How to use generative AI to scour the web for best practices
By Robert Buday and Francis Hintermann
Quantitative surveys are the dominant tool of thought leadership researchers today. Yet the most influential management ideas of the last four decades – ones that overturned conventional wisdom on business strategy, the use of digital technology, managing workforces, and other issues — came not from a series of yes-or-no or multiple-choice questions, but from a more challenging endeavor: qualitative research.
What’s more, this qualitative research was a certain type: case study interviews that explored why some companies excelled and others failed at the issue at hand. Many groundbreaking business ideas emerged only after dozens of interviews with people willing to go on the record. Determining what led to “best practice” and “worst practice” was the method behind Clayton Christensen’s disruptive innovation, Michael Hammer’s and CSC Index’s business reengineering, Tom Davenport’s and Jeanne Harris’ competing on analytics), Jim Collins’ bestselling books “Built to Last” and “Good to Great” and W. Chan Kim and Renee Mauborgne’s “Blue Ocean Strategy.” Exhibit 1 shows the vast numbers of case studies behind four of these blockbuster concepts.
Yet despite the substantial value of revealing case study research, most B2B thought leadership researchers rely on surveys. They usually administer them online, asking hundreds or thousands of managers to answer multiple-choice, rating, ranking and other questions by clicking on their best-fit response. The fixation on survey statistics is entirely understandable from a time, cost and access perspective. Researchers can field these surveys over a few weeks. What’s more, respondents can stay anonymous, which makes it easier for researchers to get their opinions.
In contrast, case study research is much harder to conduct. Lining up just a half-dozen interviews can be extraordinarily difficult. Even when the researchers work in a firm with hundreds of clients, getting a handful to speak can be arduous. Those who work with clients can block access.
For these reasons, the vast majority of thought leadership research is based on surveys. If used at all, case examples become anecdotes providing color to assertions drawn from survey statistics. Using a cake analogy, survey findings are the layers of the cake, and the case anecdotes are the icing. (See Exhibit 2.)
But if your goal is to publish groundbreaking insights that attract a multitude of clients, you need research that does the exact opposite: Your case study-based findings must become the cake and your survey analysis the icing. This requires shifting from survey-centered to case study-centered thought leadership research.
We use the term “centered” purposefully here. We don’t mean case study-only research with no surveys or desk research. We do mean that most of the insights will come from the case study findings, especially after determining what the most successful companies (the best-practice cases) did differently than the least successful ones on the topic.
Case study-centered research may sound impossible, but it no longer is, for two reasons: The first is an explosion in online information revealing bits and pieces on corporate initiatives. The second is generative AI tools that can find those far-flung pieces and put them together into highly revealing real case studies. With these two developments, thought leadership research is entering a whole new era that will make survey-centered research processes grossly insufficient.
Large language models can unearth far more best practices than is humanly possible
This era will handsomely reward researchers who embrace generative AI to scale case study research, by bringing dozens if not hundreds of case studies to the table. And it will deemphasize the surveys that have become the primary tool in their toolset.
In this article, we’ll explain why thought leadership researchers now need to use LLMs to scale case study research, and how they can do so. We’ll illustrate our method with LLM-based case research that we’ve done on the global newspaper industry.
For thought leadership researchers who want to develop groundbreaking ideas but hesitate to use LLM-generated case studies, we’ll also discuss why the terms “primary” and “secondary” research are no longer useful. The reason: Researchers may view LLM-based case studies as another form of secondary research, and thus not worthy of making a major research stream. We disagree, believing LLM output is a new form of primary research – as long as researchers verify the findings that LLMs produce.
The Limits of Surveys in Thought Leadership Research
Surveys are a fixture of thought leadership research. They have helped B2B companies show the extent of a problem they want to illuminate — whether it’s the lack of ROI on a technology, a shortage of certain workplace skills or some other malady that they address. As the recently deceased case study research expert Dr. Robert K. Yin wrote in his classic book “Case Study Research and Applications,” surveys are a research tool to answer “who, what, where, how many and how much” questions. But, he added, they’re not good at answering “why” and “how.”
