How I Create McKinsey Visuals in SECONDS With AI  🔗 **Source:** [How I Create McKinsey Visuals in SECONDS With AI](https://www.youtube.com/watch?v=Xv_aYCt-Vco) > [!note]+ Document Notes > tags: [data-visualization, ai-tools, visual-communication, content-creation, thinking-tools, diagramming] > author: Vicky Zhao > source: YouTube > reference: https://www.youtube.com/watch?v=Xv_aYCt-Vco > date created: 2026-08-31 14:00 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > # 🏷️ Title: How I create McKinsey Visuals in SECONDS with AI > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [The video under analysis](https://www.youtube.com/watch?v=Xv_aYCt-Vco) — Vicky Zhao, 11:54, published 2024-11-07 - [Napkin AI](https://www.napkin.ai/) — the sole subject of the video ## 📝 Summary - **What happened.** A single-tool walkthrough of **Napkin AI**, which generates editable diagrams from text you paste in. Zhao demonstrates the core loop — paste text, click the lightning bolt beside a passage, browse generated visual options, pick one, restyle it, export — then tours icons, Spark Search for arbitrary shapes, a sketch tool with shape detection, a hidden highlighter, and collaboration features. She states plainly that the video is not sponsored, and in the same breath invites the company to collaborate. - **Why it matters.** One genuine insight sits inside an otherwise standard tool review: browsing the generated options **changed her mind about her own content**. Seeing a linear diagram of "simplify, prioritize, structure" made her realize those three steps have no required order, so she chose a non-linear visual and decided to say so when teaching. That is visualization functioning as a *thinking* aid rather than an output aid, and it is the part worth keeping. - **Key stakeholders.** Anyone who writes explanatory content and dreads making diagrams. Napkin AI, which receives roughly twelve minutes of unqualified promotion and an explicit partnership offer. Viewers pasting proprietary or student material into a free beta service. Accessibility-minded readers, since the entire output class is images. - **Risks and tradeoffs.** The video's framing is *"use it now... before they start charging for it,"* repeated three times — manufactured urgency around a beta whose pricing has since changed, which dates the central premise. She pastes her own course transcript into a third-party beta without any mention of data handling, retention, or what happens to content you feed it. And exported PNG/SVG/PDF diagrams carry no alt text by default, which the video never raises. - **Bottom line.** The tool looks genuinely useful and the thinking-aid observation is real. The video is an enthusiasm piece with one noted limitation and no examined downsides — and its title is the clearest illustration in this whole batch of Cal Newport's argument about what engagement incentives do to advice content. "McKinsey" is never said once. ## 📖 Main Content **The problem, honestly stated.** She opens from real friction: preparing material for a cohort she was running, *"it takes so long to create visuals that go with text... Gosh, this is taking me way too long. There must be a better way."* That is a genuine and widespread bottleneck — explanatory writing benefits enormously from diagrams, and diagram production is slow enough that most people skip it. **What the tool does, and the correct comparison.** Napkin AI turns text into editable diagrams. Her distinction from image models is the technically important one: *"not like ChatGPT, which can mess up the text and create this alien language. No, this actually takes what's in the text you provided."* Diffusion image models render text as approximate glyph-shaped noise; Napkin composes actual typography into vector layouts. That is a different class of tool, and she is right to separate them. It also means the output is *editable structure* rather than a flat raster, which is what makes the rest of the workflow possible. **The loop.** Paste existing text or generate some in-app. Blue lightning bolts appear beside passages; clicking one produces a set of candidate visualizations — breakdowns, cycles, funnels — with a "generate more" for further variations. Select, restyle, export. She demonstrates on her own lesson transcript about framework thinking, which is a better test than invented copy because the text was not written to be diagrammed. **The one genuinely interesting moment.** Flipping through the options, she notices something about her own material: *"Just looking through the visuals helps me better articulate the idea."* Specifically, seeing "simplify, prioritize, structure" rendered as a linear sequence made her realize the sequence is wrong — *"You can actually simplify, prioritize, and structure without any specific order"* — so she rejects the linear option and resolves to make the non-ordering explicit when teaching. This is the video's real content and it goes by in about forty seconds. The tool surfaced a structural claim latent in her prose that she had not consciously made, and browsing alternative structures let her evaluate and reject it. Diagramming as a *diagnostic on your own thinking* is a well-known idea among people who diagram professionally and is almost never taught, because most tooling makes producing one diagram expensive enough that you never see the alternatives. Cheap generation of many candidate structures is what makes the diagnostic possible. She discovers this, names it accurately, and moves on to feature tour. **The feature tour.** Style presets (she prefers hand-drawn), label editing, colors, line thickness, patterns, and saved reusable elements. Export to PNG, SVG, or PDF, with light mode, dark mode, or transparent background — she correctly flags transparency as valuable for overlaying on video. Icons attach to titles with a lightning bolt for regenerating similar ones. **Spark Search** generates arbitrary shapes on request (her example: a full moon instead of a triangle). A **sketch tool** draws connectors with shape detection that cleans a rough rectangle into a real one. A hidden **highlighter** on the shift key, with colors usable to encode meaning. Image import, inline comments, and share permissions in the familiar view/comment/edit/share pattern, plus whole-document PDF export. She notes exactly one limitation, and it is a real one: *"once you select a visual, you can't go back to select more ideas. You have to click the blue button and generate again."* That is a meaningful workflow friction given that the browsing-alternatives behavior is the tool's most valuable use. **The disclosure, which deserves credit and scrutiny.** *"This isn't sponsored, but Napkin AI, if you're watching, I love this. Let's collaborate!"