Dan Shipper and guest David Perell test ChatGPT's ability to analyze a business model, using the New York Times as a case study. When initial answers come across as generic, Perell prompts the model to respond through the lenses of influential strategists Ben Thompson, Clayton Christensen, and Michael Porter, comparing where their frameworks agree and disagree. # Video Identity Observation: "ChatGPT Can Analyze Any Business In Seconds" ## Explicit Self-Description Evidence - **No direct channel self-description found.** The transcript contains no statements like "on this channel we…" or "I make videos about…". - The speakers reference their own work incidentally: - > "I've written about that — the cost of retaining top talent is getting harder and harder" - > "that's what this is right — this is you moving up Market, hey we're on video and stuff like that" This suggests at least one participant creates written content and the format is video, but neither is framed as a channel description. ## Intended Audience (Observable Presentation) - **Business and investing analysts / knowledge workers using AI tools.** The speaker explicitly frames the lesson around a professional analyst use case: - > "how to use GPT really effectively as a sparring partner if you're some sort of investing analyst" - **Viewers interested in prompt engineering and business strategy**, demonstrated by: - Teaching framing techniques: > "you have to give GPT shape and Direction" - Strategy frameworks cited as shared vocabulary: Clayton Christensen's disruption theory, Porter's Five Forces, Ben Thompson's aggregation theory. ## Claimed / Demonstrated Scope - **Demonstrated scope:** Practical AI-assisted business analysis using ChatGPT, shown through a worked example analyzing the New York Times' business model (revenue, subscribers, risks, strategy). - **Claimed expertise areas (via demonstration, not channel claims):** - Business model analysis and media industry knowledge (NYT's ad-to-subscription shift, ~$2.31B revenue, 9M subscribers, per the tool output read aloud). - Prompting technique: > "GPT is magical when it comes to names — names are like the ultimate compression" ## Presentation Format (Observed) - **Conversational, two-person demo format** — one person drives the analysis, another asks clarifying questions: - > "do you feel like it answered all the questions that you have" - > "tell me like how you got that — why you want to do that versus the previous prompt" - Uses **contrast-based teaching**: showing a "bad answer" from a generic prompt, then a superior one from a name-dropping prompt: - > "this is just buzzword Bingo and this is ridiculous" - > "look at how much better this is" ## Claims vs. Demonstration Distinction - **Creator claims about method effectiveness** (e.g., that naming thinkers dramatically improves output) are presented assertions supported only by the single in-video example. - **Demonstrated patterns:** real-time ChatGPT use with Bing search commentary, referencing contemporary media figures (Chris Williamson, Steven Sinofsky), indicating topical, current-events-aware content. # Delivery Observations: "ChatGPT Can Analyze Any Business In Seconds" ## Overall Format This is a **conversational, two-person demonstration** — a live walkthrough where one speaker drives a ChatGPT session while a partner reacts, reads output, and asks questions. The delivery is informal, unscripted-feeling, and built around showing a tool in real time. ## Tone & Register - **Casual and enthusiastic**, with frequent excited interjections: "I love it," "I love that I love that," "great example," "okay interesting." - **Critical and vivid when dismissing weak output**, e.g., calling one answer "just buzzword Bingo and this is ridiculous" and "I feel like I'm at you know the business school 301 seminar just like eye Rolling On A Tuesday morning." - **Confident, declarative teaching voice** on core principles: "you have to give GPT shape and Direction you have to otherwise it won't give you a good example." ## Recurring Phrases & Verbal Habits - **"So" as a sentence opener** — extremely frequent: "so one of the other things that I found with GPT," "so the New York Times they came out with a memo," "so what I would like to know." - **"Yeah" and "right" as connective filler**: "GPT critique this business model yeah," "aggregation Theory… I mean there's no doubt right." - **Repetitive scaffolding phrases for transitions**: "so why don't we do this," "so let's get super specific and let's say," "now what I would be interested in maybe for the next question is." - **"One of the things"** appears repeatedly: "one of the other things that I found with GPT," "one of the things that people Miss," "one of the things that you should know." ## Teaching Habits - **Contrast-based instruction**: deliberately shows a bad prompt/answer first, labels it — "this is the bad answer and I don't think it's going to be that helpful" — then compares it to a better result: "look at how much better this is." - **Explicit prompt-building out loud**: narrates exactly what to ask the model — "name the number one strategy… tell me a recommendation from Clayton Christensen Ben Thompson and Michael Porter give me an answer for each one that's no longer than one paragraph." - **Explains the "why" behind techniques**, framed as abstract principles: "names are like the ultimate compression… you want to think about compression right how much can I deliver in just a few words." - **Anchors concepts in named frameworks and authorities**: Clayton Christensen, Michael Porter, Ben Thompson, "Porter's 5 forces," "aggregation Theory," "jobs to be dum framer [done framework]" (apparent transcription artifact). - **Grounds abstractions