From many years of experience conducting thought leadership research, we feel close-ended surveys alone are not the optimal way to develop game-changing insights on better ways to solve a business problem. To be sure, surveys can identify important market trends: the number of companies using generative AI in marketing, or how “engaged” employees are in their jobs, for example.
But B2B companies need to use thought leadership to show off their expertise in solving client problems – whether that expertise resides in a consulting, IT services, law, architecture, financial services or another provider of knowledge. That, in turn, requires their researchers to go beyond trend identification and opinion tracking. Their studies must shed new light on effective solutions to specific client problems. Superior insights on these problems and their solutions are what separates the best thought leadership researchers from the rest.
Game-changing insights are the Holy Grail of thought leadership research. They change conventional thinking on a problem and the best way to solve it. Case study interviews can do that – if exceptional interviewers talk with people who are extremely knowledgeable about how their company solved a specific business problem. Using Yin’s framework, case study research in thought leadership should explain why and how certain organizations have been better than others at solving the problem.
Surveys can’t do that. They can’t deeply explain why some companies succeed and others fail on the topic at hand. By design, surveys with closed-ended questions can only yield superficial insights.
Not sure about that? As an example, consider a survey on Fortune 500 generative AI initiatives. One survey question might try to quantify how many companies have positive ROI, with a question such as “How successful has your company’s generative AI initiatives been on a scale of 1-5?” An average of 3.2 would indicate moderate success across companies. But asking those who said “5” a series of open-ended questions that probed why they were successful would not likely produce revealing answers to each of those questions, in part because the next logical question to ask would depend on the answer to the previous one.
Plumbing the depths of corporate successes and failures – getting to the “why” and the “how,” as Yin put it – is at the heart of game-changing thought leadership research. That, in turn, requires interviewing people involved in a certain type of initiative in dozens and sometimes hundreds of companies. This is what in-depth case study research is about.
While this hasn’t been possible at most B2B companies, we believe the means of doing it has changed dramatically. It’s not any easier to get executives to talk with researchers about initiatives that worked at their companies. However, it’s far easier to get access to what they have said publicly about those initiatives.
Two developments make this possible: a) the enormous and growing well of Web information on corporate initiatives that executives have been spewing in bits and pieces, and b) the emergence of generative AI tools to weave those bits together into coherent case studies.
Let’s look at these two developments.
The AI Alternative to Survey-Centered Thought Leadership Research
The first development has been 30 years in the making. The second one emerged just four years ago:
- A 33-year explosion in web content. Ever since 1993, when the first commercial website (Global Network Navigator) went live on the World Wide Web, more than 1.4 billion sites have followed, according to Internet security firm Netcraft While only 200 million sites are still active, it still amounts to a Big Bang event for corporate content available for the public to see. Unrealized by many people, some of this content has been revealing bits and pieces about corporate initiatives – XYZ Company’s supply chain transformation in 2002-03, ABC Company’s HR overhaul in 2018-19, etc. We liken these bit-by-bit revelations to the pieces of a jigsaw puzzle, with the pieces found in such online places as blog posts, earnings call transcripts, white papers, LinkedIn articles and online books the companies have published. But these revelatory pieces have been hard to find and, even when found, have been difficult to weave together in writing a case study. It’s akin to being given a 1,000-piece jigsaw puzzle with no picture of what it looks like when finished. Yet the pieces have been piling up for years and have been there for the taking. Using another analogy, as Bob has put it, companies have been spilling their guts online for decades. But, again, collecting and assembling the pieces have been difficult, requiring dozens of hours of skilled desk researchers. As a result, thought leadership researchers largely have ignored them.