* This is unusually honest — she is stating the current commercial status accurately and, in the same sentence, openly soliciting a future relationship. Most creators do one or the other silently. Naming both is better practice than the norm. It is also a disclosed conflict rather than an absent one. A creator hoping to be paid by a company has an interest in the review being favorable, and this review contains one limitation across twelve minutes. The honesty is real; so is the incentive. **The urgency framing.** Three times she presses the same point: *"use it now... before they start charging for it,"* *"I cannot believe this tool is free. I cannot believe they won't even let me pay for anything!"* and a closing repeat. Free-beta pricing is not a closing window in the way a sale is — nothing you build during the beta is preserved by having started early, and the scarcity does not benefit the viewer. It is, however, an effective engagement device. She does hedge fairly that the tool would be *"worth it when they do make it paid,"* which is a more defensible claim than the urgency itself. Two years on, this is the most dated element in the video: Napkin has since introduced paid tiers, so the premise that drives the video's urgency no longer describes reality. **What is never mentioned.** Three omissions matter. She pastes a transcript from her own paid cohort into a third-party beta and says nothing about data handling, retention, or training use — a reasonable question for anyone pasting client, student, or proprietary material. Nothing addresses the risk of building a production workflow on a free beta, which is precisely the risk that materialized. And the entire output class is images: a PNG or SVG diagram carries no alt text, so a workflow that adds visuals to explanatory content adds inaccessible content unless the author writes descriptions separately. For a tool whose purpose is aiding comprehension, that is a substantive gap. **The title.** The video is called *"How I create McKinsey Visuals in SECONDS with AI."* The word McKinsey is never spoken, no McKinsey framework is referenced, and Napkin has no connection to the firm. "McKinsey visuals" functions purely as a search-and-click term borrowing consulting prestige. It is a small thing, and it is a precise instance of the mechanism Cal Newport describes elsewhere in this same batch — advice content drifting toward what the algorithm rewards. Worth naming not as a moral failing but as evidence that the incentive analysis is correct. **Contested and uncertain.** Whether Napkin's output would pass as consulting-grade is never tested against any standard. Whether the thinking-aid benefit persists past novelty is unknown. And the entire walkthrough describes a 2024 interface in a product category that has moved considerably. ### 🗒️ Notes #### ⏱️ [00:00](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=0) The friction, and the tool - Real motivation: preparing cohort material, *"this is taking me way too long."* - Napkin AI: *"get visuals from your text."* Vector diagrams from pasted prose. - **Correct technical distinction:** unlike diffusion image models that render text as *"alien language,"* Napkin composes the actual text into layouts. Output is editable structure, not a raster. - ⚠️ Urgency framing begins: *"use it before they start charging for it."* Repeated three times across the video. #### ⏱️ [01:42](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=102) The core loop - Paste text or generate in-app. Blue lightning bolts beside passages. - Click → candidate visualizations: breakdowns, cycles, funnels. "Generate more" for further options. - Demonstrated on her own lesson transcript on framework thinking — a better test than purpose-written copy. #### ⏱️ [03:00](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=180) ⭐ The thinking-aid moment - *"Just looking through the visuals helps me better articulate the idea."* - Seeing "simplify, prioritize, structure" rendered **linearly** exposes a claim her prose made implicitly and she does not endorse: *"You can actually simplify, prioritize, and structure without any specific order."* - She rejects the linear option and decides to make the non-ordering explicit when teaching. - **The real content of the video, and it lasts ~40 seconds.** Cheap generation of many candidate structures turns diagramming into a diagnostic on your own thinking. Rarely taught, because one expensive diagram never reveals the alternatives. #### ⏱️ [04:03](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=243) Styling and export - Style presets (she prefers hand-drawn), label editing, colors, line thickness, patterns, saved reusable elements. - Export PNG / SVG / PDF; light, dark, or **transparent background** — correctly flagged as useful for overlaying on video. - Resolution options; copy to clipboard. - ⚠️ **Noted limitation:** *"once you select a visual, you can't go back to select more ideas. You have to click the blue button and generate again."* Real friction, given that browsing alternatives is the tool's best use. #### ⏱️ [05:00](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=300) The disclosure - *"This isn't sponsored, but Napkin AI, if you're watching, I love this. Let's collaborate!"* - Credit where due: states current status accurately **and** names the hoped-for future relationship. Most creators do one silently. - Also a disclosed conflict rather than an absent one — twelve minutes, one limitation, and an open partnership request. #### ⏱️ [06:40](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=400) Feature tour - **Icons:** hover a title to attach one; lightning bolt regenerates similar options. - **Spark Search:** type any shape and get it (her example: a full moon replacing a triangle). - Note in passing: *"even though it says 'Pro,' everything is free at the moment."* - **Sketch tool:** freehand connectors and arrows with shape detection — a rough rectangle snaps to a real one. - **Hidden highlighter:** hold shift; multiple colors usable to encode meaning. - Image import; dropper tool for color consistency. #### ⏱️ [10:47](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=647) Collaboration and close - Inline comments; share permissions (view / comment / edit / share); whole-document PDF export. - Closing framing worth keeping: use it to think alone *or* together — type out a concept you are not yet clear on and visualize it to find out what you mean. #### Omissions worth recording - **Data handling:** she pastes a transcript from her own paid cohort into a third-party beta. No mention of retention, training use, or confidentiality. Relevant to anyone pasting client, student, or proprietary material. - **Beta dependency:** no discussion of building a production workflow on a free beta — the exact risk that later materialized. - **Accessibility:** every output is an image. PNG and SVG carry no alt text by default, so adding visuals to explanatory content adds inaccessible content unless descriptions are written separately. Never raised, and it matters for a comprehension tool. #### ⚠️ Title/content mismatch - Title: *"How I create McKinsey Visuals in SECONDS with AI."