Claude for writers: a 90-day honest review
I've been using Claude daily for the better part of a year. Not experimenting with it — using it, across several publishing projects running simultaneously. Newsletters, blog posts, product documentation, editorial planning. It's embedded in how I work in a way that very few tools become.
That's the context for this review. Not a weekend test. Not a feature comparison copied from the product page. Ninety days of daily use, across real work, with real deadlines.
Here's what I found.
What Claude actually is
Claude is an AI assistant made by Anthropic. You give it instructions — write this, edit that, help me think through this problem — and it responds. The Pro tier, which costs £17/month, gives you access to the more capable models and significantly higher usage limits than the free version.
There are now several capable AI assistants on the market. Claude, ChatGPT, Gemini, and a handful of others. They're all broadly similar in that they can write, edit, summarise, and reason. They differ significantly in how well they do each of these things, and more importantly, in how they feel to use over time.
I'll compare Claude to ChatGPT specifically later in this review, since that's the most common alternative. The short version is that I switched from ChatGPT to Claude about a year ago and haven't looked back.
What I use it for
Across my publishing projects, I use Claude for the following:
Long-form drafting
The primary use case. I give Claude a detailed brief — topic, angle, target reader, tone, structure — and it produces a draft. Not a final draft; I edit everything. But a solid working draft that I can shape, which is significantly faster than starting from a blank page.
Claude handles long briefs better than any other tool I've used. You can give it 500 words of instructions and it will follow them. Most AI tools start losing the thread of complex instructions past a certain length. Claude doesn't, or at least it does so much less than the alternatives.
Voice consistency
This is underrated and it's where Claude genuinely separates itself. If you give Claude examples of your writing and ask it to match the style, it does so with more accuracy and consistency than ChatGPT manages. For anyone maintaining a distinct voice across multiple publications — which I am — this matters enormously.
It's not perfect. Claude will occasionally produce a sentence that sounds slightly off. But the hit rate is high enough that I trust it with first drafts in a way I don't trust other tools.
Structural thinking
I use Claude to help plan pieces before writing them. Give it a topic and ask it to propose five different angles, or to outline a long-form piece, or to identify what's missing from a draft. It's good at this — better, I think, than it is at pure prose generation.
Editing
I'll paste a draft and ask Claude to tighten it, flag inconsistencies, or rewrite a section that isn't working. It's a useful second pass, though I wouldn't rely on it exclusively — it sometimes removes things that were deliberate.
What works well
Beyond the specific use cases above, a few things stand out from ninety days of daily use.
It follows instructions precisely
This sounds basic but it's rare. Most AI tools follow the first instruction in a prompt and drift from the rest. Claude tends to hold the whole brief in mind. If you say 'write this in a dry, first-person voice, avoid bullet points, keep it under 800 words, and don't mention X,' it will generally do all four. That reliability compounds over time — you start trusting it with more complex tasks because it earns that trust.
It pushes back usefully
Claude will occasionally tell you it thinks your approach is wrong, or flag an ambiguity in your instructions rather than guessing. I find this useful. It's not the tool that blindly produces whatever you asked for regardless of whether the ask was sensible. Whether you find this welcome or annoying will depend on what you're after.
The context window is large
Claude can hold a lot of text in memory within a single conversation. For long-form work — drafting a full article, reviewing a long document — this matters. You can paste substantial amounts of material and ask it to work with the whole thing.
What frustrates
The knowledge cutoff
Claude's training data has a cutoff date, which means anything that happened recently it either doesn't know about or gets wrong. For a blog about AI tools — a space that moves fast — this is a genuine limitation. I supplement Claude with Perplexity for anything requiring current information.
Occasional over-caution
Claude will sometimes decline to do things or add unnecessary caveats to content that doesn't require them. This has improved significantly over the past year and it's not a daily frustration, but it happens. When it does, rephrasing the request usually resolves it.
Prose can be too clean
Claude's default register is competent and clear, which is mostly what you want. But it can produce prose that's slightly too polished — lacking the rough edges and specific detail that make writing feel human. The fix is to give it examples of writing you like and ask it to match them. But it requires that extra step.
Claude vs ChatGPT: the honest comparison
This is the question most people actually want answered. I've used both seriously and here's my genuine assessment.
ChatGPT is more capable at tasks requiring current information, because it has web browsing built into the standard interface. If you regularly need up-to-date data in your writing workflow, that's a real advantage.
Claude is better at following complex instructions, maintaining voice consistency, and handling long documents. For writers who produce a lot of long-form content with specific style requirements, these qualities matter more than web browsing.
I switched to Claude and haven't switched back. That's not a tribal preference — it's because for my specific use case, long-form publishing with consistent voice requirements, Claude outperforms ChatGPT consistently enough that it's become the default.
A full head-to-head comparison is coming. For now: if you write a lot and care about voice, try Claude. If you need current information constantly, you'll want ChatGPT or a Claude-plus-Perplexity combination.
Related: My current AI stack including Perplexity: https://thepracticalai.digitalpress.blog/my-ai-stack-2026/
Pricing: is Pro worth it?
Claude has a free tier. It's usable, but if you're using it seriously — more than a handful of queries a day — you'll hit the limits quickly. The free tier is good for experimenting. It's not enough for a daily writing workflow.
Pro costs £17/month. For the usage level I described above — daily drafting across multiple projects — it's necessary rather than optional. Whether it's worth £17 depends on what you're producing with it. If you're writing content that generates income, the maths are straightforward: one good post that drives traffic or sales covers a month's subscription several times over.
If you're writing occasionally and experimentally, the free tier is probably fine.
The verdict
Reviews of the other tools in my stack are coming over the next several weeks. Next up: whether Notion AI is worth the add-on cost, and a full head-to-head between Claude and ChatGPT.
About The Practical AI and how reviews work: https://thepracticalai.digitalpress.blog/about/
— Ellis
This post contains affiliate links to Claude Pro. I pay for this subscription myself. Affiliate status does not affect the verdict. Full disclosure at here