- The arrival nearly four years ago of generative AI as a superhuman case study research and writing tool. The first generative AI tool (OpenAI’s ChatGPT) was unveiled to the public in November 2022. It has been met with huge fanfare ever since, along with competitors from Microsoft (Copilot), Anthropic (Claude), Google (Gemini), Higher-Flyer (China’s owner of DeepSeek) and xAI (Grok), among others. But what has gone unnoticed is their ability to collect facts and write case studies about corporate initiatives in minutes rather than the days or weeks it would take humans to do so. Using generative AI to put together case studies has given us the confidence to say it has superhuman capability as a case research tool. Of course, thought leadership researchers must check every fact and inference the tools present, even though their output is getting more accurate.
These two developments have huge implications for thought leadership research. Case studies are the most important source of data to discover emerging best practices and create groundbreaking insights.
LLMs: Mining the Explosion of Web Content to Gather Case Studies
Getting dozens, much less hundreds, of executives to agree to be interviewed and discuss their companies’ practices has always been difficult. For thought leadership researchers, conducting best-practice case study research – and comparing them to average or worst practices — is like being asked to complete a 500-piece jigsaw puzzle, but with only, say, 10 pieces AND no picture of what the completed puzzle will look like.
But that has changed dramatically in the last four years and not because companies are more open to being interviewed by companies conducting thought leadership research. Rather, it’s because they (especially publicly held firms) have been “spilling their guts online” for more than a decade. By this, we mean they have been revealing online many more “puzzle pieces” of their business practices.
Seven sources have become vital to finding and piecing together these case study puzzles: investor meeting transcripts, thought leadership content, business podcasts, social media, online business books, open-access academic journals and online subscription newsletters. (See Exhibit 3 above.) Let’s look at each one:
- Investor meeting transcripts and archived videos: quarterly earnings calls, annual meetings, investor day presentations and securities filings (quarterly and annual reports). The number of conference call transcripts (including earnings calls, investor presentations and other events where English was spoken) rose from 2 in 2002 to nearly 30,000 in 2019, according to Capital IQ and Wharton Research Data Services. (See Exhibit 4; Capital IQ tracks about 8,000 publicly held companies.) For earnings call transcripts alone, the number tracked by FactSet rose ninefold from 2003 to 2024, from 3,017 to 27,556. Many companies discuss important initiatives in these investor meetings, and their executives’ statements about them can provide valuable clues on the status.
- Thought leadership content: corporate blogs, articles, white papers and other material that many companies publish to show off their expertise to clients, shareholders, influencers and investment analysts. For blogs alone, the percentage of Fortune 500 companies with public-facing blogs rose nearly 10-fold, from 8% in 2007 to 77% by 2020, according to the University of Massachusetts Dartmouth Center for Marketing Research (Exhibit 5). (The research ended in 2020.) In 2020, 99% of the Fortune 500 were using LinkedIn as a dissemination tool. Nine out of 10 B2B U.S. companies that are using AI to create thought leadership content said that they no longer need capable writers to publish readable material, according to a poll this year by the Content Marketing Institute and MarketingProfs. The floodgates are open.
- Online video and audio podcast conversations: podcast and other interviews of executives about their businesses. On Apple Podcasts alone, the number of podcasts on business management topics has grown nearly 12 times in the last seven years, from 2,644 in May 2019 to 32,248 in July 2026, according to Podcast Industry Insights by Daniel J. Lewis. (Exhibit 6.)
- Social media posts, especially longer ones (e.g., LinkedIn articles, Medium.com articles). In 2014, LinkedIn enabled about 230 million of its English-speaking members to post long-form content, averaging 50,000 posts a week. By 2017, the number of weekly articles more than doubled, to greater than 100,000. LinkedIn hasn’t posted updates on those numbers since. However, there is little doubt that they have increased given that LinkedIn membership has increased more than fivefold since 2014, to 1.3 billion. LinkedIn also said that the number of posts rose 14% between 2025 and 2026. Are LLMs using this content to train their models? LinkedIn (whose owner is Microsoft) indicates that it does (except for members who opt out), but it doesn’t say which models. Medium, which launched in 2012 as a place for authors to post articles, said that by 2022 more than 18 million articles had been published on its site, for an average of 1.8 million articles annually over those 10 years. In February 2024, Medium said 4 million new articles were posted that month alone, which on an annualized basis would be 28+ million that year.
