Recently, I embarked on a personal project with two clear and distinct objectives:
- To analyze and distill the key points from The Alchemists – J.P. Morgan’s highly anticipated 2025 investment Outlook.
- To evaluate and compare the effectiveness of three AI tools—ChatGPT, Claude, and Perplexity—in tackling this complex task.
JPM’s Outlook report, authored annually for the past 20 years by Michael Cembalest, Chairman of Market and Investment Strategy at J.P. Morgan Asset & Wealth Management, is a dense and intricate document. It is packed with economic forecasts, market analyses, and investment themes, all complemented by detailed charts, graphs, and technical terminology. For 2025, in true Cembalest fashion, the report has been cleverly titled The Alchemists.
Read the full report from J.P. Morgan, here:
To address my second objective, I utilized publicly available AI tools to evaluate their effectiveness for personal use. My goal was to experiment with input prompts and deepen my understanding of how context influences the quality and relevance of AI-generated outcomes.
Why The JPM Outlook Report Matters
As someone who has closely followed Michael Cembalest’s work since my time at J.P. Morgan’s Private Bank in 2012, I deeply admire his ability to combine original insights with contrarian thinking. Leveraging vast amounts of data (including from Chase’s extensive operational footprint) and decades of experience, he delivers analysis that is both nuanced and impactful. His foresight has often proven remarkably prescient—most notably in the 2010s, when he and Mary Callahan Erdoes boldly urged clients to significantly increase their exposure to U.S. equities. At a time dominated by widespread caution, deep pessimism, and lingering fear from the fallout of the 2008 financial crisis, this advice stood out. Those who followed their guidance are significantly wealthier today because of it.
This report is far more than a market forecast—it’s a comprehensive synthesis of economic, political, and technological trends, making it a valuable resource for informed investment decisions.
For full transparency, I do not own JPM stock (other than perhaps through an ETF), and I have no current affiliation with J.P. Morgan.
The Role of AI in the Process
To tackle this complex project and achieve my two objectives, I utilized three widely accessible and user-friendly AI tools—ChatGPT, Claude, and Perplexity—to break down the report’s intricate content. My goal went beyond simply extracting key insights; I aimed to critically assess the capabilities of these tools while exploring their potential for both personal and professional applications.
-
How effectively could they summarize, analyze, and distill such a challenging document
-
Could they serve as reliable, long-term solutions for tasks that demand synthesis and clarity?
Moreover, I sought to challenge their outputs—not only to uncover deeper insights and refine my understanding of the content but also to enhance my practical expertise with AI.
The Prompts I Used
When using AI chatbots for tasks beyond answering simple questions, a valuable tip is to craft your prompt separately before inputting it into the chatbot. I use Word and Notes, but any basic word processor will work. This approach allows you to thoughtfully structure your prompt, ensuring it includes all the necessary information and context to optimize the chatbot’s response. Taking this extra step can significantly enhance the quality and relevance of the output.
Here are the three input prompts I crafted to achieve my objectives:
First prompt – to summarize the PDF’s contents as if Cembalest himself were writing it.
“Review the attached document “The Alchemists” and summarize its contents in ten pages or less. Write the summary as if Cembalest was writing it himself and summarizing his own work. I am interested in highlights, key points, and main takeaways. Use bullet points where applicable. Include economic trends, investment themes, specific stocks, risks, and probabilities where available. Include a conclusion paragraph that can be long and wordy but concisely states the point of the report.”
Second prompt - to summarize the PDF’s contents as if an outsider were writing it, not Cembalest himself.
“Now, using the same guidelines as before, but this time come from the perspective of an outsider, not the author himself. Highlight areas that are especially distinct to Cembalest’s view, versus the consensus or accepted view. Call-out where his views are stated forcefully, or he stresses a point he feels is especially important for the reader to understand. Include a conclusion paragraph that can be long and wordy but concisely states the point of the report.”
These first two prompts were crafted to extract the insights I sought from Cembalest’s report—capturing key elements to inform my personal investment strategy for 2025.
Third and final prompt – to evaluate the AI chatbots, I employed a third and final prompt with ChatGPT.
“The attached PDF includes three summaries of JPM's Outlook for 2025 by Michael Cembalest. In the document, notice there are three summaries: one from ChatGPT, one from Claude, and another from Perplexity. Notice that I asked the same question and follow-up question prompts and consolidated the output from all three to this PDF. I want you to do this; first contrast the three summaries to see where they agree and overlap, where they disagree in their summary, and create a version of the reply to the two questions that incorporates all three replies in the PDF. Then, analyze the three replies and contrast their strengths and weaknesses; where did one chat box excel over the others, or where there were deficiencies in the outpost of one versus the others. I want to contrast the usefulness of each of the three chat bots for such a task as this one. As a reminder, the question prompts are included in the first section of the PDF, with ChatGPT, and the text is in quotes and colored in blue to distinguish the questions from the reply outputs.”