* **"McKinsey" is never spoken.** No McKinsey framework appears; Napkin has no connection to the firm. - Pure search-and-click keyword borrowing consulting prestige — and a precise instance of the incentive-drift mechanism Cal Newport describes elsewhere in this same batch. #### Dating note - Published 2024-11-07, ~22 months before this batch. Napkin has since introduced paid tiers, so the "it's free, hurry" premise no longer holds. The interface tour is dated; the thinking-aid insight is not. ## 📚 References - [Vicky Zhao, "How I create McKinsey Visuals in SECONDS with AI," 2024-11-07](https://www.youtube.com/watch?v=Xv_aYCt-Vco) — the source under analysis - [Napkin AI](https://www.napkin.ai/) — the tool demonstrated - [Excalidraw](https://excalidraw.com/) and [Mermaid](https://mermaid.js.org/) — text-to-diagram alternatives the video does not mention; Mermaid in particular produces diagrams from plain text in a version-controllable format - [WCAG 2.2 — 1.1.1 Non-text Content](https://www.w3.org/WAI/WCAG22/Understanding/non-text-content.html) — the alt-text requirement for every diagram this workflow produces ## 🗣️ Paraphrase - **Opening.** Making visuals to accompany text takes far too long. While preparing material for an upcoming cohort, the effort felt disproportionate, and since this is the age of AI there ought to be a better way — and there is one that takes seconds and is currently free, to the point of not accepting payment. - **The tool.** Napkin AI turns text into visuals, producing professional-looking graphics from what you supply rather than the garbled text that image generators produce. No design background is required. - **Getting started.** Napkin is free. You are given a text field: paste existing writing, or generate some in-app the way you would with a chatbot. Even a basic generated passage demonstrates the mechanism. - **The interaction.** Blue lightning bolts hover beside the text. Selecting a passage or clicking a bolt makes Napkin analyze it and offer many visualization options — breakdowns, cycles, funnels — with more available on request. - **A real example.** Using a transcript from one of her own lessons on the base logic of framework thinking, the first generated visual — take a complex subject, simplify, prioritize, structure — is already good, and flipping through alternatives suggests other framings, such as comparing and contrasting memorized frameworks against understood concepts. - **A limitation.** Once a visual is chosen you cannot return to the earlier options; you must generate again. - **An unexpected benefit.** Reviewing the alternatives clarifies the idea itself. A linear rendering prompts the realization that simplifying, prioritizing and structuring need not happen in order, so a non-linear visual is chosen and the point will be made explicitly to students. - **Styling.** Styles can be changed — hand-drawn is a favourite — along with labels, colours, line thickness and patterns, and elements can be saved for reuse. Export as PNG, SVG or PDF, in light or dark mode or with a transparent background useful for placing over video, at a chosen resolution. - **Disclosure.** The video is not sponsored, though a collaboration with Napkin would be welcome. - **More features.** Backgrounds and fonts are adjustable. Icons can be attached to titles and regenerated for alternatives. Spark Search produces any requested shape, such as a full moon in place of a triangle. Labels, text sizing and colours are all editable, everything marked Pro is presently free, and a dropper supports colour consistency. - **Beyond structure.** A sketch tool draws connectors and arrows and detects shapes, tidying rough drawings. Holding shift gives a highlighter whose colours can carry meaning. Images can be imported. - **Collaboration.** Comments support feedback tracking; sharing offers the usual view, comment, edit and share permissions; the whole document can be downloaded as a PDF. - **Close.** The tool supports thinking alone or with others — writing out a concept that is not yet clear and visualizing it helps establish what is meant. Worth trying before it becomes paid, though it will be worth paying for. ## 📰 Newsletter **The best thing in this video happens in about forty seconds, and it isn't the tool.** Vicky Zhao is demoing [Napkin AI](https://www.napkin.ai/), which turns text you paste into editable diagrams. Fine, useful, and the technical distinction she draws is the right one: unlike image models that render text as *"alien language,"* Napkin composes your actual words into vector layouts. Editable structure, not a picture of a diagram. But then she pastes in her own course transcript about "framework thinking," flips through the generated options, and catches herself: *"Just looking through the visuals helps me better articulate the idea."* The tool had rendered "simplify, prioritize, structure" as a **linear sequence**. Looking at it, she realizes that ordering is wrong — those three things have no required order — and that her prose had been implying a claim she doesn't actually believe. So she picks a non-linear visual and decides to say so explicitly when she teaches it. **That is diagramming as a diagnostic on your own thinking**, and it is the part worth stealing. It almost never gets taught, for a boring structural reason: making one diagram has always been expensive enough that you never get to see the alternatives. When generating twelve candidate structures costs nothing, the alternatives become the point. The diagram you *reject* tells you what you meant. She names it accurately and then moves on to the feature tour, which is a shame, because the feature tour is the replaceable part. **Things that are genuinely handy:** transparent-background export (for overlaying on video), shape detection in the sketch tool, a highlighter hidden behind the shift key, and Spark Search for arbitrary shapes. One real limitation, which she flags: once you pick a visual you can't go back to the other options without regenerating — annoying precisely because browsing alternatives is the tool's best use. **Three things the video doesn't mention, and one of them is ours.** She pastes a transcript from her own **paid cohort** into a third-party beta and says nothing about data handling or retention. Worth a thought before you paste client work in. Nothing about the risk of building a workflow on a free beta — which is the risk that actually materialized, since Napkin has had paid tiers for a while now. The video's repeated *"use it before they start charging"* urgency has aged into a footnote. And the one that matters for us: **every output is an image.** PNG and SVG carry no alt text by default. A workflow that adds diagrams to explanatory content adds *inaccessible* content unless you write descriptions separately — for a tool whose entire purpose is aiding comprehension, that's a real gap. If you adopt this, budget the alt text as part of the work, not after it. For anything version-controlled, [Mermaid](https://mermaid.js.org/) is worth a look as a complement — plain-text diagrams that live in your repo and diff properly. Different tradeoff, and it doesn't do the browse-the-alternatives thing at all. **One last note, and I mean it kindly.