What the wording does
A useful default for this project is a capable colleague making the work understandable. That is an editorial recommendation drawn from the available texts, not an official OpenAI voice rule. The recurring move is to give someone a recognizable task, make a change understandable, then return to a person's judgment or next action.
The collection has real variation. The beginner lessons normalize trying again. Product demonstrators sound exploratory on the page. Technical explanations keep prerequisites in view. Customer stories allow emotion, and launch testimonials sometimes use expansive praise. A universal instruction to sound calm, avoid adjectives or remove every hesitation would flatten those differences.
For a product segment in an all-hands, start with the working demonstrator. Use the technical peer when a mechanism matters; use a personal story when you have a real person's experience to tell. The six approaches above are editorial groupings. Several films move between them.
Choices that shape the tone
Give each pronoun a job
I can own a choice, an observation or an opinion. We can invite the viewer to inspect something together. You can give the learner a useful action. Moving between them changes the relationship with the audience. It also changes who appears responsible for the result.
The dashboard tour uses shared inspection to guide attention; the beginner lesson gives the learner choices. For your script, decide who is choosing, who is doing the work and who must check it. Avoid an unqualified “we” that could mean either the presenter, the product or the whole company. Dashboard passage · First-steps lesson
Make reassurance actionable
The prompting lesson treats an imperfect answer as a reason to clarify a request. Its practical details matter more than a general promise that the tool is easy. The organization lesson similarly allows a shorter format and returns the final choice to the person using it. Prompting lesson · Organization lesson
When writing instruction, attach reassurance to a next step: try one item, add the missing context, change the format, inspect the answer. Repeatedly calling a task simple can make a learner feel at fault if it is difficult. A useful action is a stronger invitation.
Preserve the words that limit a claim
Can, might, if, before, approximately and in this example carry meaning. In the leadership-meeting demo, the presenter distinguishes a drop-off from two possible explanations and asks for more context. That uncertainty is part of the reasoning. The semantic-layer explanation keeps a knowledgeable reviewer and ongoing organizational work in the account. Decision demo · Trust explanation
Editing away every qualifier can turn a competent explanation into an unsupported promise. Keep uncertainty that changes what a viewer may conclude. Remove a hesitation only when it adds no meaning and its removal still respects the speaker's performance.
Name the object before you evaluate it
The design demo names adjustable properties; the dashboard demos name labels, charts and views. Concrete nouns give conversational commentary something to point to. Technical detail earns its place when it explains a step in the task, as in the incident-investigation example. Design exploration · Customization demo · Technical walkthrough
In a draft, underline this, that, it and better. Check whether the listener already knows the referent and the standard of comparison. Replace only the ambiguous instances. Repeating the object's name at a change of view can be more useful than finding a new synonym.
Let enthusiasm keep its owner
The first-impressions montage includes strong praise and personal ambition; the evaluation interview includes candid limitations. These are people reporting their judgments. Treat their enthusiasm as part of a situated account. Do not silently promote a customer's estimate into an independently verified product result. First impressions · Personal evaluation
For a benefit claim, retain who experienced it, what task it concerned, what the comparison was and whether the outcome was measured or estimated. An aspiration also needs its tense: the molecular-science story describes an imagined benefit and a motivating hope, not a demonstrated clinical outcome. Research story
Sentence rhythm on the page
These texts support a reading of thought order and rhetorical repetition. They do not establish vocal pace, pauses, emphasis, pitch, accent, warmth of performance or music. Machine punctuation and caption line breaks are especially poor evidence of breath or cadence. No words-per-minute target is derived here.
For a new script, a useful pattern to try is an orientation, one action, then a consequence or qualification. It can take three sentences or a single well-shaped sentence. Vary the length according to what the listener has to understand. A list of three equally short declarations can be less natural than a connected explanation.
The following is an original rehearsal example, with proposed thought breaks rather than measured timing:
We have three open items for Friday's review. / I'll add an owner to each one. / Then the team can decide what is ready and what still needs work.
Read it aloud while looking at the planned picture. Check where the listener needs time to locate an item or read a label. Establish the actual holds, stresses and edit points against the supplied recording. Keep a purposeful hesitation or an exploratory maybe when it expresses the person's real decision process.