- Online business books (especially those authored by CEOs): We believe books by CEOs rather than those who reported to them are more likely to reveal key details of initiatives (especially successful ones) that happened at a company. In many cases, lower-level executives might have to get permission from their CEOs to write such books. However, statistics on CEO-authored books are hard to find. In business books overall, book tracker Bowker says 18,105 business and economic book titles were published in the U.S. in 2025, compared with 4,068 business titles in 2000 (a narrower category than the one in 2025). From our experience, CEO books have become more prevalent – e.g., Pfizer CEO Albert Bourla’s book on how the drug company developed the world’s first Covid vaccine in 2020 to be approved by Western healthcare regulators.
- Academic management journals: Management journals such as Harvard Business Review, MIT Sloan Management Review, California Management Review and Entrepreneur & Innovation Exchange (an academic site that Buday TLP edits) have been publishing articles about corporate initiatives for years. Many authors are executives at companies; others are consultants or professors who studied the companies. But regardless of the author, some of these articles contain bits and pieces on corporate initiatives that LLMs can use to assemble and write corporate case studies. The number of open-access academic journals on business and management increased dramatically between 2004 and 2020, from 1 to 427, according to the Scopus service of journal publisher Elsevier (Exhibit 7). (Open access means available free to the public.) This does not include closed-access academic journals on business management and accounting, open only to paid subscribers, which numbered more than 1,300 in 2025. Since an estimated 50% of academic articles are behind paywalls, this apparently would limit LLM’s access to their content. However, some academic publishers have licensed their content to LLMs.
- Online subscription newsletters: Companies for years have published articles and other content for years in email newsletters, using such platforms as Constant Contact and Mailchimp. In more recent years, publishing and subscription services such as Substack, beehiiv and Ghost let authors make money from their writings, handling payments and other chores. Their growth in subscribers has been dramatic. For example, Substack claims the number of paid subscribers to its authors’ newsletters increased 20-fold between 2020 and 2025, from 250,000 to 5 million. Motivated by the financial incentive, many authors no doubt are tapping the expertise they’ve built over the years. Case in point: Former General Electric CEO Jeff Immelt has had a Substack newsletter called “The Long View” since May 2026. Joining CEOs and former CEOs in writing newsletters are a growing number of other C-suite executives. Mike Fisher, former CTO of Etsy is one. (His Substack newsletter is “Fish Food for Thought”.) Another is Sol Rashidi’s Substack newsletter, “The Sol of AI.” Rashidi draws on her years as chief analytics officer at Amazon Web Services, Estee Lauder, Merck, Sony Music and Royal Caribbean. Current and former executives who have been codifying and spreading their expertise through their subscription newsletters are giving the generative AI tools another deep well of insights to draw upon.
Generative AI tools have vacuumed up billions of pieces of content from these seven primary and other sources on the web. Following well-worded prompts, LLMs in minutes can find the right pieces on certain corporate initiatives and write them as highly readable case studies. (Try this out with your favorite generative tool. For example, write “Tell me why Pfizer’s Covid vaccine was the first and most effective one to come to market” and see what appears.)
Of course, the multitude of hallucinations and other errors associated with LLMs still means that case study researchers must verify all information collected and analysis drawn from them. That will take time. However, from our personal experience, the additional fact-checking hours most likely will be far less than the human time saved by having the LLMs do the fact-gathering and case study writing.
The best thought leadership researchers in one industry – IT services – appear to have noticed the capabilities of LLMs. A 2025 study by Buday TLP and Curious Insights of thought leadership programs in this global industry found that the most common use of generative AI was in gathering and analyzing secondary research, including collecting case studies. Some 47% of the best IT services firms said they use LLMs this way. A much smaller percentage (31%) said they use LLMs to write prose for thought leadership content. (See Exhibit 8.)
This is a new opportunity for every B2B company: to collect dozens or perhaps hundreds of case studies through LLMs. As such, it will require a radical rethinking of the research process. Case examples no longer need to be confined to anecdotes dredged up at the end of the analysis process to spruce up superficial, dry and often obvious insights drawn from survey statistics.
