Although I cover artificial intelligence now, earlier in my career I was a comedy writer and wrote jingles. As I was going through my Google Drive recently, I came across a folder from my very early writing days.
The old piece was fun to see, and had been used for a writing workshop I took at Harvard about 20 years ago. The story is very much a product of its time because it includes instant messaging, Clippy (that animated paper clilp) and a reference to a fax machine. But the humor still holds up because it's about workplace struggles with technology.
I figured that piece would be a good one to test ChatGPT, Claude and Gemini as writing coaches.
I wasn’t looking for the chatbot that could rewrite it most aggressively or smooth out every sentence. I wanted to know whether any of them could recognize what was already working, find the places where the story dragged and help me improve the comedy without stripping away my voice.
Here's what happened when I gave all three of them the draft.
The draft
I gave all three chatbots the same draft and the same instructions. All three offered useful advice, but one understood the story on a much deeper level.
The piece begins with what should be a very simple request from a boss: reserve the main conference room for an important meeting in an hour.
Unfortunately, the narrator has never used the email system to book a room before. What follows is an escalating workplace crisis involving a 21-year-old intern, a sleepy coworker, a phone call from the office bathroom and a woman named Sharon who brushes her teeth at her desk.
I left everything untouched for the test. That meant keeping the dated references, awkward sentences and obvious mistakes, including “starred” instead of “stared,” “the the” and “marshmellows.”
Those errors gave the chatbots something easy to catch, but proofreading wasn’t really the point. I wanted to know whether they understood why the story was funny.
The entire piece builds toward the boss casually revealing that he solved the problem by putting a sticky note on the door — the same solution the narrator rejected when the intern suggested it. A good writing coach should understand why that callback matters and avoid “improving” the story in a way that weakens it.
The prompt I used
I entered the same prompt each time: Act as my writing coach. Don’t rewrite this piece for me. Identify what is already working, where the humor or momentum drops, and the three changes that would make the biggest difference. Preserve my voice, comedic timing and the period in which the story takes place. Be specific and quote brief examples from my draft when useful.
I judged the responses on how specific the feedback was, whether each chatbot understood the comic escalation and whether its suggestions would actually make the piece better.
I also paid attention to restraint. I wanted help seeing the draft more clearly, not a chatbot taking over and writing a different story.
ChatGPT focused on timing
ChatGPT zeroed in on the rhythm of the piece and gave me some of the most precise line-level edits.
One of its strongest suggestions was to make better use of the introduction and then it told me how to better raise the stakes in the middle of the story to keep things moving. It also suggested trimming certain aspects of the story to give the dialogue more room and help the jokes land without changing what's actually happening.
It also suggested minor tweaks that involved creative elements that I may not have considered to help connect two of the strongest visual jokes in the middle of the story.
ChatGPT understood the timing of the piece and gave me edits I could use immediately. However, it spent more time polishing individual moments than examining whether every step in the story made sense.
Claude found the hole in the story
Claude approached the piece differently. It looked beyond individual lines and explained how the comedy was built.
It identified the central joke as the gap between the size of the task and the size of the narrator’s panic. Reserving a conference room should take a minute. Instead, it becomes a crisis involving several coworkers, a bathroom phone call and increasingly desperate attempts to avoid admitting that she doesn’t know what she’s doing.
Claude also understood why the seemingly random details belonged in the story. Sharon’s toothbrush isn’t there just because it’s funny that someone brushes her teeth at her desk, but it elevates another element of the story — something that I hadn't even noticed that I had subconsciously done.
More importantly, Claude spotted a problem the other chatbots missed: we never learn what the employees in the story actually do for work. The narrator moves through the story too quickly, which is something that impacts the entire story. Claude noticed it.
So while the chatbot didn’t try to decide the joke for me. It pointed out that I needed to choose. Claude worked to protect my ideas and text without interjecting it's own ideas. It seemed to understand the story and knew exactly which punchlines were holding it together.
Gemini gave the clearest advice
Gemini’s feedback was the easiest to absorb and apply.
It quickly identified the strongest moments in the story, recognized why the premise is relatable and the punchlines. Its suggestions were direct. Like ChatGPT it suggested I tighten the opening, trim some of the descriptions and expand the dialogue.
Gemini even showed me exactly where the opening could lose a few words and how the dialogue could be made punchier. I wouldn’t have had to interpret much before beginning a revision.
That clarity was Gemini’s biggest strength, but its analysis didn’t go as deep as Claude’s. It helped me tighten the story in front of me without finding the one place where the story’s logic breaks down.
The best AI writing coach
Claude gave me the best writing advice because it understood both the jokes and the structure supporting them.
ChatGPT was especially strong on rhythm and gave me the most inventive suggestion with the returning Tetris image. Gemini delivered clear advice that I could apply without overthinking it. But Claude found issues with the piece that I hadn't even noticed, which is exactly what a writing coach or editor does — and would have immediately improved the story.
Claude also gave me a choice instead of trying to write the solution itself. It showed me where the gap was, explained why it mattered and left the joke to me. That’s exactly what I want from a writing coach.
All three chatbots are certainly capable of offering useful writing advice, but they were useful in different ways.
ChatGPT was the strongest line editor, especially when it came to pacing and recurring imagery. Gemini offered the clearest path into a revision. Claude was the best at stepping back and seeing how every part of the story worked together.
The test also reinforced why I prefer asking AI to respond to my writing instead of writing it for me. When a chatbot rewrites an entire draft, it becomes difficult to separate your choices from its choices. Asking for a limited number of specific recommendations keeps you in control. You can consider the advice, reject what doesn’t fit and make the changes yourself.
The wording of the prompt helped. Telling the chatbots to preserve my voice and the period of the story discouraged them from modernizing the technology or replacing the dated details. Asking for only three changes forced them to prioritize instead of burying the useful feedback inside a long list of corrections.
I still wouldn’t accept any chatbot’s writing advice without questioning it. An AI can correctly identify a slow section and then suggest cutting the exact detail that makes the story sound like you. But if I were returning to an old draft and wanted help seeing what I could no longer see, I’d choose Claude.
Let me know in the comments what you think about using AI has a writing partner.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom's Guide
