Wednesday I got an email from You.com, one of the coolest AI tools I haven’t talked to y’all about yet. It was inviting me to be on the waiting list for its newest AI model, ARI, which stands for Advanced Research and Insights.
This AI engine searches up to 400 sources to create a custom report with visuals, charts and clickable citations. In grad school this kind of work took me 2 years. And ARI says it’ll finish it in 5 minutes.
As a writer who has spent a kabillion hours researching articles, books and theses, I’m both thrilled and bitter at the same time.
Deep Research Experiment One: Family Tree
You.com moved me through the waiting list right away, and I received confirmation of my acceptance late last night. So my first experiment was to research my last name: Ziesenis.
Within minutes, I had a 13-page report called “Ziesenis Family History, Geneaology, and Current Locations.”
It was awesome, but it didn’t mention ME, so I was a little miffed. I inquired about “modern achievements of the Ziesenis family” and “notable modern figures of the Ziesenis family.” Whew. I finally made the list.
Each additional question created a new report, complete with its own custom cover, table of contents, citation list and more.
My final prompt asked ARI to combine all of them into one document.
I’m on pages 8, 13 and 15 (and maybe more).
Deep Research Experiment Two: AI Training Data
I realize the story of my family is captivating, but I thought I’d try a more methodical approach with a more relevant topic. ARI is brand new from You.com, and it joins brand new deep research offerings from ChatGPT, Google Gemini and Perplexity as well.
I prompted all 4 platforms with this:
Conduct a thorough analysis of AI training programs in the workplace by examining scientific studies and credible sources. Focus on how companies are implementing these programs, the skills being taught, and the measurable impacts on employee performance and job satisfaction.
All of the platforms took several minutes to go out and research their little hearts out for final reports that ranged from 5 to 17 pages (you can read them all here). Here’s what happened:
| Platform | Final Product Format | Sources | Time to Produce | Comments |
|---|---|---|---|---|
| You.com ARI | Polished 16-page report with front cover, TOC, back cover and 2-column layout | 328 sources read and 71 citations | 3-5 minutes | Looks very fancy and professional. |
| ChatGPT Deep Research | Just a plain old ChatGPT response that I had to cut and paste into a 17-page Word doc | 24 sources but no separate citation list | ChatGPT said it took 6 minutes, but I clocked 10+ for the final report | Readable content but definitely not what I’d consider a “report.” The only one to ask clarifying questions before starting |
| Perplexity Deep Research | Just a plain old Perplexity response, but it had a button to turn it into a 5-page PDF | 66 sources read and 12 citations | Maybe 3 minutes | Nice and concise with the added citation list. |
| Google Gemini 1.5 Pro with Deep Research | Created the report in a sidebar from the regular chat then gave a button to open as a 9-page Google Doc | 75 sources read and 22 citations | Told me to “come back later” and took about 6 minutes total | Nice summary with citations. |
Other Insights
I found it interesting that all of the models showed their work as they were doing the research. You could see how they broke down the steps and evaluated the sources. You could also see them change direction and go out and get more info when necessary. If it wasn’t so cool, it would be downright creepy.