** The video is titled *"How I create McKinsey Visuals in SECONDS with AI."* The word McKinsey is never spoken. No McKinsey framework appears. Napkin has nothing to do with the firm. I watched this the same afternoon as Cal Newport explaining that advice content drifts toward whatever the recommendation algorithm rewards until *"it's the algorithm specifying your content."* This is that, in miniature, in a video I otherwise liked. Not a scandal — just unusually clean evidence that his argument is right, and a reminder to read titles as bids for attention rather than as descriptions. ## ✅ Action Items #### DevRel - Developer advocacy: adopt generate-many-then-reject as a standard step when diagramming architecture. The rejected structures are the diagnostic. `#action/h1` - Education: teach diagramming as a thinking tool, not a documentation chore. This is a genuinely underserved topic. `#action/h1` - Developer experience: any diagram we ship needs alt text budgeted as part of production, not appended afterward. `#action/h1` - Developer marketing: not applicable to this source. `#action/h3` - Community management: not applicable to this source. `#action/h3` - Internal/external/cross-project: evaluate text-to-diagram tools against version-controllable alternatives like Mermaid before standardizing on any. `#action/h2` #### OSPO - Policy & Governance: write guidance on pasting internal or client material into free beta services. This video models the behavior without the caveat. `#action/h1` - License Compliance: confirm output licensing and commercial-use terms for any AI diagram tool before it enters a publishing workflow. `#action/h1` - Security Management: beta services with no stated retention policy are a data-handling question, not just a procurement one. `#action/h2` - Software Inventory: track tool dependencies that sit on free tiers — pricing changes are a foreseeable disruption. `#action/h2` - Developer Support: not applicable to this source. `#action/h3` - Community Engagement: not applicable to this source. `#action/h3` - Strategic Alignment: not applicable to this source. `#action/h3` #### Community - Maintainers: cheap diagram generation lowers the cost of good architecture documentation. Worth a try on an under-documented subsystem. `#action/h2` - Contributors: not applicable to this source. `#action/h3` - Sponsors: not applicable to this source. `#action/h3` - Ecosystem: an accessible-diagram convention — required alt text, high contrast, no color-only encoding — would be a small shared standard with real reach. `#action/h1` #### WordPress - Core: not applicable to this source. `#action/h3` - Core Performance: not applicable to this source. `#action/h3` - Core AI: not applicable to this source. `#action/h3` - Core Program Team: not applicable to this source. `#action/h3` - Mobile: not applicable to this source. `#action/h3` - Playground: not applicable to this source. `#action/h3` - Community: diagram-heavy meetup slides need alt text and contrast checks; most decks fail both. `#action/h2` - Training: **strong fit.** Learn WordPress content would benefit from more diagrams, and this makes them cheap — provided alt text is part of the lesson template. `#action/h1` - Marketing: not applicable to this source. `#action/h3` - Support: a diagram in a support answer often beats three paragraphs. Low-cost improvement. `#action/h2` - Design: **the strongest fit.** Generate-and-reject is a legitimate design-thinking method, and the design team should have a position on AI diagram tools. `#action/h1` - Documentation: handbook pages are diagram-poor because diagrams were expensive. That constraint has changed; the alt-text requirement has not. `#action/h1` - Photos: not applicable to this source. `#action/h3` - TV: transparent-background export is directly useful for video overlays in WordPress.tv content. `#action/h3` - Polyglots: text-bearing images are translation-hostile — a reason to prefer SVG or text-based diagram formats. `#action/h2` - Accessibility: **the most important item.** Publish guidance on accessible diagrams before AI-generated visuals proliferate across project materials. `#action/h1` - Meta: not applicable to this source. `#action/h3` - CLI: not applicable to this source. `#action/h3` - Hosting: not applicable to this source. `#action/h3` - Tide: not applicable to this source. `#action/h3` - Openverse: not applicable to this source. `#action/h3` - Themes: not applicable to this source. `#action/h3` - Plugins: not applicable to this source. `#action/h3` #### The WPCC - Programs: not applicable to this source. `#action/h3` - Projects: not applicable to this source. `#action/h3` - Funding experiments: not applicable to this source. `#action/h3` ### FAIR - Technical independence: text-based diagram formats (Mermaid, Graphviz) keep diagrams portable and diffable; proprietary tool output does not. Worth stating as a preference. `#action/h2` - Policies and covenants: not applicable to this source. `#action/h3` - Productized governance artifacts: governance diagrams should be maintainable in plain text so they can be reviewed and versioned. `#action/h2` ## 🧪 Hypotheses 1. **Browsing multiple generated structures surfaces implicit structural claims that the author did not intend.** Demonstrated once, on herself, and the most interesting claim in the video. `#atom/literature/h1` 2. **AI-generated diagrams increase the total volume of inaccessible content, because alt text is not part of the generation loop.** Near-certain in direction; the size is measurable. `#atom/literature/h1` 3. **Free-beta urgency framing ("use it before they charge") increases adoption without improving user outcomes.** Nothing about early adoption is preserved when pricing changes. 4. **Titles containing prestige-brand keywords absent from the content outperform accurate titles on click-through.** This video is a clean instance, and it is the mechanism Newport describes. 5. **Diagram-as-diagnostic benefits decay after the novelty period, as users stop browsing alternatives and accept the first option.** The selection friction she flags would accelerate this. `#atom/literature/h2` 6. **Text-based diagram formats survive tool churn better than proprietary editors, measured by diagram availability after three years.** Directly relevant given this video's own dating problem. ### 📊 Methodology - **H1 (structures surface implicit claims).** Give writers a passage of their own prose and generate 8–10 candidate structures. Interview for whether any rendering revealed a claim they had not consciously made, and whether they revised. Blind a second coder on whether the revision changed meaning. Small-n and qualitative by nature — the construct is insight, so report it as such rather than dressing it in statistics. - **H2 (accessibility debt).