The AI Writing Tools I Actually Use - And the Ones I’ve Quietly Uninstalled

The single best piece of AI-writing advice I can give you, after three years of actually working with every major tool on the market while trying to write fiction, essays, and client copy that does not sound algorithmically assembled, is this: use AI for the parts of writing you already know how to do, and never for the parts you don’t. That is the whole rule. If you already know how to write a decent sentence, AI can help you write more of them faster. If you don’t know how to write a decent sentence yet, AI will help you produce a plausible-looking simulation of decent sentences, and you will never learn to write, because the tool will always be slightly better than you at the thing you haven’t mastered yet, and the surrender is easy.

The second-best piece of advice is this: almost all the “AI writing tools” sold directly to writers are worse than just using a raw frontier model (Claude, ChatGPT) with a careful prompt. The wrappers add UI, sometimes add a few writing-specific features, and charge you more money for a thin layer over an API call. I’ve tried most of them. I’ve kept three. I’ll walk you through what survived and why.

If you’re looking for the broader landscape of writing software (not just AI), including the non-AI tools I still use every day, the comprehensive rundown over on creativindie covers the full map. This piece is just the AI-specific slice of that landscape, from a working writer who cares about the output sounding like a human.

The frontier models (the actual tools)

Claude (Anthropic). My primary AI assistant for all non-generative work. Research, outlining, editing feedback, brainstorming, weird requests like “read this 4,000-word essay and tell me where the argument softens.” Claude’s voice is the least AI-patterned of the major models, which matters when I’m using it as a sounding board; its suggestions don’t bend my prose toward the generic neutral-pro register that ChatGPT reliably does. I pay for the Pro plan. I use it approximately three hours a day. It has become, honestly, part of how I work, and I think about that with some ambivalence.

ChatGPT (OpenAI). I keep an account but I use it for specific tasks: image-generation-prompt writing, quick factual lookups when I don’t want to open a browser, and occasionally for “generate 30 headline variants” brainstorming where the sheer volume matters more than any single one’s quality. ChatGPT’s writing voice is the most distinctive AI voice online right now; it has specific tics (the em-dash overuse, the “here’s the thing” pivots, the three-part argument structures, the smug-reassuring closing paragraph) and once you can see the tics, you can’t unsee them in other people’s articles, which is uncomfortable. I don’t use ChatGPT for anything that’s going directly into my own prose.

Gemini (Google). I’ve tried it. Competent for research-synthesis tasks because of the Google integration. Worse than Claude for voice. Better than Claude for current events, because it can actually search. I don’t pay for it.

Everything below here is a wrapper on one of the above.

Wrappers aimed at creative writers

Sudowrite. The best-known AI tool specifically for fiction writers. It wraps GPT (and I think now also Claude, for some features) with a UI designed around novelists: “describe” a scene to add sensory detail, “expand” a paragraph, “rewrite” with different tone, “brainstorm” plot directions. In principle this is useful. In practice, I found Sudowrite actively degraded my prose. The “describe” feature adds adjectives the way a stoned teenager adds adjectives; indiscriminately and in the wrong places. The “rewrite” feature tends toward the generic literary-fiction register that AI models default to. The pricing is steep ($19-$59/month depending on tier).

I used Sudowrite for three months and came away with pages that were technically more detailed than my drafts but had lost something specific; a rhythm I hadn’t realized I had until it was gone. I stopped paying. If you’re a plotter-type novelist who struggles with sensory detail and you’ve already established your voice, it might work for you. If you’re still developing voice, stay away. It will absorb you.

NovelCrafter. A newer competitor, browser-based, project-management features (scene cards, character sheets, world-building databases) plus AI generation. It’s the best “all-in-one” novelist tool I’ve tried, and the AI features are less aggressive than Sudowrite’s; NovelCrafter treats AI as one feature among many rather than the central value prop. I’ve seen working novelists swear by it. I found the AI output still subtly generic; my own Scrivener + Claude workflow produces better prose, so I didn’t switch.

Lex.page. A clean browser-based writing environment with a “+++” shortcut that lets you ask an AI to continue or suggest. I like Lex specifically because the AI is optional and un-pushy; you can write 20,000 words in Lex without ever invoking the AI, and it just behaves as a distraction-free writing environment. When I do invoke it (rarely), it’s usually to break past a specific stuck paragraph, and I cut most of what it suggests. I still use Lex for first drafts of essays.

Jasper. Marketing-focused, sub-par for any literary prose. If you write SEO blog content for a living and you don’t care about the output being distinctive, Jasper will save you time. If you care about voice, Jasper will destroy it. I tried it. Hated it. Canceled.

Rytr, Writesonic, Copy.ai. Same category as Jasper, varying quality, all roughly interchangeable, all focused on marketing content. I have no specific opinion about any of them individually. Don’t use them for fiction.