Instead, with LLM-generated case studies now available in abundance at the beginning of the process (in data gathering), researchers can use them to drive what issues they survey and where they go for even deeper insights on the case studies they’ve collected.
Here’s our take on how the thought leadership research process should be revised.
Rethinking the Stages of Thought Leadership Research
We believe that generative AI-based case study research is so important that we recommend it be used in the first stage of data gathering, with surveys and real interviews coming in the second stage. The result will be a sharpening of the problem studied, and a narrowing of the inquiry on the best solutions. (See Exhibit 9.)
When the survey questionnaire is developed after AI gathers case studies, the questions can be more surgical. Take the hypothetical study topic mentioned before: how Fortune 500 companies are capitalizing on generative AI. It would be a huge finding if LLMs found that companies with the biggest gains from generative AI all used data from service call transcripts, customer product reviews on third-party websites, and social media complaints. The survey component of this study should then devote a higher-than-initially believed number of questions about these types of data. For example, what types of such unstructured data are most useful, on a scale of 1 to 5? And which product review sites are most important?
As we stated earlier, this is where surveys can be highly useful: at quantifying previously underappreciated practices unearthed from numerous qualitative case examples. But, again, in using this example, we wouldn’t expect a survey alone to unearth certain types of data as a key success factor. One reason is the people taking such a survey may not use the same term in referring to these types of data. Since a survey’s questions and answers must be short for participants to quickly get through it, those questions and answers often resort to verbiage that may be unfamiliar to many. For example, you could call the previously mentioned data “unstructured,” but that might not capture all of it. A survey’s unfamiliar taxonomy may go over your survey respondents’ heads.
A key role of case research, both generative AI-based and interview-based, should be creating the right taxonomies – categories that make sense to most or all survey respondents.
What’s more, a survey can also point you to best- and worst-practice companies that should, and are willing to be, interviewed. In the past, each of our organizations have fielded surveys that asked participants for follow-on phone interviews. While typically only a very small minority have said “yes,” they nonetheless have been valuable interviews.
In addition to fielding a quantitative survey, the second phase of data gathering should include additional case study research – especially interviews with best- and worst-practice companies. (See Exhibit 10.) In this stage, the research team should also reach out to companies they find as “best practice” from their extensive generative AI-developed case studies. In case study research, there is no substitute for interviewing people in companies, some of which excel and others that trail on some common aspect of business. Moreover, interviewers should benefit immensely from having LLM-written case studies in hand before they talk to people in those companies.
What about using the LLMs to analyze all this, both “quant” and “qual” data? The jury is out. Consider the ecommerce site Shopify. It’s been using AI tools to create case studies, but not for thought leadership research purposes. Those case studies are customer success stories. The company built an AI agent called Matchie to research and write customer stories.
However, generating insights about how customers should sell through Shopify remains a human task. “I’ve learned AI can automate everything but this, which is truly the most important part: seeing the … trends … among all these conversations,” said Genesis Miranda Longo, Shopify’s head of customer and industry marketing, in a podcast interview (Marketing Beyond, with Alan Hart).
Sidebar: Using AI Agents to Reveal Best Practices in the Global Newspaper Industry
Newspaper Industry Best Practices: Trying Out a Generative AI-Centered Case Study Approach to TL Research
To illustrate our “qual before quant” approach to thought leadership research, we conducted the first phase of a study on one of Bob’s favorite topics: the future of the newspaper industry. (Bob worked at two newspapers and two trade publications at the beginning of his career, before entering the thought leadership profession in 1987.) We used OpenAI’s ChatGPT and Google’s Gemini for less than an hour recently to conduct the first three steps of the thought leadership research process we described earlier.
Before we set these two LLMs loose, we formulated an initial set of hypotheses and issues to explore, based on our knowledge of the newspaper industry, and on a study on college newspapers, whose financial fortunes have seen a decline that’s very similar to the local U.S. newspaper industry. link:
Formulating initial hypotheses is a critical early step in research design. They help researchers increase the odds of exploring new territory – if those hypotheses buck the findings of previous research on a topic and drive the questions that they will ask. Our initial set of hypotheses were these:
- Newspaper companies that focused on print died; those that focused on their online editions survived.