** Sample published content containing AI-generated diagrams; measure the proportion carrying meaningful alt text against a baseline of hand-made diagrams in comparable content. Prediction: substantially lower, because generation and description are separated in the workflow. - **H3 (urgency framing).** A/B the same tool review with and without the scarcity framing; measure signups and 30-day retention. Prediction: signups rise, retention does not. Requires a creator willing to run it, which is the practical obstacle. - **H4 (prestige keywords).** Corpus of tool-review videos coded for whether the title's brand keyword appears in the content. Compare click-through and watch-time against matched accurate titles. Confounder: channels using this tactic may differ systematically in production quality, so match within channel where possible. - **H5 (novelty decay).** Longitudinal usage instrumentation: count "generate more" clicks per session over the first 90 days. Prediction: monotonic decline toward accepting the first option. - **H6 (format durability).** Retrospective: sample diagrams published three years ago in proprietary versus text-based formats; measure how many are still viewable, editable, and updatable. Cheap, and the answer would settle a recurring argument. ### ❓ Questions - Would Napkin's output actually pass as consulting-grade? The title claims it and the video never tests it against any standard. - What happens to text pasted into the tool? Retention, training use, confidentiality — unaddressed, and she pasted paid-cohort material. - Does the thinking-aid benefit survive the selection friction she identifies? If reviewing alternatives requires regenerating, most users will stop. - How do you write alt text for a generated diagram — does the tool have the structured representation needed to draft one automatically? If so, that is the obvious missing feature. - Is the linear-vs-non-linear insight generalizable, or did it depend on her already knowing the content well? A novice might accept the wrong structure precisely because it looks authoritative. - What is the right division between generated diagrams and text-based formats like Mermaid for content that must be version-controlled and translated? - Why does the title say McKinsey? The honest answer — click-through — is the interesting one, and it validates Newport's argument from elsewhere in this batch. - How much of this 2024 walkthrough still matches the product? The method outlives the interface, but a viewer arriving today will not find these screens. ### 🔮 Future Research - **Diagramming as a diagnostic on prose.** The strongest idea here and effectively unstudied. If generating alternative structures reliably surfaces unintended implicit claims, that is a writing-and-thinking technique with reach well beyond any single tool. - **Accessible-by-default diagram generation.** The tool has the structural representation needed to draft alt text and does not. Whether generated descriptions are adequate compared to human-authored ones is testable and would matter a great deal. - **Tool churn in AI-assisted workflows.** This video is its own case study: a workflow demonstrated on a free beta that no longer exists in that form. Understanding which parts of a documented workflow survive tool change would improve how such content is written. - **Prestige-keyword titling and its effects.** Newport supplies the mechanism; this video supplies an instance. Systematic measurement of how far titles drift from content in advice media would make the argument empirical. - **Linear bias in generated visualizations.** Her rejected option imposed an ordering the content did not have. Whether generators systematically default to sequential structures is a checkable and consequential question. ### 💡 Implications - **Practice.** Generate several candidate structures for anything you are about to explain, and pay attention to the ones you reject — the rejection is the insight. Then budget alt text into the same working session, because it will not happen later. - **Policy.** Organizations need a stated position on pasting internal material into free beta tools. The behavior in this video is entirely normal and entirely unexamined, which is how policy gaps get discovered after the fact. - **Security.** Beta services with no published retention policy are a data-handling decision. That is not an argument against using them, only against using them for client, student, or proprietary content without asking. - **Education.** Diagramming deserves to be taught as thinking rather than as documentation. The cost structure that made that impractical has changed, and curricula have not caught up. - **Funding models.** Not applicable to this source, beyond the observation that free betas are customer-acquisition instruments and the urgency framing serves the vendor rather than the viewer. ### 🧠 Learning Styles Assessment - **Reading level estimate:** spoken, informal; Flesch-Kincaid Grade Level roughly 7–8, Flesch Reading Ease around 68–75. Short sentences, plain vocabulary, high enthusiasm. The most accessible video in this batch by reading level. #### 👀 Visual learners - Present: excellent, as it must be. Continuous screen recording of every feature, with the generated visuals themselves as the subject matter. - Improve: no summary of the workflow, and no side-by-side of a hand-made diagram against a generated one — which is the comparison a viewer actually wants before adopting the tool. #### 🔊 Aural learners - Present: energetic, clear narration. - Improve: the content is irreducibly visual; an audio-only listener gets almost nothing. Naming what each feature does *before* demonstrating it would salvage some of it. #### 🗣️ Verbal learners - Present: the *"alien language"* description of image-model text output is vivid and technically accurate. - Improve: "professional-looking" and "McKinsey visuals" are never defined against any standard, so the quality claim is unfalsifiable. #### 🧩 Kinesthetic learners - Present: strong — a complete click-by-click walkthrough on a free tool, immediately reproducible. - Improve: no sample text provided, and no suggested first exercise. The obvious one — paste something you have already written and look for a structure you did not intend — is the video's own best idea and is never assigned. #### 🧠 Logical learners - Present: the technical distinction from image generators is correct and well-reasoned. - Improve: the quality claim rests entirely on assertion, and the urgency argument does not survive examination — early adoption of a free beta confers nothing when pricing changes. #### 🧑‍🤝‍🧑 Social learners - Present: collaboration features are covered — comments, share permissions — and the closing frames the tool for thinking together. - Improve: no example of collaborative use, and no discussion of how a team would keep generated diagrams consistent with each other. #### 🧘‍♀️ Solitary learners - Present: well suited; the whole workflow is solo and self-contained. - Improve: no guidance on when a diagram helps versus when prose is better. Not every idea benefits from a picture. #### 🌿 Naturalistic learners - Present: nothing environmental. - Improve: not a meaningful gap for this source; noting it rather than manufacturing a connection. ### 📊 Accessibility Assessment *Assessment of the video as published, and of the workflow it teaches.