Editing / proofing tools

Grammarly. Good for obvious errors, bad for anything subtle. I run finished drafts through Grammarly’s free tier as a last-pass safety net (typos, basic grammar) and ignore every “tone” suggestion, every “clarity” suggestion, and every “engagement” suggestion, because those suggestions are reliably awful. Grammarly’s premium tier is, in my opinion, not worth the money; it upsells aggressively and its substantive suggestions are mostly wrong for any prose that’s trying to have a voice.

ProWritingAid. Slightly more useful than Grammarly for fiction, especially its “pacing” and “overused words” reports. Same caveat: ignore the tone/clarity advice. I do a ProWritingAid pass on long manuscripts once before final. Monthly subscription.

LanguageTool. Open-source alternative to Grammarly. Works almost as well for basic corrections. I’ve started using this instead because I’m increasingly uncomfortable with Grammarly’s data practices.

Hemingway App. Highlights complex sentences and passive voice. It’s a teaching tool, not an editor. I used it while I was learning to write, found it mildly useful, and stopped using it when I developed enough sentence-rhythm sensitivity to self-correct. If you’re early in your writing life, worth trying.

The category I think most writers miss: research tools

This is where AI actually earns its keep for me, and where almost no writing-tool review bothers to go.

Claude (again) for literature reviews, fact-checking, timeline construction, and character-background research. I can feed Claude an 80-page transcript of a historical source and ask for a specific thread of argument, and it pulls the thread accurately. This is a genuine productivity multiplier. Two hours of work becomes twenty minutes.

Perplexity. Search engine with AI summarization. For when I need current information (news, recent trends, contemporary scholarship). Perplexity links out to sources, which matters for fact-checking; I never trust the summary, I always click through. I pay for the Pro tier.

NotebookLM (Google). Let me upload research documents and ask structured questions about them. Much more limited than Claude at the research-synthesis level, but the “audio overview” feature (generates a podcast-style conversation about your uploaded docs) is genuinely useful for listening to source material while I’m walking.

Scholar.google.com (not AI, still the best academic search engine). For citations, for verifying that a fact I half-remember from a book I read in 2019 is actually in the book.

What I actually use, daily

My working AI stack, for the record:

Claude Pro: three or more hours a day, for editing feedback, research synthesis, outline generation, and weird requests
Perplexity Pro: twenty to forty minutes a day, for current-events research
Lex.page: open in a browser tab during first-draft essays
Grammarly free: final-pass typo check
ChatGPT free tier: occasional one-off tasks, generally headline brainstorming

Everything else I’ve tried and uninstalled or canceled. The writing stack I’d recommend to a working writer in 2026 is: (1) a frontier model for thinking with, (2) a clean writing environment, (3) a minimal grammar-check on the final pass, (4) research tools for whatever you need to know. That’s it. The wrappers and the specialist tools are mostly charging you for UI over features you could replicate by writing a careful prompt.

The voice problem, specifically

The deepest problem with writer-AI tools is that they push your prose toward the AI models’ training-data median. The median is, generically, well-educated neutral-American literary-magazine voice; competent, readable, mid-register, faintly pleasant, indistinguishable from eighty thousand other competent writers. This is the problem. It is not a bug. It is what LLMs do; they generate the statistically-average version of the thing you asked for.

Your voice is not the statistical average. Your voice is a specific deviation from the average, and that deviation is the entire commercial case for your work being worth reading. Any tool that nudges your prose toward the mean is, over time, destroying the thing you’re actually selling.

This is why every AI-assisted writer I know has developed a private ritual to undo the AI smoothing: a pass where they deliberately add back their own quirks, tangents, soft misspellings, weird word choices, cadence breaks. The pass is not optional. If you don’t do it, your prose will slowly become what Robin Sloan (writer, Oakland) has called “the faint gray of the median”; a tone so broadly acceptable that nobody will remember a single sentence.

Sloan’s essays on writing-with-the-machine are the clearest I’ve read on this specific problem and I recommend them unreservedly. Craig Mod’s writing on tools is similarly good; he’s spent more time thinking about the infrastructure of a contemporary working writer’s life than almost anyone online.

What’s actually worth paying for

If you’re on a budget and you can only pay for one AI writing tool, get Claude Pro. Use it for thinking. Keep your writing in a text editor. Don’t pay for any writer-specific wrapper. Write your first drafts unassisted. Use Claude to review and edit after the draft exists. Never let Claude write a paragraph you haven’t already written a worse version of. That one discipline protects your voice more than any other single rule.

If you can pay for two things, add Perplexity for research.

If you can pay for three, add Lex.page or a Substack-like writing environment if Lex doesn’t fit you.

Anything else is optional and most of it is worse than useless. The future of useful AI-for-writers is not a series of ever-cleverer wrappers. It is a small set of general-purpose tools that you learn to use carefully, ruthlessly, and with a permanent suspicion that they’re doing something to your voice when you’re not looking.

Because they are.

That’s the whole job. Use the tools. Watch the tools back. Keep writing the weird, specific, wrong-in-interesting-ways sentences that a frontier model would have smoothed out, because those sentences are the only thing you have that an LLM does not.