Anatomy of an answer

When the assistant replies, the most important thing is always the answer text at the top: the plain-language response to what you asked. Everything else on the card is supporting detail you can use when you want it and ignore when you don't. This page names each piece so you know what you are looking at.

The answer text

The first thing in every reply is the assistant's written answer. It is the protagonist of the card. If a reply also includes a step-by-step plan, the answer text still comes first, and the plan sits below it (and folds away on its own when you can't act on it right now, so the text stays the focus).

Inside the text you may see a few clickable highlights:

• An amber underlined phrase points at something on the page: a whole region, or a single button, field or link the assistant saw while answering. Click it and the assistant briefly outlines it so you can find it. If the page has changed since the answer was written and the thing is gone, the phrase tells you so for a moment instead of doing nothing.
• A cyan superscript number is a citation: it links to a knowledge-base article available during the turn.
• An indigo link is a resource the assistant found for you, like a file or a contact. Click it to open it.

The trust glyph

Next to the Appilot name at the top of the reply you may see a small colored icon telling you how the assistant arrived at its answer. Hover, focus, or tap it and a small bubble shows the full label and what it means:

Badge	Meaning
🟢 Verified action	Backed by something your organization set up: a control the steps point at, or one of its tools that ran for this answer.
🟡 From documentation	Cites an article available during the turn. Open it to check whether it supports the answer.
🔵 Based on this page	Read from what is on the screen, or steps built on things found there. Nobody wrote this down for you in advance.
🟠 Best guess	Nothing documented was cited and nothing on the page backs it. Double-check it before you act.
🔴 Not enough info	The assistant didn't have enough to answer confidently.

A single citation can earn the documentation badge for the whole answer. Green can come from one authored step in a mixed plan or a successful tool with no plan card. The badge identifies evidence to review; it is not a confidence score. Backoffice qualifies older records without a recognized rule version and distinguishes interrupted turns from completed answers.

The action plan (only when you asked it to do something)

If your request was an action ("create a folder", "fill this form"), the reply includes a plan: an ordered list of steps the assistant can walk you through. The header shows whether it's proposed, running, done, paused, or cancelled. When the plan belongs to a different screen than the one you're on, a short line says so (for example "This plan is waiting for you on Dateien-App") with a Take me there button. When you're already on the right screen, there's no extra line and you can run it straight away. Full details: see Guided actions.

You can read the plan two ways, and a small Narrative / Table switch at the top of the plan lets you choose:

• Narrative (the default) reads as a short numbered list of plain sentences, the same way you would say the steps out loud: "Click + New", "Select New Folder", "Enter My folder into the field labeled Folder Name", "Click Create".
• Table lays the same steps out in columns (step, action, field type, and a locate button), which some people find easier to scan. It is the same plan, just shown differently; switching back and forth never changes anything.

The highlighted control name in each step is clickable: tap it and the assistant briefly points out that exact button or field on the page so you can find it. A small colored icon before the name tells you its kind at a glance: a button to press, a text field to fill, a dropdown to pick, or a link to follow.

When a step means filling in a form, the whole form is treated as a single step, not one line per field, so even a long form never turns the answer into a wall of text. A one-field form shows its value right in the sentence; a longer one appears as a Complete the form step you can open to see the values the assistant will enter (a folder name, a description, a checkbox shown as Checked or Unchecked). Each value has a Preview on page button: press and hold it and the assistant shows that value right inside the real field on the page, then clears it when you let go. Nothing is saved until you run the plan. And because the form is one step, there is an Apply all button that fills every field on the page in one click (when those fields are on the screen in front of you); after it fills them, the assistant points out the button you'll press to finish, but never presses it for you.

If you ask the assistant to change one of those values after it has shown you the plan ("actually, call the folder Reports instead"), it updates just that field in the existing plan, in place, without rebuilding the whole list, so your place in the steps is never lost.

When you can't act on a plan from where you are, it collapses to a single "N steps · Show steps" line so it doesn't crowd the answer; tap it to see the steps any time.

The detail row at the bottom

A row of small chips sits under the reply. The two you'll use most:

• Sources lists the documents and pages the answer used.
• Feedback lets you give a 👍 / 👎 or report a problem, which helps your organization improve the assistant.

Open Feedback and you get two separate things, for two different situations:

• Report a problem is for an answer that was wrong or misleading. It appears once you have given a 👎.
• Ask the team to answer this is for an answer the assistant could not give, because nobody has documented the topic yet. It is always there, whether or not you rated the reply. Press it and a short form opens with your question already filled in; edit it if you want, then send. Your question and the page you were on go to the people who run the site, they write the answer, and you get a notification when it is published. This is the same knowledge request described in Knowledge requests, reachable at the moment you need it instead of from a separate view.

If the request cannot be sent, the form says so and keeps your text so you can try again. It never disappears silently.

Everything technical is tucked behind a Details chip so it doesn't clutter the reply. Open it only if you need it: inside are the response timing and a Debug ID. If something looks wrong, open Details, copy the Debug ID, and share it with your administrator so they can look up exactly what happened.

When something goes wrong

• A reconnecting message means the connection dropped briefly; the assistant retries on its own.
• An error reply shows what failed and, when it's safe, a Try again button. If it asks you to contact your administrator, the issue is on the configuration side and retrying will not help.
• Stopped means you (or a timeout) ended the reply before it finished.