Words and pictures need the same evidence
The earlier words-and-shots studies provide sampled visual evidence for 12 films. This transcript pass adds a language reading; it does not add a continuous shot-by-shot or listening review. Use the following as proposed production checks.
| What the line asks the audience to do |
What the picture needs to supply |
Check before recording or editing |
| Inspect a named item or “this” result |
The object, legible at the moment it is named |
Can a viewer locate it without already knowing the interface? |
| Follow an action |
The relevant starting state and the action's consequence |
Are you describing a request, a preview or a confirmed result? |
| Understand a condition |
Enough continuity to keep that condition attached to its claim |
Does the cut imply certainty while the sentence is still qualifying it? |
| Compare two things |
A stable basis of comparison |
Are the compared values, versions or objects the ones actually being discussed? |
| Believe a reported benefit |
A concrete example or clearly attributed testimony |
Does the visual support the stated scope, rather than a broader implication? |
| Understand a person's stakes |
A particular place, object, activity or memory from their real account |
Is the detail theirs, and does it explain why the work matters? |
One useful restraint comes from the first-steps lesson: a simpler answer is requested in the studied sequence, but the revised answer is not shown there. The narration should preserve that distinction. The website-building demo likewise contains requests about sharing and future updates; those requests alone do not prove recipient access or later execution. Lesson study · Website study
Four useful draft edits
These are invented writing exercises, not quotations or corrections of a source speaker. The revised version is appropriate only when its factual assumptions are true.
| Draft line |
A more specific version |
What the change does |
| “This makes it better.” |
“The owner column tells us who needs to check each item.” |
Names the object and the particular improvement. |
| “The report explains why people left.” |
“The report shows where people left. We still need to understand why.” |
Separates an observation from an explanation. |
| “The update has been shared with everyone.” |
“I'll review the draft, then share it with the team.” |
Fits a sequence where sharing has not happened yet. If it has, state and show the confirmed result. |
| “This saves the team hours.” |
“In this example, the list puts the open items in one place.” |
Describes the demonstrated benefit when there is no time-saving measurement. |
Specificity does not require flattening a person's voice. A customer's surprise, a designer's uncertainty or an instructor's invitation can carry the story. Keep those when they belong to the person and the task.
A working brief for the all-hands
Use this as a proposed writing brief. It is not an approved storyboard or a verified description of a Poe feature.
- Relationship: speak as a colleague who understands the work and can show what changed.
- Opening: name the audience's task or decision before listing features.
- Middle: show one coherent attempt. Explain the choice that matters and what the visible result supports.
- Confidence: be direct about an observed fact; retain conditions, uncertainty and review where they affect the conclusion.
- Ending: say what a person can now decide or do, and what remains open.
Before recording, write down the one thing the audience should understand, the actual product behavior you can demonstrate and the human decision at the end. Verify Poe's behavior before drafting the demonstration as fact. Prefer a concrete, modest claim you can show to an expansive claim the footage cannot establish.
After recording, use its actual speech and pauses to time the presentation. Keep the talking-head and presentation videos separate; the presentation should not duplicate the spoken narration. The start cue, manual playback offset and ending buffer belong in the timed storyboard review. Neither this guide nor source-film timestamps substitutes for those production decisions. Script writer · Timing worksheet
What this reading covers
The reference library contains 100 uploads frozen on September 18, 2026, covering July 24 through September 18. Usable retained text was available for 25 films, and all 25 texts were read for this guide. Twelve primary readings use machine speech recognition; 13 use additional caption tracks. Eight of the 12 machine-transcribed films also have captions, making 21 captioned films in total. Those eight overlaps do not add eight more films to the coverage.
The other 75 films remain in the wider visual/storytelling library but are outside this transcript-based tone guide. Availability is uneven, so the six groupings describe this text sample; their sizes are not estimates of the channel's overall style.
Caption tracks include automatic captions and subtitle tracks; a track being labeled subtitles does not establish that a person checked every word. None of these 25 readings was verified by listening in this pass. Names, technical terms, punctuation and speaker changes can be wrong. The short work film has only a very small recognized sample. Sparse text is not proof of a speech-free soundtrack.
The notes below are original analysis, not reproduced source scripts. Passage links locate approximate spans on the official YouTube uploads. They are useful for finding the evidence, not certified word timings or edit boundaries. Interpretations are labeled as readings; fictional adaptations are labeled as examples. Existing close readings supply the separately documented visual observations.