Looking for a write-up comparing ChatGPT and NotebookLM on how they summarize a study? I decided to perform a little experiment to do just that!
As I discussed in last week’s Kitchen Sink, OpenAI commissioned a study from the National Bureau of Economic Research (NBER) and Harvard economist David Deming, which “draws on a large-scale, privacy-preserving analysis of 1.5 million conversations to track how consumer usage has evolved since ChatGPT’s launch three years ago.”
It’s a 64-page study and to say it’s a bit in-depth – well, that’s an understatement! 😉
So, I decided to perform a little experiment to not only see if I could get the highlights of it more easily, but to also try comparing ChatGPT and NotebookLM (from Google) to see how each platform summarizes the study.
First, I downloaded a PDF version of the study. Then, I loaded the study into a new notebook for NotebookLM to digest it, then created a Briefing Doc of the new notebook to get a summarized briefing. The result of that briefing titled Briefing Document: Analysis of ChatGPT Usage Patterns and User Demographics by NotebookLM, is saved to a PDF file, is here.
I then uploaded the same PDF file to GPT-5 and gave it the following prompt: “Please provide a briefing doc overview of this document of approximately 1,000 to 1,500 words.” The result of that briefing, titled Briefing Document Overview: How People Use ChatGPT by GPT-5, is saved to a PDF file, is here.
Then, I created a new chat in GPT-5, uploaded both PDF file briefings and gave it the following prompt: “Please compare these two documents and identify key differences between them in terms of findings from the study.” Here is the response from GPT-5:
I compared the two briefing documents you uploaded:
- Briefing Document: Analysis of ChatGPT Usage Patterns and User Demographics (Doc A)
- Briefing Document Overview: How People Use ChatGPT (Doc B)
Here are the key differences in findings from the study:
1. Scope and Emphasis
- Doc A (Usage Patterns & Demographics)
Goes deeper into quantitative details, breaking down percentages, message volumes, demographics, and trends across specific categories (work vs. non-work, types of activities, professions, education levels). It highlights shifts over time (e.g., non-work usage rising from 53% to 73% in one year). - Doc B (Overview: How People Use ChatGPT)
Provides a higher-level summary of the same NBER study, focusing on broad takeaways (e.g., ChatGPT as decision support, main categories of use, limited role in coding/companionship) but without the same level of numeric detail.
2. Work vs. Non-Work Usage
- Doc A:
Provides hard numbers showing non-work usage dominance:- June 2024: 53% non-work vs 47% work.
- June 2025: 73% non-work vs 27% work.
It interprets this as a shift in consumer surplus and personal applications.
- Doc B:
Notes that non-work dominates but does not give the detailed table or percentages. Instead, it frames this as evidence of ChatGPT’s expansion beyond the workplace.
3. Conversation Categories
- Doc A:
Provides a granular table of categories (Practical Guidance, Writing, Seeking Information, Technical Help, Multimedia, Self-Expression, Other) with percentage changes over time. Examples:- Writing: 36% → 24% of usage.
- Seeking Information: 14% → 24%.
- Technical Help: 12% → 5%.
- Multimedia: 2% → 7%.
- Doc B:
Mentions the same broad categories, but without detailed statistics or year-over-year change.
4. User Intent (Asking, Doing, Expressing)
- Doc A:
Introduces a new taxonomy:- Asking: 49%
- Doing: 40%
- Expressing: 11%
Notes growth in Asking and Expressing, and links Asking to higher satisfaction.
- Doc B:
Mentions that ChatGPT is as much about asking for decision support as doing tasks, but does not provide the specific breakdown or satisfaction correlation.
5. Professional and Occupational Patterns
- Doc A:
Links usage to O*NET work activities, showing 81% of work-related tasks fall into information interpretation and decision-making/problem-solving. Also breaks down usage by education and occupation (graduate-level users = 48% work-related use vs 37% for less-educated users). - Doc B:
States that writing dominates in work contexts and mentions differences by profession, but does not tie to O*NET categories or provide comparative percentages.
6. Demographics
- Doc A:
Gives detailed demographic findings:- Gender: early male skew closed; by June 2025, 52% feminine vs 48% masculine names.
- Age: 18–25 account for 46% of all adult messages; work usage highest in 36–65 group.
- Geography: strongest growth in low- and middle-income countries.
- Occupation: computer-related jobs = 57% work usage; non-professional = 40%.
- Doc B:
Mentions closing gender gap and global spread, but does not give detailed numbers or breakouts by age/education/occupation.
7. Coding and Companionship
- Doc A:
Provides precise numbers: coding = 4.2%, companionship = 1.9%. - Doc B:
Notes they are relatively small compared to media narratives, but omits percentages.
8. Economic Implications
- Doc A:
Concludes that ChatGPT is a general-purpose technology with impact on consumer welfare and workplace productivity, emphasizing decision support as the most powerful application. - Doc B:
Frames ChatGPT’s impact more generally as a co-pilot for knowledge work, but without the economic lens of consumer surplus vs productivity gains.
✅ In summary:
- Doc A is data-heavy, offering detailed tables, percentages, and longitudinal comparisons (2024 → 2025).
- Doc B is narrative-focused, providing a broad overview of the same study’s findings, with less emphasis on statistics and more on thematic conclusions.
A bit unscientific, but fun! So, when comparing ChatGPT and NotebookLM, which one is better? Depends on whether you want a data-heavy summary or a more narrative summary. A combination of both could make a really good summary!
So, what do you think? Comparing ChatGPT and NotebookLM, which one do you prefer? Read both and let me know what you think! Please share any comments you might have or if you’d like to know more about a particular topic.
Image created using Microsoft Designer, using the term “robot lawyer comparing two documents”.
Disclaimer: The views represented herein are exclusively the views of the author, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.