* #### Screen reader semantics - Present: accurate title relative to the tool, if not to the content. - Improve: **the workflow itself is the accessibility problem, and it is never mentioned.** Every artifact produced is an image; PNG and SVG exports carry no alt text, so a viewer who adopts this pipeline will generate inaccessible content at speed. For a tool marketed on aiding comprehension, that is a substantive omission. The video's own narration is also heavily deictic — *"here we go,"* *"click into this"* — making it unusable without sight. WCAG 2.2 AA 1.1.1 Non-text Content, 1.2.3 Audio Description. #### Keyboard navigation - Present: standard YouTube player. - Improve: every demonstrated interaction is mouse-driven — hovering for lightning bolts, dragging in the sketch tool, the shift-key highlighter. Whether Napkin is operable by keyboard is never addressed. WCAG 2.2 AA 2.1.1 Keyboard. #### Contrast and readability - Present: the generated diagrams appear to use reasonable default contrast. - Improve: the customization tour encourages free choice of colors and line thickness with no mention of contrast minimums, and several style presets rely on thin lines. Diagrams that encode meaning by color alone fail regardless of contrast. WCAG 2.2 AA 1.4.3 Contrast (Minimum), 1.4.11 Non-text Contrast, 1.4.1 Use of Color. #### Text resizing and reflow - Present: SVG export is genuinely the right answer here and she mentions it — vector output scales without degradation. - Improve: text embedded in an exported PNG does not resize with user settings. Recommending SVG for that reason, rather than only as a format option, would help. WCAG 2.2 AA 1.4.4 Resize Text, 1.4.10 Reflow. #### Media and captions - Present: captions available and noticeably better punctuated than the other Zhao videos in this batch. - Improve: no audio description for a video that is almost entirely screen action. WCAG 2.2 AA 1.2.5 Audio Description (Prerecorded). #### Forms and interactions - Present: not applicable to video content. - Improve: Napkin's own editor is a complex interactive canvas whose accessibility is unassessed here and would need separate evaluation before institutional adoption. #### Standards reference - Primary gap, and it is the workflow rather than the video: **1.1.1** — this pipeline produces images at speed with no alt-text step. Anyone adopting it should treat description as part of generation. Secondary: 1.4.1 (color-only encoding in customization) and 1.2.3 (deictic narration). ### 🌍 DEIB Assessment #### Cultural breadth - Present: the tool and the technique are language- and culture-agnostic in principle. - Improve: "McKinsey visuals" as the framing appeals to a specific professional prestige hierarchy — Western management consulting — and treats it as the aspirational standard without examining it. The video assumes a knowledge-worker context with slide decks and cohorts. #### Inclusive language - Present: warm, plain, encouraging. *"You don't have to be a designer"* is genuinely welcoming and lowers a real barrier. - Improve: enthusiasm markers are dense (*"I'm so excited,"* *"It's genius!"*, *"This tool is amazing"*) in a way that crowds out evaluative language a viewer could use to decide. #### Neurodiversity support - Present: consistent structure — introduce a feature, demonstrate it, move on — and diagrams themselves aid many readers who struggle with dense prose. The tool arguably serves neurodivergent readers well. - Improve: rapid pace, no chapter markers, and the feature tour has no organizing logic, so re-finding a specific capability requires scrubbing. Generated visuals can also be visually busy by default, and nothing addresses reducing that. #### Representation - Present: she demonstrates on her own real teaching material rather than invented copy. - Improve: no other voices, no user examples beyond her own, and no acknowledgement that the "professional-looking visuals" standard being borrowed is itself a narrow aesthetic. #### Psychological safety - Present: reassuring toward people who find design intimidating, which is the video's most inclusive quality. *"you don't need a huge amount of experience creating visuals"* removes a real barrier to explanatory writing. - Improve: viewers are shown how to paste their own teaching material into a third-party beta with no mention of confidentiality, and are encouraged to build a workflow on a free product with no discussion of what happens when that changes — which it since has. Both are foreseeable harms introduced without warning. ### ✅ Logical Validity #### Strong reasoning - **Correct technical distinction (00:00).** Separating Napkin from image generators on the grounds that diffusion models produce *"alien language"* where text should be, while Napkin composes the real text into layouts, is accurate and identifies the property that makes everything downstream possible: the output is editable structure rather than a raster. - **Testing on real material (01:42).** Demonstrating on her own lesson transcript rather than purpose-written copy is a harder and more honest test, because the text was not composed with diagramming in mind. - **The thinking-aid observation (03:00).** The strongest reasoning in the video, and it arrives as genuine discovery rather than a scripted point. Seeing a linear rendering exposes an ordering claim her prose implied and she does not endorse, so she rejects it and resolves to make the non-ordering explicit. Generation → comparison → rejection → revised understanding is a complete and valid inferential loop. - **Naming a real limitation (01:42).** *"once you select a visual, you can't go back to select more ideas"* is a friction that specifically undermines the tool's best use, and reporting it against her own enthusiasm is creditable. - **Explicit non-sponsorship disclosure (05:00).** Stating the commercial status accurately, and naming the hoped-for future relationship in the same breath, is better disclosure practice than the norm in this genre. #### Fallacies - **Title/content mismatch** — the title promises *"McKinsey Visuals"* and the word McKinsey is never spoken; no McKinsey framework appears and Napkin has no connection to the firm. Prestige borrowing for click-through, and a textbook instance of the incentive drift Newport describes elsewhere in this same batch. **Fix:** name the tool in the title. `#fallacy/loaded-language` - **Manufactured scarcity** — 00:00 and repeated twice more: *"use it now... before they start charging for it."