This is a republication of Reedo’s Naver article. It is not a new experiment or a DMS client case study. The cover and in-article photographs are AI-generated staged scenes, not evidence of actual work.
If you are deciding between ChatGPT, Claude, and Aside, it is better to start with the work you want to delegate and the criteria for checking the results, rather than a ranking. This video compares four tasks using the same prompts: organizing email and schedules, creating presentation slides, identifying contract risks, and rehearsing for an evaluation.
This summary is based on the video description and subtitles reviewed on October 16, 2026. Filming took place from October 14 to 16, and the emails, contracts, and other materials were fictional. Because personalization was enabled during the comparison, these observations should not be generalized into an overall performance ranking or results for later versions.
Whichever AI you choose, first check the connected accounts, personalization settings, recheck results, and sources for numbers and quotations.
AI-generated staged image, not an actual work screen.
How Were the Matching Prompts Structured?
The prompts in the video have five parts: the task, tools to use, classification criteria, output format, and prohibited actions. For email organization, for example, the tools were asked to sort messages into urgent, reference, and ignore categories, prepare reply drafts, and identify scheduling conflicts, while being prohibited from sending messages or changing the schedule. The important point is to specify both the scope of the task and the authority to act.
All three tools classified at least 19 of the 20 fictional emails in line with the predefined criteria, and all found the overlapping appointments. However, the ChatGPT results include a rerun due to a recording problem. Since results can vary even within the same experiment, these figures should not be read as the overall accuracy of each product.
The video also explains that ChatGPT’s reply draft included information from the main account rather than the account used for filming. Although it does not establish a single cause, this clearly illustrates why connected accounts and personalization settings should be checked first. Before supplying real materials, it is safer to test only reading and draft preparation with fake emails.
AI-generated staged scene 2View original
AI-generated staged image, not an actual work screen.
What Differed in the Files and Review Results?
For the presentation task, the tools were instructed to ask about anything they did not know first. ChatGPT asked three questions, Claude four, and Aside eight. After receiving the same answers, they produced files with nine, nine, and ten slides, respectively. All three labeled estimates and items requiring confirmation, and all carried out the revision to reduce the presentation to five slides.
Aside applied saved design rules. However, Aside is not an AI model of its own; it is a tool used with models such as GPT or Claude. Both this difference in what is being compared and the influence of existing settings need to be considered. One instance of saved rules being applied should not be turned into a guarantee of consistency across all tasks.
In the fictional contract, all three found the five clauses deliberately written to disadvantage the creator. When asked to check again, they produced additional items, and Aside corrected an earlier calculation. This was an exercise in spotting risks, not legal advice. The video creator also stated that the sources of the case numbers and industry figures cited by Claude had not been verified.
AI-generated staged scene 3View original
AI-generated staged image, not an actual work screen.
What Should You Check When Applying This to Your Work?
The evaluation rehearsal involved entering the same practice answers in sequence. All three tools asked whether the figures had been measured directly. However, the video reports that Claude’s subsequent model answer included time figures absent from the notes and wording claiming that actual measurements had been taken. Asking sharp questions and grounding the final answer accurately are separate things to check.
In the observations shown in the video, ChatGPT was fast in the timed stages, Claude asked probing follow-up questions, and Aside used saved rules. However, claims such as Aside having visually inspected the presentation were self-reports from the tool; the creator had not directly observed that process. Reports of actions performed need to be distinguished from actual verification.
If you try this yourself, first choose non-sensitive fictional materials and specify prohibited actions. Second, compare outputs produced from the same requests and revision instructions. Third, recheck calculations and citations against the original sources. Recording not only how quickly the task finished but also what you still need to verify makes it easier to choose what to delegate next time.
Watch the Original Video for the Four Comparisons
Rather than choosing a winner from a single result, a practical takeaway from this video is to attach a check to each task: verify accounts for email, actually open generated files, and recheck review findings. This article summarizes observations from that video; it is not a report of a separate experiment.