with concrete anecdotes**: the US Airways Hudson River crash, Chris Williamson's podcast, Steven Sinowski (likely "Sinofsky" — transcription artifact) on Twitter. ## Question-and-Answer Rhythm - The speakers alternate between **prompting GPT and prompting each other**: "do you feel like it answered all the questions that you have," "so you want you want to read it out for us sure." - Meta-questions about the method drive the teaching: "tell me like how you got that like why why you want to do that versus the previous prompt which was which was much more General." ## Vocabulary & Style Notes - Business/strategy jargon used fluently: "tapestry of knowledge," "vectors of strategy," "fault lines of disagreement," "Mount Rushmore of business strategists," "moving up Market." - Occasional **hedging and uncertainty markers** for facts: "I want to say it was 2015 or 2016," "I don't actually know what these are so let's look at them." ## Stable Patterns vs. Isolated Moments - **Stable (repeated)**: "so" openers, enthusiasm interjections ("I love it"), contrast-then-apply teaching structure, citing named frameworks, walking through prompt construction aloud. - **Isolated**: the "buzzword bingo / business school 301" mockery appears once as a sharp tonal spike; the "Mount Rushmore" metaphor is a one-off flourish. Uncertainty is low for the stable patterns given frequent repetition; the two-person division of roles cannot be fully attributed per speaker throughout (transcript lacks speaker labels), so individual delivery patterns are stated with that limitation. # Position Extraction: "ChatGPT Can Analyze Any Business In Seconds" ## Speaker Attribution Note The transcript contains two speakers in dialogue, but no curated metadata identifies them by name. Positions below are attributed as **Speaker A** (host/interviewer, driving the GPT demonstration) and **Speaker B** (interviewee/analyst, providing most of the commentary). Assignment is based on conversational flow only. --- ## Core Claims ### 1. Broad prompts produce poor GPT output - **Speaker A (stated):** When users ask GPT to "critique this business model" without constraints, "your answer is going to be very poor." - **Reasoning:** You must "give GPT shape and direction" — specificity is required for quality output. ### 2. Names act as "ultimate compression" in prompting - **Speaker B:** Invoking "Clayton Christensen," "Michael Porter," and "Ben Thompson" lets GPT load entire frameworks (jobs-to-be-done, disruption theory, five forces, aggregation theory) into the prompt. - **Reasoning stated:** "You want to think about compression... how much can I deliver in just a few words." - **Follow-on position:** Referencing multiple named thinkers creates a "tapestry of knowledge" — a conversation "amongst those three people" where you can find agreement, disagreement, and "the fault lines of disagreement" to explore strategy vectors. ### 3. Web-search mode changes answer quality - **Speaker B** (observation, offered neutrally): GPT "just searched used Bing to search for the answer," meaning it was "summarizing things that other people have written" rather than reasoning from its own knowledge — "that's just one thing to keep in mind." - **Speaker A** accepted the setup but flagged the resulting answer as the "bad answer" for comparison purposes. ### 4. Generic risk answers are "buzzword bingo" - **Speaker A** critiqued the first answer: "cyber security threats, disruptive innovation and digital transformation... pandemic related challenges, supply chain and vendor risk... these are just ridiculously broad... these aren't the actual risks." --- ## Case Study: New York Times (as presented by the speakers) ### Business context claimed - **Speaker A:** NYT issued a memo (~2015–2016, hedged: "I want to say") about needing to change to stay in business; the company makes money as "the authoritative source" on news, politics, and culture; the strategic shift is **from advertising revenue to subscription revenue**. - Note: Speaker A framed this partly from memory ("I want to say it was 2015 or 2016") — a stated qualification, not verified fact. ### GPT-reported figures (read back by Speaker A) - Annual revenue ~$2.31 billion, an 11% increase from $2.07 billion the prior year. - Subscription revenue ~$1.5 billion in 2022; ~9 million subscribers in 2022. - Advertising revenue ~$523 million. - Speaker B briefly noted uncertainty about whether all questions were answered; Speaker A judged it "a really good answer." ### Named-framework analysis (Speaker A's commentary on GPT's output) - **Disruption theory:** Smaller, nimbler competitors moving up market — Speaker A cited Chris Williamson's podcast as an example, then **qualified it**: "not a direct competitor." - **Porter's Five Forces:** - *New entrants:* Content production/distribution "has been democratized." - *Bargaining power of customers:* Willingness to pay may drop unless NYT offers "distinct value" (Speaker A flagged "distinct value" as undefined and worth returning to). - *Substitutes:* Rise of citizen journalism; witnesses on the ground (e.g., the US Airways Hudson River crash) can distribute information faster than media companies. Speaker A contrasted this with the late-20th-century era of privileged access to high-profile people. - *Supplier power:* "The cost of retaining top talent is getting harder" — journalists leaving to launch their own ventures ("I've written about this," Speaker A noted). - *Industry rivalry:* Speaker A: "I don't think that one is quite as interesting." - **Aggregation theory:** Risk from platform dominance by Google and Facebook aggregating news content — Speaker A: "there's no doubt." --- ## Unresolved / Tentative Points - Date of the