Curiosity and a good experiment go hand in hand. I wanted to test just how deep (and accurate) ChatGPT and Perplexity AI’s Deep Research outputs really were. While I was at it, I also wanted a refresher on Prime Minister Justin Trudeau’s time in office.
To this end I wrote a decent enough prompt and settled into an afternoon of waiting for the Liberal leadership vote results while poking around in AI.
The prompt:
My goal: I want to understand and then to share my understanding of PM Justin Trudeau’s time in office
Response format: For each achievement and failure provide a:
Clear title
Short summary
Bullet point details
Warnings: Keep this as objective as possible. Just the facts. Be sure to compare to outcomes achieved by other leaders in G7 countries.
Formating: Do not use title case, use sentence case for titles and headings. Do not use emojis.
Tone and reading level: This needs to be easy to understand with any acronyms or jargon explained. The tone should be friendly and on the casual side.
Include his entire tenure and across all areas of politics. Include a section that summarizes and compares public opinion and polling data to his actions. That could help in understanding why sentiments changed.
I saved their responses in Google Docs (which you can explore for yourself—links below). But I didn’t stop there. To take the comparison a step further, I uploaded both responses to NotebookLM and asked it to compare them. As an extra layer, I had NotebookLM generate a podcast (aka fraudcast – thank you for that Kimberly Dunn) and a timeline based on the responses. To add context for the fraudcast, I added a video about Trudeau stepping down.
I used the paid versions of all three tools. For ChatGPT, my paid version is the $30/month plan, and NotebookLM is included in my Google Workplace subscription.
Here are the links to the output form each app.
NotebookLM fraudcast (podcast)
First impressions: ChatGPT vs Perplexity
Each AI tool has its strengths, but right away, I noticed differences in the responses. ChatGPT’s output felt more structured, while Perplexity AI seemed to lean into surfacing sources and citations. Both claimed to provide deep research, but did they really?
Looking at the ChatGPT response, the length alone spoke volumes. Here are some key observations:
- I appreciated the ChatGPT layout—it was easier to read than the Perplexity output.
- Trudeau’s accomplishments and failures were easy to find in ChatGPT’s output for each area.Lees os in Perplexity.
- Perplexity had a lot of trouble with “Use sentence case for titles and headings.”
- I found Perplexity’s tone friendlier especially after I re-prompted to make the language simpler.
- ChatGPT did way better in generating a graphic showing a visual timeline of PM Trudeau’s approval ratings. While ChatGPT’s timeline graphic wasn’t perfect, it was significantly better than Perplexity’s attempt.
Accuracy and depth: did they get it right?
AI tools can sound confident, but confidence doesn’t always mean accuracy. I looked at the outputs for factual correctness, depth of insight, and how well they contextualized information.
NotebookLM’s comparison helped highlight differences. I thought it was far too kind – but maybe that’s my bias. Having it generate a podcast and a timeline added an interesting dimension. The timeline was particularly useful for visualizing key moments in Trudeau’s leadership.
To be honest, I don’t have a lot of political savvy. I try to keep up, but I’d have to really dig to determine if the details offered by either AI were 100% accurate. That said, I do pay attention and recall a lot of the bigger stories. Here again, I liked the thoroughness of ChatGPT. ChatGPT also took more time to produce the responses, while Perplexity was pretty quick.
Readability and style: how digestible was the information?
A research tool is only as useful as its ability to communicate clearly. ChatGPT tends to generate polished, easy-to-read responses, while Perplexity AI focuses on pulling in cited sources.
ChatGPT’s output was a clear winner in readability and scannability, in my opinion.
Final thoughts: which AI did it best?
Now, the big question—who won overall? I invite you to check out the Google Docs and see for yourself.
That said, my own takeaway was… ChatGPT for the win.
Ultimately, AI research tools are evolving quickly, and what’s “best” depends on what you need. If you’re serious about AI-assisted research, I encourage you to try the same experiment using a variety of AI models and apps, with a topic of your own, and see what works best for you.
If you run your own tests, I’d love to hear how it goes on LinkedIn.