I’ll admit that I was hoping for a clean sweep when I did my poll of AI programs.

I’ll admit that I didn’t get one.

But the results of the “which AI program did the best job of writing out a scene” are in. When I asked you guys last week, there was one vote for each of them (God bless the three of you who voted and hats off to you for not being influenced by one another). But with each of you picking one, that left me no clearer than I was when I asked the question.

But I also did that PickFu poll of 50 anonymous people (college educated non-fiction readers)1.

So, out of Claude, Grok and Manus, which was the winner?

THE RESULTS

Shockingly (to me), it was a tie—between Claude and Grok (each got 25 votes), with Manus coming out at the bottom with 12 votes.

(FWIW, my favorite was the one done by Claude.)

Some of the comments include:

B (Claude) wins out because the wording is simple, as a child would use. Simplicity is a good thing. Look at Hemingway. Purple prose is always something to avoid.

I chose A (Grok) first because of the clarity of it. Chose B (Claude) second because it flowed and C (Manus) was just ok. I could do without it.

Then again…

C (Manus) made me feel like I was in the scene the most. It felt the most connected and real.

So what does this tell us?

Essentially, what we already know: taste is subjective. Why do I love Martin Amis’ writing while someone I know and respect worships at the altar of Glennon Doyle? Some people find Seth Rogan and Jack Black hilarious! Taste is bizarre and humans even more so.

That being said, when there’s one clear loser, that’s something to pay attention to.

So then I decided to ask Manus, Claude and Grok which they thought was best and here’s what I got:

From Claude:

For pure writing quality, Claude appears to be the strongest choice based on these comparisons, especially if you value natural-sounding, thoughtful, and well-structured content. Claude is often described as producing the most human-like writing among the major AI tools.

You could say, well, of course Claude would pick Claude but guess what Grok said?

If you prioritize polished writing with a human touch, Claude is the consensus leader based on user sentiment and reviews.

Only Manus wouldn’t give Claude the edge. It was very Switzerland-y and just talked about what Claude prioritizes (safety, accuracy, and ethical considerations), where Grok excels (at real-time information and platform integration, with fewer content restrictions, suitable for social media and marketing content and THEN why Manus is actually the most future focused (it offers “a glimpse into the future of AI agents that can independently execute complex writing tasks”). Nice one, Manus!

And so that really is, in the end, what matters. What do you want to use it for? I’m not into creating AI social media content and I’m not sure what a glimpse into the future of AI agents even means so even Manus can’t convince me Manus would be better.

In the Substack Live that Jonathan Small and I did about this topic on Friday, we both agreed on one thing: what separates great writing from not great writing is an ability to show rather than tell. While none of these programs can convert a piece of bad writing that’s all “tell” into a piece of excellent writing that’s all “show,” with multiple prompts and your own edits (and continued reminders to the LLM that it should not make anything up), you can come away with a piece of writing that’s god damn impressive.

But you need to know what god damn impressive writing is to make it into that. If you’re reading this, you clearly have excellent taste but as Carrie Fisher’s character so eloquently said in When Harry Met Sally, everybody thinks they have good taste and a sense of humor but they couldn’t possibly all have good taste and a sense of humor. So if you’re confident you know what god damn impressive writing is, these tools can be magical. If you’re not sure, use them with caution.

And my money is still on Claude.