- Newspapers that largely offered content that readers could get for free on other sites did poorly, while those made readers pay for proprietary information succeeded.
- Those that couldn’t tailor online advertising to specific audiences lost significant revenue to social media, search and other online sites that could.
- Buyouts that piled debt onto newspaper companies became burdensome and prevented or curbed necessary investments in online innovations.
Had we embarked on this study without the benefit of generative AI tools, these would have been the hypotheses that guided most of our inquiry into the best practices in the global newspaper industry in staying financially solvent. However, knowing that LLMs could rapidly gather information on best financial practices in the industry, we wanted LLM-based case study research to push our thinking on these hypotheses. Specifically, we asked these two LLM’s the same three questions/prompts:
- “What have been the 10 most financially successful newspaper companies in the world – and the 10 least financially success ones – since the year 2000 (‘financially successful’ defined as growth in revenue, profitability and readership)?”
- “Write 500-word case studies on each of those 20 newspaper companies (the most and least successful ones you found), explaining the practices that led to financial success or decline.”
- “What five management practices did the 10 most financially successful newspaper companies have in common, and what five management practices did the 10 least financially successful newspaper companies have in common? And in 200 words, how did the five most important management practices of the most financially successful ones compare to the five most important management practices of the least financially successful ones?”
We won’t provide the entire output; it’s too long for this paper. And we won’t reveal the names of the least financially successful newspaper companies to avoid embarrassing them. But we will show the highlights. More important, we explain how the AI-generated case study research changed our thinking on the initial hypotheses and research questions to pursue.
On the 10 most and 10 least financially successful newspaper companies:
ChatGPT and Gemini agreed on only four of the 10 most financially successful newspapers, and on only five of the 10 least financially successful ones. What’s more, ChatGPT had one newspaper company as a laggard, and Gemini had the same company as a leader. (Exhibit 11.)
What did the LLMs tell us about the key practices of what they claimed were the most and least financially successful newspaper companies since 2000? In short, a lot that we didn’t think about in our first set of hypotheses, especially these points:
- The most successful newspaper companies aggressively diversified into digital non-news content (mentioned by Gemini’s output to our prompts). Our first initial hypothesis – about focusing on their online editions – was not specific enough. It didn’t include non-news digital content.
- In our second initial hypotheses, we said the leaders made readers pay for proprietary information. ChatGPT was more specific about this, pointing to such information as elite analysis, global business intelligence, premium national coverage, and “political identity.” (The last three are somewhat unclear to us.)
- The leaders prioritized reader revenue over ad revenue (Gemini). ChatGPT said something similar (the leaders “built direct reader-revenue businesses … central – not secondary – to the business model”). We missed this completely in our four initial hypotheses.
- The leaders sold off print edition assets (including office buildings and presses) to fund acquisitions of digital companies (Gemini), and the laggards borrowed billions of dollars to buy out other print newspaper chains before the Great Recession of 2008-09. We somewhat missed this (saying only that those that focused on their print editions died, and that buyouts piled on debt that curbed digital innovations). Gemini’s best practice gave us more depth. ChatGPT’s answer was helpful here as well. It said the laggards were often newspaper chains that sold out to consolidators, which cut costs that weakened the product (layoffs that resulted in less news coverage, etc.).
One more item to mention: Because interviewing best-practice companies is such an important part of thought leadership research, we asked both ChatGPT and Gemini to give us a short list of executives from the newspaper firms they identified as leading. ChatGPT listed 15 of them, including seven it said would be the “most valuable interviews,” why it chose them, and the best practices to focus on if we were to interview them. Gemini provided 13 names, including four mentioned by ChatGPT.
