The composite scenario
Mira is the fictional founder of Northstar, a B2B workflow product. She wants to explain why the company removed a "smart recommendations" feature before launch.
She has no finished script. She has a public product update, a page explaining the workflow, notes from three pilot reviews, and a concern that the decision may sound anti-AI.
Her goal is one accurate explanation for customers and product peers. She does not need a viral claim.
Step 1: provide the source link and context
Input: the public product update, workflow page, and Mira's notes.
REC's role: organise the context for research and question preparation.
Human responsibility: confirm that the material is safe to use and distinguish public facts from private notes.
The public update says the feature was removed because recommendations were inconsistent. The private notes add an important detail: users could not tell which data each recommendation used.
That distinction becomes a likely interview thread. It should not be published until Mira decides it is accurate and appropriate.
Step 2: research the subject
REC's AI-assisted preparation can examine the supplied material and public context. It looks for the product mechanic, claim, unresolved question, audience, and likely objection.
Candidate tension: Was the feature removed because the model was poor, or because the product could not make its reasoning inspectable?
Annotation: AI has identified a question. It has not established the answer.
The research brief gives Mira a chance to catch a mistaken assumption before recording.
Step 3: prepare a question path
A useful sequence might ask what job the recommendation feature was meant to do, what happened in the pilot reviews, which part of the experience made the team uncomfortable, what alternatives the team considered, why the team removed the feature instead of labelling it beta, what would need to change before reconsidering, whether the decision is evidence against AI recommendations generally, and what a customer should understand about the current product.
Annotation: The questions move from mechanics to evidence, decision, objection, and limit. They do not script Mira's conclusion.
Step 4: check the recording setup
Before the interview, Mira checks camera, microphone, framing, and the environment.
She also reviews boundaries: no pilot customer names, no unannounced roadmap dates, no claim that all users disliked the feature, and a pause if a number needs checking.
Annotation: Technical readiness and editorial readiness are different. A clear camera cannot make an unsafe answer publishable.
Step 5: record the solo interview
Mira answers the questions on camera. One answer is initially vague: "We wanted to keep trust at the centre."
A follow-up prompt asks for the exact moment the team changed direction. She answers: "In the third pilot review, the user asked which activity produced the recommendation. We could show the output, but not a useful evidence trail. We considered adding a confidence label. That would have described uncertainty without explaining the source, so we removed the feature."
Annotation: The second answer contains an event, limitation, rejected alternative, and decision. The answer becomes source material, while remaining Mira's recollection and requiring review.
Step 6: create and review the transcript
The recording becomes a transcript. Mira checks "third pilot review," the wording of "confidence label," whether "removed" accurately describes the product state, and whether any customer detail slipped in.
She corrects a mis-transcribed product term.
Annotation: A transcript improves search and traceability. It is not automatically accurate.
Step 7: identify grounded highlights
AI helps surface candidate passages tied to the transcript. Possible highlights include the difference between uncertainty and traceability, why a confidence label was rejected, the condition for reconsidering the feature, and the answer to "Are you anti-AI?"
Each candidate includes enough surrounding answer to review the meaning.
Annotation: Selection is a recommendation. Mira and the editor decide what deserves publication.
Step 8: choose and cut the clip
The team selects the passage about the confidence label. They include the sentence before it so the viewer understands the problem.
They reject a shorter cut that says: "We removed AI recommendations because users could not trust them."
That version overgeneralises one pilot observation and implies a broader conclusion.
The approved clip says the team could not provide a useful evidence trail in that workflow and removed the feature until it could.
Step 9: write accurate framing
Caption: "Northstar removed a recommendation feature after a pilot user asked which activity produced the result. In this composite example, the founder explains why a confidence label did not solve the traceability problem."
In a real session, the caption would not say "composite." Here it must, because no real customer or product result is being presented.
Step 10: final review
Before publication, the team checks whether the clip matches the transcript, whether the product information is current, whether the pilot story is safe and permitted, whether the title preserves scope, whether factual claims are linked, whether the important limit is visible, and whether the team can retrieve the source answer later.
What this walkthrough does not show
A REC session does not guarantee a strong answer, accurate claim, good clip, or business outcome. AI can prepare a weak question or select a passage that loses context. Transcripts can contain errors. Human review remains necessary.
REC is not an automatic customer-proof generator, fact-checker, social scheduler, or replacement for legal and privacy review.
Try the workflow on one real decision
Choose one public link and one decision you can safely explain. Let REC prepare the research-guided questions. Record the answer, check the transcript, and select only the highlight that remains accurate when it leaves the full session.