The AI Chatbot Inputs and Outputs
We aggregated the input and output material into a PDF document including:
- The Three Prompts
- Transcripts of responses from the three AI chatbots
-
Contrasting the three responses and identifying their differentiating styles
I found all the responses fascinating on many levels, including the concise presentation of details.
Engaging in a process like this sparks neural connections, igniting ideas for how to leverage AI in new and innovative ways. Frankly, this is where the highest value of the exercise lies—developing the ability to think critically about AI and identifying practical applications wherever possible.
In an era defined by hyper rates of change, the ability to adapt will be the ultimate determinant of survival and success. The real test will come in how we navigate the next two years, particularly as individuals and within small, agile groups.
If you’d like to access the PDF document with the details of prompts and responses, please refer to the link at the end of this article.
The Verdicts
- ChatGPT: Delivered a structured, balanced, and thematic summary that adheres closely to the prompts - optimal for a comprehensive summary. ChatGPT has become my go-to AI chatbot. I like it's memory and memory management features. Memory gives the benefit of past context and reference which makes a big difference over time. I think of my ChatGPT's as a SLM - Small Language Model - which brings personalization to the LLM. I address it as 'Uma' to add a sense of directness and continuity to my daily interactions.
- Claude: Added narrative flair and a creative, thematic approach, making it highly engaging. Claude's response emphasized metaphorical elements like “policy alchemy” and systemic risks. Claude is also "smart" in that it works exceptionally well with data, statistics, and charts, and in my estimation brings an intellectual element to the process. I use Claude to gather specific data and statistics and find I can quantify just about anything with a sentence or two in my prompt.
- Perplexity: Detailed, quantitative insights and actionable probabilities, appealing to data-driven decision-makers. I think of Perplexity as my search engine. I do almost all of my internet searches on Perplexity and ask it the types of questions I would normally take to a search engine on a browser. It is fast, thorough, and provides 'deeper' replies than the traditional search engine you use now.
To see more detail about the contrasting responses, refer to the PDF entitled "Contrasting the three responses and finding differentiating points."
What I Learned
I’m banking on a 10% plus correction in the broad equity markets in 2025 and preparing a clear plan to respond when it occurs. I found the concept of the Thucydides Trap fascinating and felt enlightened as I reflected on its relevance to today’s world.
More so, this project reinforced the importance of thoughtful preparation and critical thinking when using AI tools. Crafting effective prompts is not just about asking a question—it’s about structuring a request that provides the AI with the right context, details, and objectives to generate meaningful results. By experimenting with different prompts and comparing outputs across ChatGPT, Claude, and Perplexity, I gained valuable insights into how these tools interpret nuanced instructions and adapt to complex tasks.
The process also highlighted that while AI can be incredibly powerful, its effectiveness hinges on the user’s ability to guide it strategically. Taking the time to refine prompts and evaluate the results critically ensures that the outcomes are not only accurate but also actionable. This experience has enhanced my understanding of AI’s potential as a decision-making aid and reaffirmed the value of engaging with these tools as a dynamic, iterative process.
Ultimately, this exercise was not just about the answers I received—it was about how the process itself sharpened my ability to frame questions, analyze results, and leverage AI as a tool for clarity and insight. This learning will inform my ongoing use of AI, both personally and professionally, as I continue to explore its potential for solving complex problems and enhancing productivity.
Looking Forward - A Deeper Dive into AI for the Family and Family Offices
While this is an example of using widely available AI tools, it speaks to potential ahead. For example, we are building personal knowledge bases from which to interconnect, in a modular fashion, with AI tools that will use the information as a personal chatbot and for agentic functions. The goal is an autonomous system built with interconnecting modules that form an AI enabled platform that is private, secure, and highly functional in the age of AI. The system will be multi-generational and act as a family knowledge hub for many, many years to come.
One more thing: I’m following some of J.P. Morgan's advice and am anticipating a 10% or greater correction in the broad equity markets. I 'm preparing a deliberate plan of action for when it occurs.
To access the PDF with more detail and transcripts of the AI-generated responses, click below:
Thanks for reading,
Brad Deflin
from the CyberAdvisor Letter, January 2025