* Free-beta pricing is not a closing window; nothing about adopting early is preserved when pricing changes, so the urgency serves adoption rather than the viewer. She partly redeems this by conceding the tool would be *"worth it when they do make it paid,"* which is the more defensible claim. `#fallacy/false-urgency` - **Unfalsifiable quality claim** — throughout: *"professional-looking"* and the title's consulting-grade implication are never tested against any standard, comparison, or rubric. The claim cannot be checked, and the video's central promise rests on it. **Fix:** show a hand-made diagram beside a generated one. `#fallacy/unsubstantiated-claim` - **Suppressed evidence — data handling** — 01:42: she pastes a transcript from her own paid cohort into a third-party beta with no mention of retention, training use, or confidentiality. Viewers are shown the behavior and not the question. `#fallacy/suppressed-evidence` - **Suppressed evidence — accessibility** — throughout: the workflow's entire output class is images without alt text, so adopting it generates inaccessible content at speed. Unmentioned, and materially relevant for a tool whose stated purpose is aiding comprehension. `#fallacy/suppressed-evidence` - **Enthusiasm substituting for evaluation** — throughout: twelve minutes containing one limitation, delivered in a register (*"I'm so excited,"* *"It's genius!"*, *"This tool is amazing"*) that supplies affect where assessment belongs. Compounded by the open partnership solicitation, which gives a disclosed interest in the review being favorable. `#fallacy/appeal-to-emotion` #### Tone of voice Unreservedly enthusiastic — closer to a friend showing you something they just found than to a review. The register is warm, fast, and exclamatory, and the delivery is genuinely infectious. Its virtue is accessibility: *"You don't have to be a designer"* lowers a real barrier, and the excitement communicates that the tool is pleasant to use, which is information of a kind. Its cost is that the tone leaves no room for assessment. Across twelve minutes there is one limitation and no discussion of tradeoffs, alternatives, data handling, or durability. The register is uniform whether she is describing an export format or pasting proprietary teaching material into a beta service, so nothing signals which decisions carry consequences. The partnership solicitation — honest as it is — sits inside that uniform enthusiasm rather than prompting any additional care. The most instructive tonal fact is the gap between the title and the voice. The video's actual register is a delighted personal recommendation; its title is optimized consulting-prestige bait. Those were produced by different pressures, and the seam between them is visible. ### 🛡️ DARVO Analysis DARVO describes an accused party's response pattern — Deny, Attack, Reverse Victim and Offender. **No DARVO pattern is present.** This is an enthusiastic tool review with no allegation, accuser, or defensive posture. Component by component: - **Deny:** Absent. Nothing is alleged and nothing denied. The one statement with denial's grammatical shape — *"This isn't sponsored"* — is a voluntary disclosure, not a rebuttal, and she immediately adds the countervailing fact by soliciting a partnership. Volunteering the interest rather than concealing it is the opposite of the DARVO move. She is also willing to state a limitation against her own enthusiasm. - **Attack:** Absent. No person or company is disparaged. ChatGPT's image generation is criticized on a specific technical point — text rendered as *"alien language"* — which is an accurate description of a known limitation of diffusion models, not an attack on a competitor's character. No other tool is even mentioned, which is itself a gap (Mermaid and Excalidraw go unnamed) but not hostility. - **Reverse victim/offender:** Absent. No grievance is claimed anywhere. The framing is entirely positive-sum: she found something useful and wants viewers to have it. **Emotional cues and rhetoric.** High enthusiasm throughout, with dense excitement markers. The persuasive devices in play are manufactured urgency around the free tier and prestige borrowing in the title — commercial rhetoric aimed at adoption and click-through, not defensive rhetoric. **Related fallacies.** The real weaknesses are captured above: `#fallacy/false-urgency` on the scarcity framing, `#fallacy/loaded-language` on the title, and two `#fallacy/suppressed-evidence` findings on data handling and accessibility. These are disclosure and evaluation gaps in good-faith content with no adversary — a different and less serious class than DARVO, worth stating plainly. `#darvo/deny` `#darvo/attack` `#darvo/reverse` — all three assessed. None present. ## 🎯 Conclusion - **Bottom line.** A warm, competent tool review of Napkin AI containing one idea worth far more than the tour around it. When she flips through generated structures and realizes a linear rendering imposes an ordering her content does not have — *"You can actually simplify, prioritize, and structure without any specific order"* — she is doing something rarely taught: **using generated alternatives as a diagnostic on her own thinking**. The diagram she rejects tells her what she actually meant. That works because cheap generation removes the cost that historically let you see only one structure. Everything else is a feature tour with a single noted limitation, delivered at uniform enthusiasm, and three omissions that matter: she pastes paid-cohort material into a third-party beta with no word on data handling; she urges adoption of a free beta three times without addressing what happens when pricing changes, which it since has; and the workflow's entire output class is images with no alt text, which for a comprehension tool is a real gap. The title promises McKinsey visuals and the word is never spoken — a small, clean instance of exactly the incentive drift Cal Newport describes elsewhere in this batch. - **Next steps for viewers.** 1. Steal the technique, not the tool: for anything you are about to explain, generate several candidate structures and pay attention to the one you **reject**. That rejection is the insight. 2. Budget alt text into the same session as diagram generation. It will not happen afterward, and this pipeline produces images faster than descriptions. 3. Decide before you paste. Client, student, or proprietary material into a beta service with no published retention policy is a choice worth making deliberately. 