How to Write Quality Content with Claude, ChatGPT, and Kimi

In a recent mastermind session, a founder brought up a problem I hear constantly: his product has grown past what he can market himself, content is the bottleneck, and he's never hired a content writer before. He'd been compiling AI use cases on his own but said flatly that he doesn't trust AI to do the last mile writing. That skepticism is common, and it used to be well founded. It isn't anymore, at least not in the way most founders assume. The real issue isn't whether AI can write well. It's whether you're feeding it the right raw material and putting a human in the right place in the process.

The model that gets used to write it matters, and the differences are real

Founders in the group have started running actual side-by-side comparisons rather than picking a model on reputation, and the results are worth knowing if you're deciding where to route your content work.

I've done my own direct comparison between Opus 4.8 and ChatGPT 5 for writing, and Opus 4.8 was noticeably better. Then I compared Opus 4.8 against Claude Fable 5, and Fable 5 came out ahead of that too. I know this because I write a twenty-page research report every week using an internal tool that now runs on Fable 5, and I recently finished a 400-page book, written roughly half by me and half by Fable 5, on the topic of living a purpose-driven life. It's the best writing quality I've seen from a model so far, though I'd expect that ranking to keep shifting every couple of months as the labs leapfrog each other.

Another founder in the group runs formal evals on writing quality specifically for cold email copy, which is a narrower and more measurable task than general content, but still a useful data point. In his testing, Opus 4.8 scored first, Kimi K2.6 came in second, and Sonnet 5 came in third. Interestingly, his team's actual production default is Kimi K2.6, not the model that scored highest, because it hits a strong enough quality bar at a meaningfully lower cost when you're generating volume through a provider like Fireworks. That's worth internalizing: the best model in an eval and the right model for your production workflow aren't always the same choice once cost and volume enter the picture.

• Opus 4.8 scored highest in both informal and formal comparisons shared in the group, for general writing and for cold email copy specifically.
• Claude Fable 5 outperformed Opus 4.8 in a direct side-by-side for long-form writing, including a full nonfiction book.
• Kimi K2.6 scored competitively in formal cold-email evals and is the production default for at least one team, largely on a cost-to-quality basis at scale.
• ChatGPT 5 underperformed Opus 4.8 in a direct comparison for writing quality, though it wasn't tested head-to-head against Fable 5 or Kimi in this group.

The workflow that actually works: human guided, AI written

The model comparison matters, but it's the second half of the conversation that actually changes outcomes: what you feed the model in the first place. The founders getting the best results aren't asking AI to generate content out of thin air. They're using a structure where the human does the parts a model genuinely can't do, and the AI does the parts where it's now faster and, in a lot of cases, better than a human working alone.

The value of talking to actual customers and collecting real examples with real, specific detail is very high, and no model can manufacture that for you. A content hire's most valuable work is often exactly that: interviewing customers, digesting existing case material, and pulling out the specific stories, numbers, and language that make a piece of content credible instead of generic. Once that person has a one-paragraph outline in a shared doc describing what the piece should cover and which real example or story it should be built around, handing that outline to a strong writing model can shorten their workload by roughly half and double how much finished content they can put out in the same amount of time.

The condition that makes this work is having someone who actually understands the product and the customer review the outline before the AI ever touches it, and then do a real final edit after. Skip either of those steps and quality drops fast. Keep both in place and the AI can genuinely outperform what an unaided human writer would produce on a tight deadline, especially on volume, and increasingly on quality too.

Building the input pipeline: specialized stories and case study outlines

Several members described versions of the same underlying system: a pipeline for capturing the raw material of good content before AI ever gets involved. One vendor in the group described running agents that scan competitor content, industry threads, and search behavior on a recurring basis, cluster it into topics organized by where a buyer sits in their journey (informational questions, comparison shopping, ready to buy), and surface content gaps nobody has claimed yet. From there, a human decides which pillar topics and supporting pieces are worth building, and only then does a model draft the actual piece, whether that's a blog post, a LinkedIn post, or a landing page.

The detail worth underlining in that setup is where the human sits in the loop. It's not at the start, brainstorming ideas from scratch, and it's not skipped entirely either. Humans review and fact-check at the end of the process, after a first draft exists, rather than trying to originate everything themselves. That mirrors what other members in the group described independently: humans are best used to gather the true, specific, differentiated material (real customer conversations, case study details, competitive positioning) and to do quality control on the output, while the model handles turning a good outline into finished prose.

• Collect the raw material first. Customer interviews, case study details, and specific outcomes are the input that makes AI-written content credible instead of generic.
• Write a one-paragraph outline, not a brief. A short, specific description of what the piece covers and which real example anchors it is enough for a strong model to work from.
• Have an expert review the outline before drafting. Catching a wrong angle at the outline stage is far cheaper than catching it after a full draft.
• Put the human at the end, not just the beginning. Final fact-check and edit is where a person's judgment adds the most value once a draft exists.

Building your buyer persona so the content actually lands

One founder made a point in this same conversation that's easy to skip past but shouldn't be: none of this works well if you haven't first done the unglamorous work of defining your buyer persona in real detail. Understand their triggers, how they search on Google and other search engines, and increasingly, how they phrase questions to ChatGPT, Claude, and other AI assistants directly. Content built around a real persona compounds; content built without one just generates volume that doesn't convert.

The same founder framed content as something that has to become an asset rather than an expense, meaning you need to actually track how many leads and how much revenue come from what you publish, through proper landing pages and funnels, rather than treating content spend as a black box. Without that tracking, you can't tell whether a given piece, format, or writer is actually working, and you end up unable to double down on what's succeeding or cut what isn't.

What this means for hiring

This shift changes what you should actually be hiring for if content is your bottleneck. The founder who opened this conversation said it plainly once the discussion wrapped: he's looking for someone AI-capable, not someone who's just going to write. That's the right instinct. The job isn't disappearing, but its center of gravity is moving away from staring at a blank page and toward gathering real material, structuring it well, directing a model with a clear outline, and doing a sharp final edit.

If you're hiring your first content person right now, look for someone comfortable interviewing customers and digesting a messy pile of internal material, someone who can write a tight, specific outline rather than a vague brief, and someone who treats AI-assisted drafting as a normal part of the job rather than something to resist or something to lean on entirely without editing. The founders in this group who've gotten the most out of Claude, ChatGPT, and Kimi are the ones who built a real pipeline around the model, not the ones who just typed a prompt and hoped for the best.