The AI Interview

After your recording or upload is processed, the AI interviewer reviews everything and asks targeted follow-up questions. This is where the platform captures the context that makes your documentation valuable — the decision logic, edge cases, roles involved, and business reasons behind each step.

Open the interview from the "Continue Interview" link in the email notification, or by clicking your workflow in Scout. Most interviews take about 5–10 minutes — longer for more detailed or complex processes.

The interview page

Left side — the conversation:

• The AI asks questions about your workflow; you answer by typing, or click the microphone icon to dictate (it transcribes what you say and auto-sends). A long note above the question scrolls inside that note. The question and the answer box stay on screen.
• You can paste or attach images (screenshots, diagrams) with the paperclip icon — the AI can see them.
• A "Step X of Y" counter with a progress bar sits at the top of the question, filling in as you cover each topic. Click it to see which topics are covered, which are still to come, and anything left to clarify.
• Upload more documents during the interview with the plus icon by the message box, or from the "..." menu beside the question ("Add documents"). A small document count badge in the top bar shows what you've added — hover to see names and processing status.
• The "..." menu also has "Your answers so far" — a panel of everything you've already answered, where you can edit an earlier answer.
• If the interview came from a screen recording, a small video icon in the top bar replays the recording. If the AI references timestamps in its questions, click them to jump to that moment.
• Record more of this process — if you realise mid-interview that you left part of the process out, the "..." menu has "Record more of this process". It starts another screen recording through the Linc Chrome extension (same as your first one); when you stop, the interview picks up exactly where it was. A "Watching your new recording…" chip shows in the top bar while the new footage is processed, and the AI's next question acknowledges what it saw there. You can keep answering in the meantime. Each extra recording is listed under the video icon ("Recording 2", "Recording 3"…) and is used when the documents are generated. Available while the interview is open, for screen-recording interviews only (not recording days), one new recording at a time, and it needs an up-to-date extension. Finishing the interview waits until the new recording has been processed. If it can't be processed, the chip goes away and you can record it again.

Right side — your recording & info: the video with playback controls (play, pause, speed, fullscreen); for meeting recordings, a searchable transcript where any timestamp jumps to that point; plus the workflow name, status, department, and dates.

Voice mode

For a hands-free conversation — similar to voice mode in ChatGPT or Claude — use the audio-waves icon next to the message box (hover shows "Enable voice mode"):

1. Click the audio-waves icon to enable voice mode.
2. Speak naturally. The AI listens, auto-sends after you pause, and speaks its responses aloud. You can interrupt it while it's speaking — just start talking and it stops to listen.
3. Click the icon again ("End voice mode") to return to typing.

Tip: Voice mode is ideal for explaining complex workflows quickly without typing. It works best in a quiet environment with a good microphone.

Check what the AI has understood

A link reading "See what I've understood so far" appears under the message box once you've answered a few questions. Click it to open a panel showing the draft flow chart the interviewer has built from your answers, plus the pain points it has heard you mention (steps with a pain point are outlined in red).

It's there so you can spot a misunderstanding while there's still time to fix it. The map redraws after each answer. The map is not editable — if something on it is wrong, just say so in the chat and the interviewer will correct it. To point at a specific box, use the "Mention a step" buttons at the bottom of the panel — they drop that step's name into your reply.

You can answer as soon as the question appears — you don't have to wait for the topic tracker or the suggested-reply chips underneath. While those chips are still coming, faint placeholders sit in their row so the page doesn't jump; they turn into real shortcuts a moment later, or disappear if the question doesn't have short answers. If you reply first, they restart from the new answer.

Questions about what you didn't mention

After you've described how the process works, the interviewer may ask about things your account didn't touch — "Is there an approval before the order is placed? Who gives it?", "Does anyone check the receipt against the order before the invoice is paid?". It asks because a process of that kind usually has such a step somewhere, so it wants to know whether yours does. There is no right answer: "no, there isn't" is a complete answer, it's recorded as such, and nothing is added to your SOP that you didn't describe. It works the other way too — if you mention a step that isn't usual for this kind of process, expect "what triggers it, and who does it?" and, once, "is there a reason it's done that way here?" — that's to capture the reasoning, not to question it.

These questions are the same for every organisation and need no setup; the interviewer works them out from the workflow name, the process the session is tagged to (if any) and your organisation's industry context. A session with a name too vague to tell what kind of process it is (say "misc") skips them and sticks to the standard topics. In Architect's Benchmark tab, your answers to these questions appear as interview evidence next to the comparison, so a reader can see what was said rather than only what the SOP writes.

Questions that name your colleagues and other processes

When your session belongs to an initiative, the interviewer also knows who else is in it — the departments the initiative covers and the people listed in them in your organisation's People directory — and which processes in the Library sit near yours (the ones your process's documentation hands off to or is fed by, and those in the same department or with a similar name). So after you've described who does what, it may check the edges by name: "Is Maria Souza involved in this, or only Treasury as a team?", "Does this hand off to 'Invoice Matching', or is that a different flow?". These are confirmation questions, not assumptions: nobody and nothing is added to your SOP unless you say they're part of it, and "no, that's a different flow" is a complete answer. It only uses names from departments the initiative has in scope, and it never reads other people's interviews. A session with no initiative, or an organisation with no People directory yet, simply gets the usual "who else is involved?".

Tips for a great interview

• Be thorough. Mention edge cases, workarounds, and informal steps you normally take — the more context, the better your SOP.
• You can redirect the AI. If a question isn't relevant, tell it to skip and move on. The interview is conversational.
• You can hold the interview in any language — though the generated documents are always written in English.
• Don't leave without completing. Deliverables only generate when you explicitly finish the interview (see Completing the Interview).