, posts, training courses, forms, AI Notepad notes, custom-app records and more — each of which you can toggle on or off individually from Agent Guidelines → AI Features. Second, Verified Knowledge entries that admins have approved. Third, live data pulled from connected apps through agent tools (schedules, leave balances, tickets, etc.). Fourth, the files an employee attaches to a message. Optionally, you can include platform help-centre articles by turning on Include Platform Help Articles in Answers in the Settings page.

How does agent routing work, and can employees force it?

Each question is classified and routed to whichever specialist agent owns that domain. There are over 100 built-in domain agents, and each one is gated by the same licence and audience rules as its parent app. If a question spans two areas, the first agent hands the remainder to another mid-conversation. Employees can also force routing by @mentioning an agent inline (e.g. @hr, @scheduling).

What happens when I approve a Verified Knowledge entry?

The entry moves from the Pending review tab to the Active tab and immediately becomes part of the retrieval corpus. Future answers can cite it. You can also edit the entry before approving. To remove it later, retire it — it moves to the Retired tab and stops influencing answers.

If the entry answers an unanswered question group, approving it also closes that group. Reject this correction hands the question back to the person who wrote the answer (see “I’ve been assigned an unanswered question” below).

How do Journeys work and who controls them?

Journeys track multi-step goals (onboarding a hire, migrating a policy) across sessions. Each employee can have up to 3 active journeys at a time, each with up to 12 steps. The assistant updates progress during conversation, and the employee can override any step from their My Journeys page. Journeys are toggled from Agent Guidelines → AI Features: the Journeys toggle controls whether goals are tracked at all, and Journey inference controls whether the assistant infers progress from conversation context. Both are on by default.


Unanswered questions

How do I see what people asked that Ask AI couldn’t answer, and why are some questions missing?

Open Ask AI → Agent Guidelines → Questions (admin or above) and click Group questions, or Regroup questions after the first run. Ask AI takes the last 30 days of questions it couldn’t answer from company knowledge and groups them by meaning. Each run considers up to the 200 most recent questions and uses your organisation’s AI usage. Groups keep their identity across regroups, so an owner and a history carry over while people keep asking.

A group is shown only when at least 5 different people asked it. Smaller groups are counted in a note under the list but never displayed, so nobody can be picked out from what they asked. The Narrow to filter works on one department or one country at a time, and each slice must also reach five people.

Who owns an unanswered question group?

Each shown group becomes a Content Governance finding with the issue type Unanswered question. Ask AI assigns it to a topic expert for the group’s subject who can open Ask AI; if there is none, it stays unassigned and the admins get the reminder. The finding is rated high when 20 or more people asked, otherwise medium. Use Assign or snooze on the group’s row to change the owner or snooze it in Admin → Content Governance.

I’ve been assigned an unanswered question — what do I do?

Open it from the reminder, or from Content I own in your profile menu. The page shows the merged question and how many people asked, counted but never named. Click Create an answer, write it, and choose Send for review. Your admins then approve or reject it in Verified Knowledge. While it waits, the question stays with you and the reminders stop. If they reject it, you get a “Your answer wasn’t approved” notification and the question is yours again.

What happens when I click “Create an answer” as an admin?

A drawer opens with the merged question already filled in. You can describe the facts and click Draft to have AI fill the question and answer — it never saves on its own. Choose who the answer can influence, then Save answer. An admin’s answer goes live in Verified Knowledge at once, closes the group and resolves its finding.

Can a question group close without anyone writing an answer?

Yes, in two ways. Once a week Ask AI re-asks each open group’s questions as up to 3 of the people who asked, with their own permissions and nothing saved. The group closes only if every one of them now gets an answer from company knowledge. An admin can also resolve or dismiss the group’s finding in Content Governance. Dismissing is permanent: the same question keeps landing in that dismissed group.

How do I stop the unanswered-question reminders?

Owners, or the admins when a group has no owner, get a weekly reminder. Each person can mute it in their notification preferences under Ask AI unanswered questions, which is on for in-app and email by default. That setting also covers the “answer wasn’t approved” notice. For the whole organisation, open Admin → Content Governance → Settings. On the Ask AI card, turn off Notify on governance review to stop reminders. Turn off Enable automated governance scanning to stop filing findings altogether; it is on by default.


Voice

What are the voice limits, and which voices can I choose?

Each voice session is capped at 15 minutes. The per-user daily limit defaults to 30 minutes and is configurable from the Ask AI Settings page; setting it to 0 removes the daily cap entirely. When an employee hits the daily limit, the voice button is disabled until the next day.

Ask AI supports 10 voices: Marin (default), Cedar, Alloy, Ash, Ballad, Coral, Echo, Sage, Shimmer, and Verse. Marin and Cedar are the newest and recommended for best quality. The admin chooses one voice for the whole organisation from the Settings page, where an in-settings preview lets you hear each voice before selecting it.


Attachments

What files can I attach to a question, and what are the limits?