NYT memo stated from memory only. - Speaker B openly questioned whether the first GPT answer covered all requested questions. - Web-search vs. internal-knowledge mode left as a user choice ("you may want it to do" either way). - "Distinct value" explicitly bookmarked but never defined in the transcript. ## Final Direction Stated **Speaker A** proposed a next prompt: name the **number one strategy** for NYT, with a one-paragraph recommendation from each of Christensen, Thompson, and Porter — yielding "three different perspectives." (No answer shown in transcript.) # Speaker Attribution & Provenance Record ## Video **Title:** ChatGPT Can Analyze Any Business In Seconds **Video ID:** `01a0f15b-3338-7027-a7db-e8661fbdfa80` ## Speakers - **Speaker A (host/creator, unnamed):** Drives the demonstration, sets up prompts, narrates the framing ("so one of the other things that I found with GPT..."), explains prompting strategy and the "names are compression" technique. - **Speaker B (guest/co-presenter, unnamed):** Reads GPT outputs aloud, reacts ("I love that"), asks clarifying questions ("why you want to do that versus the previous prompt"), and interjects commentary on Chris Williamson's podcast and Steven Sinofsky's writing. - No speaker is identified by name in the transcript; no curated metadata was supplied to establish identities. ## Roles Observed - **Creator/host:** Speaker A (implied by framing of the video lesson). - **Guest/co-analyst:** Speaker B. - **Narrator:** None distinct. - **Tool output (not a speaker):** ChatGPT responses read aloud by both speakers, including the revenue figures and risk analysis. ## Referenced Sources & Third-Party Claims - **New York Times internal memo** (2015 or 2016, per Speaker A's recollection) about business-model shift — details uncertain in source ("I want to say"). - **Named business strategists** invoked as prompting anchors: **Clayton Christensen** (disruption theory, jobs-to-be-done), **Michael Porter** (Five Forces), **Ben Thompson** (aggregation theory). - **Third parties mentioned in commentary:** - **Chris Williamson's podcast** — cited as a small media example "moving up market" (Speaker B). - **Steven Sinofsky** on Twitter — referenced regarding rise of citizen journalism (Speaker B). - **US Airways Hudson River crash** — anecdote about on-scene reporting beating media companies. - **GPT-retrieved data** (provenance: ChatGPT, possibly via Bing search): NYT annual revenue ~$2.31B (11% up from $2.07B), $1.5B subscription revenue, 9M subscribers, $523M advertising revenue in 2022. Speaker A explicitly notes one answer was generated via **Bing web search** rather than model knowledge. ## Promotional Segments - None identified. Self-referential mention ("this is you moving up Market, hey we're on video") is incidental, not flagged as promotion. ## Availability & Uncertainty - **Comments availability:** Not stated in source material. - **Uncertainties:** NYT memo date (2015/2016 — speaker unsure); speaker identities not established; GPT financial figures are tool output, not independently verified in the transcript. so one of the other things that I found with GPT is it's really good at looking at something like a business model and one of the things that people Miss is they'll put in their business model and they'll say GPT critique this business model yeah and what they end up doing is when you're that broad your answer is going to be very poor you have to give GPT shape and Direction you have to otherwise it won't give you a good example and let's say that you want to you know let's do the New York Times let's do the New York Times that would be really interesting cuz there'll actually be a tapestry of knowledge that it can build on yeah so the New York Times and maybe you want to write this so the New York Times they came out with a memo I want to say it was 2015 or 2016 and it's actually extremely interesting where they spoke about how they need to change as a company in order to stay in business and the New York Times they make money by being the authoritative Source on news stories around the world politics and culture and what they have needed to do is shift from advertising Revenue to subscription Revenue so what I would like to know and maybe we can ask GPT is what is their annual revenue how many readers do they have per year what are the most profitable parts of the business what are the least profitable parts of the business and then we can just ask GPT to critique the business model yeah so have that be the final sentence we asked it tell me more about how the New York Times makes money what we want to know is what their annual revenue is how many readers they have and what are the most profitable parts of the business and least profitable parts so New York Times reached an annual revenue of approximately $ 2.31 billion uh which was an 11% increase from the previous years previous year's 2.07 billion the revenue was sub from subscriptions was significant amounting to 1.5 billion in 2022 uh and they had 9 million subscribers in 2022 advertising Revenue was 523 million um and um yeah I don't know if it I did do you feel like it answered all the questions that you have that's a really good answer so why don't we do this tell me about the biggest risks in the business over the next 10 years so I just want to set this up this is the bad answer and I don't think it's going to be that helpful what I want to do is I want to look at this and then I want to compare it to the next answer and I want to set something up about how to use GPT really effectively as a sparring partner if you're some sort of investing analyst where we're going to get some really specific answers after this I love it so one of the things