In a study, this is hugely valuable information for preparing for interviews. Here’s a sample of ChatGPT’s advice:
Sidebar: Today’s Primary/Secondary Research Fallacy
Management researchers have long talked about the differences between primary and secondary research. In his classic marketing research textbook, former Georgia Tech business professor Naresh K. Malhotra defined primary data as that collected by a researcher from the source of that data, to answer the study’s questions (with or without, for example, a research panel firm that fielded a survey). Secondary data, he said, was data already collected from the source by another entity, and often for a different research purpose.
But the dichotomy between primary and secondary research reflects a bygone, pre-LLM era, and we believe it is no longer useful. It was based on getting access to the ultimate sources of data (in the case of thought leadership here, data about a company’s initiative), or gaining access through other parties that had secured the data from the source. In short, primary and second research indicated whether you had direct or indirect access to the sources of some data on the topic you studied.
Today’s explosion of information on corporate initiatives supplied by those companies on the web (in podcast interviews, executives’ subscription newsletters, blog posts, earnings call transcripts, etc.) means researchers have much more direct access to source data – if they can find it on the web. That was difficult before generative AI. Now it is easy.
Before the web, the only way to find out about a company’s HR, supply chain or other improvement initiative was by talking to its employees directly or by learning from someone else who had talked to them: journalists (writing news articles), academics who studied them (and wrote journal articles or case studies), employees who left (e.g., through expert network firms like Gerson Lehrman Group), and so on. In other words, these companies didn’t say much about their internal initiatives, and if they did, finding out what they said was difficult.
The Web and LLMs for combing the web have made the distinction between primary and secondary research far less useful. If an LLM comes across a speech from an executive about a certain initiative in his company, that “data” about the initiative was not collected as part of some other study; it was put online directly by that executive in a LinkedIn post, podcast interview or other online source. Is that primary or secondary data?
We argue it’s primary data because it comes direct from the source, unfiltered by third-party analysis. But that begs the question of whether “primary” and “secondary” research are useful categories anymore. We don’t think they are – if researchers continue to view LLM-found statements issued by a company’s executives, findable on the web and woven into case studies, as secondary research. As long as researchers verify LLM-generated “facts” and analysis, we see these AI tools as instrumental to understanding corporate practices on a topic.
Since access to source information is far easier now, perhaps a better taxonomy for research data should be “direct from the source” vs. “indirect.” Direct would include interviews with employees involved in corporate initiatives, as well as their blog posts, podcast interviews, subscription newsletters and other statements they’ve made and which can be found on the web. Indirect from the source would mean the observations came from entities outside those companies: analysts, media, professors, etc.
Sidebar: In Thought Leadership Research, LLMs Bring a Superhuman Capability
We don’t use the term “superhuman capability” lightly. However, our usage of the large language models (LLMs) in case study development dazzled us – even if the output was not 100% correct.
For example, earlier this year, we asked ChatGPT, Gemini and Claude to tell us why the digital strategy of The New York Times has been far more successful than the LA Times’ digital strategy. (We asked each LLM the same question: why The New York Times’ digital strategy since the 1990s has been far more successful in generating paid subscribers and ad revenue than The Los Angeles Times’ online moves.)
These newspapers, headquartered in America’s two largest cities, saw themselves as competitors 30 years ago with ambitions to be national newspapers, even though they were on opposite coasts. In fact, the former CEO of LA Times (Otis Chandler) wanted to surpass The New York Times as the best U.S. paper. But their fortunes have diverged greatly since then. In the last 12 years, The New York Times has become a digital star among newspaper companies and the LA Times a fading one. (See Exhibit 12.)
Researching and writing case studies about them would have taken weeks, with desk researchers spending hours collecting information, interviewing executives and deciding what all that information added up to. But thanks to LLMs, what their executives have stated publicly since then and can be found on the Web provides a wealth of information for case studies. Generative AI tools unearthed those statements from podcast interviews, conference presentations, media interviews, LinkedIn posts, quarterly earnings calls, and other online places where their executives appeared. (The New York Times has been a publicly held company since, and The LA Times stopped being one in 2000.)