Employees can attach up to 3 files per message. Images (PNG, JPG, WebP, GIF), PDFs, Word, Excel, CSV, and plain text have an 8 MB limit each. One of the files can be an MP4, MOV, or WebM video, up to two minutes and 100 MB. Supported video codecs are H.264, HEVC, VP8, and VP9, at up to 4K resolution.

Images use vision; documents use text extraction. PDFs are read up to 100 pages. Scanned or unreadable content produces an explanation when text cannot be extracted.

How do I ask about a video?

Use Attach files in the composer. On mobile web, Attach files or a video opens the full composer for the current conversation. Wait until the chip says Video ready, then ask your question. Processing combines timestamped speech, when available, with up to six sampled frames. Answers should cite timestamps and explain unclear details; sampled frames do not show every action between them and do not certify compliance.

Remove the attachment to cancel processing and clear its extracted evidence. If processing fails or is interrupted, use Retry processing. Each clip allows up to three processing attempts; each employee can start six attempts per hour. Organization AI usage limits also apply. Private video uploads and extracted evidence expire after 24 hours; answers already in the conversation follow the conversation’s retention policy.

Why was my video refused or left waiting?

Use a supported export within the size, duration, and resolution limits. Corrupt files, unsupported codecs, and over-limit clips are rejected before processing. No audio track is supported: the assistant uses visual samples. If a worker is interrupted, the status offers retry after ten minutes without progress. You can remove a failed attachment; the composer blocks sending while an attachment is still processing or failed.


When something looks wrong

“The AI said it can’t help and offered to open a ticket — what does that mean?”

When the assistant detects that it cannot confidently answer — a capability refusal, an apology answer, or an empty result — it offers a Get human help button. People can also just ask for a person in the conversation (“can I talk to someone about this?”), and the assistant sends it the same way.

Where it goes:

  • With Service Desk enabled for the organisation and open to that employee, it files a Service Desk ticket carrying the question, the AI’s answer and the turn reference, routed by your Service Desk rules on the request type the question is classified into.
  • Without Service Desk (or when the employee’s audience rules exclude them from it), it goes to the admins of the app the question was about — for example the Leave Management admins for a time-off question — as a notification with the question, the answer and a link to the employee’s profile. If that app has no admins of its own, your Ask AI admins (or, failing that, your workspace administrators) receive it. The person asking never receives their own request.

The button is hidden only when nobody could receive it — for example when the only administrator is the person asking.

“My employees say they can’t clear their conversation”

Check whether Allow users to clear their conversation is turned on in the Ask AI Settings page (it is on by default). When this is off, the Clear conversation and Reset AI completely controls are hidden on every surface — web, mobile, and API — and the server returns a “disabled by your administrator” message if the endpoint is called directly. Clearing is a view-level action (the governance record is retained for the retention period), so this toggle controls a UX affordance, not data retention.


Shared AI conversations

Can I invite coworkers into a private conversation, and remove them later?

From History, choose Share conversation. Edit the summary, select up to seven coworkers, choose contributor or viewer access, and review the exact content before sending invitations. Each coworker must accept before seeing the shared conversation. The original private history, attachments, memories and tool traces stay private.

The owner opens Participants to change contributor/viewer roles or revoke access. Revocation blocks subsequent reading, search and exports; it cannot remove copies someone already downloaded. Contributors send as themselves. Shared conversations do not become personal AI memory.

What can Ask AI use in a shared conversation?

It uses the reviewed summary and shared messages. You can include a Service Desk ticket when every active participant can open it. Answers and follow-up context are checked again when access changes. Other sources and actions remain available in your private Ask AI conversation.

What happens if an answer is interrupted?

Your message stays in the conversation. A failed reply offers Retry reply to its sender. Live updates preserve your unsent draft. Reusing a submitted form with edited text produces a clear error instead of silently ignoring the edits; refresh and submit it as a new message.


Conversation retention and data

How long are conversations kept?

The default is 24 months. An admin can change this from the Settings page; the allowed range is 1–36 months. Conversations older than the configured period are permanently deleted by a nightly sweep. Shortening the retention period removes history immediately on the next sweep and cannot be undone. Diagnostic flow logs are kept separately for 180 days or the tenant’s retention period, whichever is shorter.

How do I handle a GDPR data-subject request for Ask AI data?

Data requests are handled for every app at once, Ask AI included. Go to Admin → Retention & Legal Holds → Requests about a person (admin-only). Find the person, check what each app holds about them, then choose Export a copy (Article 15/20) or Erase their data (Article 17). Erasure is irreversible. Each request covers the current organisation only — a person who belongs to several organisations needs a request in each.


Licensing and limits

What are the hard limits I should know about?

What Limit
Files per message 3
Image/document size 8 MB
Video size / duration 100 MB / 2 minutes; one video per message
Video processing Up to 6 sampled frames; 3 attempts per clip; 6 attempts per employee/hour
Video evidence retention 24 hours
PDF page extraction 100 pages
Voice session length 15 minutes
Voice daily limit (default) 30 minutes per user (0 = unlimited)
Active journeys per user 3
Steps per journey 12
Conversation retention 1–36 months (default 24)
Diagnostic flow-log retention 180 days or tenant retention, whichever is shorter

More help

On this page