4. For anything version-controlled or translated, evaluate [Mermaid](https://mermaid.js.org/) alongside this — plain-text diagrams diff, review, and survive tool churn. Different tradeoff, and the video mentions no alternatives at all. 5. Discount the urgency. Free-beta scarcity confers nothing on early adopters, and this specific window closed years ago. 6. Read the title as a bid for attention rather than a description — here and generally. It is the cheapest habit available and this video is a good place to practice it. ## 🔄 Refactor Appendix — How I create McKinsey Visuals in SECONDS with AI ### 📚 Literature Atoms (H3) #### [H1] Generated alternatives expose structural claims your prose made implicitly > [!quote] ID:rw_napkin_h1 > "Just looking through the visuals helps me better articulate the idea." **Context & use**: - ~03:00 — a linear rendering of "simplify, prioritize, structure" reveals an ordering she does not endorse; she rejects it and resolves to teach the non-ordering explicitly. - Diagramming as a diagnostic on thinking. Works only because generating many candidates is now cheap — one expensive diagram never shows you the alternatives. **Tags**: #xref #articleseed #aphorisms **Highlight**: [view highlight](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=180) <!-- atomicity: 1 idea ✓ --> #### [H2] Composed text differs in kind from generated text-shaped pixels > [!quote] ID:rw_napkin_h2 > "not like ChatGPT, which can mess up the text and create this alien language" **Context & use**: - 00:00 — the technical distinction that makes the whole workflow possible. - Diffusion models render glyph-shaped noise; Napkin composes real typography into vector layouts, so the output is editable structure rather than a raster. Correct, and the reason the tool is worth separating from image generators. **Tags**: #xref **Highlight**: [view highlight](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=0) <!-- atomicity: 1 idea ✓ --> #### [H3] Free-beta urgency serves adoption, not the viewer > [!quote] ID:rw_napkin_h3 > "use it now. Now that I found it, use it before they start charging for it." **Context & use**: - 00:00, repeated twice more. - Nothing about early adoption is preserved when pricing changes — and it since has, which dates the video's central premise. A clean specimen of manufactured scarcity in tool-review content. **Tags**: #xref #addendum **Highlight**: [view highlight](https://www.youtube.com/watch?v=Xv_aYCt-Vco&t=0) <!-- atomicity: 1 idea ✓ --> ### 🧠 Permanent Candidates (H4) #### [P1] The structure you reject tells you what you meant **Claim** (one sentence, Courtney's voice): Generating several structures for the same content and rejecting the wrong ones reveals the claims your prose was making without your knowledge. **Evidence pointer**: [[#H1]] **Scope & conditions**: Holds wherever content can be structured more than one way — explanatory writing, documentation, curricula, talk outlines, information architecture. Requires that generating alternatives be cheap; when one artifact costs hours, you never see the comparison set and the diagnostic disappears. Depends on the author knowing the domain well enough to recognize a wrong structure, so it sharpens expertise rather than substituting for it — a novice may accept an incorrect structure precisely because it looks authoritative. **Links** (≥1 with annotation): - [[Diagramming as thinking]] — because extends - [[Rubber duck debugging]] — because extends the same externalize-to-see-it mechanism - [[Information architecture]] — because applies-in structuring content <!-- atomicity: 1 idea ✓ --> #### [P2] A generation workflow inherits an accessibility debt unless description is part of the loop **Claim** (one sentence, Courtney's voice): Any pipeline that produces artifacts faster than it produces their descriptions manufactures inaccessible content at the speed of the tool. **Evidence pointer**: [[#H2]] **Scope & conditions**: Holds for images, diagrams, charts, video, and audio produced by generative tools where description is a separate manual step. Sharpens as generation cost falls, because the ratio of artifacts to descriptions worsens. Does not apply where the tool retains the structured representation and can draft a description from it — which is the obvious missing feature here, since Napkin has the structure and does not use it. Practical test: is alt text produced in the same session as the artifact, or deferred? **Links** (≥1 with annotation): - [[WCAG 1.1.1]] — because applies-in the requirement - [[Accessibility debt]] — because extends - [[Automation and quality]] — because contrasts on what speed costs `proof_state: working` <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Try the generate-and-reject technique on a piece I've already written — look for a structure I implied but don't endorse. Cheapest experiment in this batch. #fleeting #inbox - [F2] Compare Napkin against Mermaid for the deep-read and audit workflows. Mermaid diffs and lives in the repo; Napkin browses alternatives. Different jobs. #fleeting #inbox - [F3] Post idea: "the structure you reject tells you what you meant" — diagramming as a diagnostic on prose. #fleeting #articleseed - [F4] Write the accessible-diagram guidance before AI visuals proliferate in WordPress project materials. Alt text, contrast, no color-only encoding. #fleeting #addendum - [F5] This title says "McKinsey" and the word never appears — put it in the incentive-structure post as the concrete example alongside Newport's argument. #fleeting #articleseed - [F6] Do diagram generators default to linear/sequential structures? If so that's a systematic bias worth knowing about. #fleeting #inbox ### → Things 3 (max 2 actions) - [ ] Run generate-and-reject on an existing draft and record whether it changed the structure — writing #things - [ ] Draft accessible-diagram guidance (alt text, contrast, color-only encoding) — accessibility #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/How I Create McKinsey Visuals in SECONDS With AI.md|How I Create McKinsey Visuals in SECONDS With AI by Vicky Zhao]] ## Highlights - create this alien language. No, this actually ([View Highlight](https://read.readwise.io/read/01m1c8xb04y4xgcn2t9w2sq4x9)) - Note: The correct technical distinction, and the reason the whole workflow is possible. Diffusion image models render text as glyph-shaped noise; Napkin composes real typography into vector layouts. That means the output is EDITABLE STRUCTURE rather than a raster — which is what makes restyling, SVG export, and browsing alternatives work at all. Right to separate it from image generators. - Tags: atom/literature/h2 - through the visuals helps me better articulate ([View Highlight](https://read.readwise.io/read/01m1c8sr25zrp7gtzz542bth0y)) - Note: The real content of the video, and it lasts about 40 seconds. Seeing "simplify, prioritize, structure" rendered LINEARLY exposed an ordering claim her own prose was making that she doesn't endorse — those three have no required order — so she rejects the linear option and decides to say so when teaching. Diagramming as a diagnostic on your own thinking. It works only because generating many candidates is now cheap: one expensive diagram never shows you the alternatives. The structure you REJECT tells you what you meant. - Tags: atom/literature/h1, atom/zettel/z1