that you should know is it just searched used Bing to search for the answer to your question so rather than answering from its own sort of knowledge it just Googled like what are the biggest risks for media companies over the next 10 years which is something that you might want it to do but it will affect the answer because it's basically just summarizing things that other people have written directly from its first search rather than like just creating it from it from from what it knows so um that's just that's just one thing to keep in mind you we may want to redo it and have it not not look on the web but no I love this I mean this is just buzzword Bingo and this is ridiculous cyber security threats disruptive innovation and digital transformation like I feel like I'm at you know the business school 301 seminar just like eye Rolling On A Tuesday morning it's like oh my goodness pandemic related challenges supply chain and vendor risk like these are just ridiculously broad swath climate risk like what right no like these aren't the actual risks right so let's get super specific and let's say don't search the web look up the work of Clayton Christensen Michael Porter and Ben Thompson and focus on how the New York Times can continue to shift from advertising Revenue to subscription Revenue um and tell me like how you got that like why why you want to do that versus the the previous prompt which was which was much more General great question so GPT is magical when it comes to names names are like the ultimate compression and that's what you want to do whenever you're communicating something you want to think about compression right how much can I deliver in just a few words and if I say Clayton Christensen what it's going to do is it's going to look up his jobs to be dum framer it's going to look up his disruption Theory and just in two words I can subsume all those ideas into that prompt and I can do the same thing with Porter's five five forces I can do the same thing with Ben Thompson's aggregation Theory and it'll probably give me something about aggregation theory about Porter's 5 forces and now what we've done is we have that tapestry of knowledge and now we can start having a conversation amongst those three people to see where the Mount Rushmore of business strategists agree and then what do they disagree on and then the fault lines of disagreement now we can follow those and we can see what the different vectors of strategy can be I love that I love that so like so we got our answer so you want you want to read it out for us sure so I mean this is exactly it so now we have look at how much better this is so we have disruptive innovation smaller nimbler competitors who offer cheaper more accessible or innovative solutions that gradually move up market so that's what you're seeing with somebody like Chris Williamson's podcast is doing this he is a oneman show basically and he's moving up Market with better and better production quality okay interesting not a direct competitor but these small media companies that's what we're seeing there like every there we go I mean that's what this is right this is you moving up Market hey we're on video and stuff like that great example now Porters 5 forces I don't actually know what these are so let's look at them so threat of new entrance can enter the market with lower barriers especially in the digital space now that's a great answer okay so the ability to produce and distribute content has been democratized absolutely bargaining power of customers huh what's this have more options for news and content their willingness to pay might decrease unless the New York Times offers distinct value I love those words distinct value what does that mean let's index that bookmarket we'll come back threat of substitutes altern alternative sources of news and information I don't know if you saw what uh Steven sinowski wrote about Twitter a few months ago but this is what's happening we're seeing this rise of Citizen journalism and media companies can no longer compete on certain vectors so what happened in the late 20th century is a company like the New York Times they could have privileged special access to certain highle people now with Google and even just all the people who have boots on the the ground the people who are there can often distribute information faster the famous example of this was when the solely US Airways airplane crashed the people on the plane the people who were by the Hudson River they were reporting better than a lot of the media companies because they were just there sharing information bargaining supply of suppliers I mean I've written about this that the cost of retaining top talent is getting harder and harder you have people who are leaving these traditional media companies and they're going to launch their own thing right and the media companies have to do a better and better job of convincing the journalists who they work with to stay industry rivalry okay I don't think that one is quite as interesting aggregation Theory a risk lies in the potential dominance of digital platforms like Google and Facebook that aggregate news content I mean there's no doubt right so look at how much came from this we have very specific things and now it's given us it's given us some strategies now what I would be interested in maybe for the next question is name the number one strategy yeah for the New York Times to use period tell me a recommendation from Clayton Christensen Ben Thompson and Michael Porter give me an answer for each one that's no longer than one paragraph so now we're going to get a one sentence strategy we're going to get paragraph to summarize and we're going to get three different perspectives