Two years into the generative content boom, the strongest brands aren't the ones that have automated everything. They're the ones that figured out which parts to automate and which parts to protect. The tool you pick matters less than knowing where to draw that line.
We write a lot of content. For ourselves, for clients, across blogs, emails, ad copy, product descriptions, and the long-form work that builds authority. AI is a part of that workflow now. It is not most of it.
Here's the honest breakdown of what's actually working — and what we're seeing fail.
The tools we keep paying for.
Claude (the long-context one).
Claude is our default for long-form drafting, brand-voice work, and anything that benefits from a model that won't get squirrelly halfway through. The 200k context window means you can drop your entire brand guide, three example pieces, and a brief into a single conversation and get output that actually sounds like you.
ChatGPT for ideation and short-form.
Faster turn-taking. Better at brainstorming. We use it for headline batches, hook variants, social copy, and the kind of throwaway 'give me twenty options' prompting where Claude's deliberate cadence is overkill.
Grammarly and Hemingway for the final pass.
Boring. Essential. Hemingway in particular is the antidote to the long-sentence, semicolon-laden style that LLMs default to. If your content reads like an LLM wrote it, run it through Hemingway and rewrite anything red.
The tools we use selectively.
- Jasper — useful only with a tightly tuned brand voice profile and a rigorous human edit. Without those, you ship beige.
- Copy.ai — fine for short-form ad copy when you need volume.
- Writer.com — strong for enterprise teams that need governance, terminology controls, and audit logs more than they need creative output.
- Notion AI — useful for summarization and meeting-note cleanup, weak as a writing partner.
Where AI fails — and we let it.
There are categories of writing where we deliberately don't use AI. Not because we couldn't, but because the cost of the model getting it 80% right is higher than the time saved.
- Founder voice and thought leadership. If a CEO is going to put their name on a 1500-word essay, the ideas have to come from them. AI is in the room — as a transcription tool, as an editor, never as the author.
- Anything with legal or regulatory weight. Healthcare claims, financial advice, terms and conditions. Cite a hallucination once and the cleanup costs more than every word AI ever saved you.
- Customer stories and case studies. People can tell. The texture of a real interview survives every layer of editing in a way generated content does not.
The workflow that ships.
A piece of content that we'd publish under our name or a client's name typically goes through this loop:
- Brief written by a human — including the angle, the audience, the specific outcome we're after, and the things we don't want to say.
- Research synthesized by AI — competitive pages, source material, transcripts. AI is good at compression. Use it.
- Draft written collaboratively — usually a writer plus Claude, in a back-and-forth where the human is steering and the model is fetching the next paragraph.
- Edit by a different human — line edits, voice checks, fact checks. This step is non-negotiable.
- Polish pass with Hemingway and Grammarly — to strip the LLM tells.
- Ship.
The thing nobody says out loud.
Generative content has a smell. Readers can pick it up even when they can't articulate what it is. The cadence is too even. The transitions are too smooth. Every paragraph carries the same weight.
Good writing has rhythm. It surprises you. It earns the reader's attention sentence by sentence. AI can scaffold that — it can't manufacture it.
Use AI like an espresso machine. Pull a shot when you need one. Don't drown the cup in milk.