Now consider another case study example: Netflix. Imagine running a study on the keys for streaming companies to making profits. You would probably quickly figure out that your team could learn a lot from Netflix, the only streaming company to consistently be profitable in its streaming incarnation. The rest of the industry has collectively lost billions of dollars this decade.
If you couldn’t get to a Netflix executive to answer this question, you could let LLMs conduct a massive search about what Netflix executives have said publicly to be the keys to profitability in streaming. For example, those LLMs would find a comment that co-founder Reed Hastings made in a 2025 a podcast interview (Shane Parrish’s “The Knowledge Project”). Parrish asked him what data was most important to Netflix in determining how much customers valued a certain show. Hastings said that the number of customers indicating a thumb’s up or down after watching a show was most important. You can hear that at the 43:20 mark.
This fact would most likely be a key piece of qualitative data for TL researchers to find in trying to discern the secrets to profitability in the $108 billion global streaming industry. But many researchers would miss it because it’s too hard to find. It was an idle comment made toward the end of a one-hour interview. It shows there’s a growing goldmine of thought leadership research data to be found online, waiting to be collected and analyzed by LLMs, and which help researchers make enormous strides in explaining why certain companies are leading in their markets.
Scaling Case Study Research: A New Era for Thought Leadership
Overall, what did we learn from our example of a newspaper industry study that would use generative AI to better inform the first three steps of thought leadership research (design, data collection and analysis? It’s that using LLMs upfront to conduct massive case study research quickly can make the ultimate design, data collection and analysis far more surgical, and thus far more likely to yield bold new insights. Using generative AI purposefully can have three primary impacts:
- Clarifying more precisely what issues to probe: narrowing “the problem” aspect of a study (i.e., the specific challenges that organizations have in dealing with the research topic at hand), as well as focusing on specific practices that matter most in solving it, and therefore which ones to zoom into (and not).
- Identifying which companies to study: determining which ones would be best- and worst-practice case examples, and having dozens or even hundreds of them to learn from– not a handful or fewer, as is typical.
- Revealing which executives to interview: identifying who is most likely to be highly knowledgeable about the initiative that a research team is investigating — from the online statements that your LLMs have unearthed. Reminder: While LLMs enable us to rapidly gather what these people have stated publicly, asking questions no one else has asked and getting answers they haven’t stated publicly before could give your research team important data no one else has, and therefore the ability to create unique insights.
We are confident that generative AI will change the game in thought leadership research. Companies that use it to tap dozens, perhaps even hundreds, of case study examples on the topics they research will gain a big advantage over companies that continue to focus on one-dimensional surveys that spew statistics and superficial insights on complex issues.
As more companies share details about their key initiatives online, and as the LLMs evolve even further and tap more sources of online information on these initiatives, the opportunities for thought leadership researchers will multiply.
It’s a new world for those who figure out how to scale qualitative case study research and shed new and necessary light on the increasingly complex issues that businesses face. Using LLMs to compile case studies from the explosion of company disclosures on the Web is a major opportunity for every thought leadership research group. This could potentially make the “n’s” of qualitative data (# of companies studied) as high as the “n’s” in quantitative survey data. It would give researchers much more time to analyze a much bigger pool of qualitative data, and thus a far greater ability to identify important patterns in “best practices” and create big insights about them.
Generative AI is both raising the playing field for thought leadership researchers and changing how the game must be played. Anyone who knows how to write LLM prompts now can churn out a fact-filled white paper now in minutes on any topic. No research or writing ability is necessary anymore to sound smart on any topic.
But at the same time, generative AI tools enable every thought leadership research to conduct massive qualitative case study research. Those who master it but also conduct additional interviews to dissect the key differences between best- and worst-practice firms stand to be the next round of innovators of game-changing management concepts.
About the Authors
Robert Buday is CEO of Buday TLP. He can be reached at bob@budaytlp.com. Francis Hintermann is global executive director of Accenture Research, the thought leadership institute of the $70 billion (FY25 revenue) management consulting and IT services company. He can be reached at Francis.hintermann@accenture.com.