I Tried Every AI Writing Tool — Here's What I Actually Use

Over the past year I've used Claude, ChatGPT, Gemini, Jasper, Copy.ai, Writesonic, Rytr, Sudowrite, Lex, Wordtune, Grammarly's AI features, and Notion AI for writing tasks. Not brief tests — I ran actual work through each of them. Client emails, article drafts, marketing copy, documentation, social media posts, and the kind of miscellaneous business writing that eats up two hours a day if you let it.

The AI writing tool space is crowded in a way that suggests most of these tools will not exist in two years. There are roughly three categories: the general-purpose LLMs that happen to be good at writing (Claude, ChatGPT, Gemini), the dedicated AI writing platforms that wrap an LLM in a specialized interface (Jasper, Copy.ai, Writesonic), and the writing-enhancement tools that augment your existing process (Grammarly, Wordtune, Lex). I've formed opinions about all three categories, and the opinions are not evenly distributed.

The General-Purpose LLMs: Where the Actual Writing Happens

Claude, ChatGPT, and Gemini are not "AI writing tools" in the marketing sense. They're general-purpose language models that you can use for writing, among many other things. They don't have templates for "blog post" or "product description" or "Instagram caption." You just tell them what you want and they write it.

This turns out to be the correct approach to AI-assisted writing, and here's why: templates constrain the output in ways that make it worse. When Jasper gives you a "blog post" template, it's imposing a structure — intro, three body sections, conclusion — that may or may not fit what you're actually writing. When you tell Claude "write a 1,500-word article about X, in this voice, with these specific points," you get exactly what you asked for. The specificity of your prompt replaces the rigidity of a template, and specificity produces better output than rigidity every time.

Claude is what I use for 80% of my writing work. I've covered this in the Claude vs. ChatGPT comparison, but the short version: Claude's prose is less generic, follows complex instructions more consistently, and requires less editing. When I write with Claude, the output is a draft I can edit. When I write with most other tools, the output is raw material I have to reshape.

ChatGPT handles about 15% — specifically the tasks where I need inline code execution, web search integrated into the writing process, or image generation alongside text. When I'm writing a piece that requires current data and I want the research and writing to happen in the same conversation, ChatGPT's browsing capability makes it the better choice.

Gemini gets the remaining 5%, mostly for tasks involving very long source documents. When I need to read a 200-page PDF and produce a summary, Gemini's context window handles the full document without chunking. The writing quality of Gemini's output is a tier below Claude and ChatGPT — more generic, more prone to the kind of safe, corporate phrasing that sounds like nobody — but for summarization and extraction tasks, writing quality matters less than comprehension accuracy.

The Dedicated Writing Platforms: The Problem They Solved Doesn't Exist Anymore

Jasper, Copy.ai, Writesonic, and Rytr all launched in the GPT-3 era, when getting good output from an LLM required significant prompting skill and the raw API wasn't accessible to non-technical users. These tools solved a real problem: they gave you a friendly interface, pre-built templates, and prompt engineering baked into the platform so you didn't have to learn it yourself.

That problem no longer exists. ChatGPT has a conversation interface anyone can use. Claude has a conversation interface anyone can use. The barrier to getting good writing from an LLM has dropped from "you need to understand prompt engineering and API calls" to "type what you want." The middleman value that Jasper et al. provided — making the LLM accessible — has been eliminated by the LLM providers themselves.

What these platforms offer now is essentially a wrapper around the same models you can access directly, with templates, brand voice features, and team collaboration tools added on top. The wrapper costs $49-$125/month depending on the platform and tier. Claude Pro costs $20/month. The question you should ask is whether the templates and brand voice features are worth the $29-$105 premium. For most people, the answer is no.

I tested Jasper extensively because it's the most established of the group. The brand voice feature — where you train the tool on your existing content and it mimics your style — is the headline differentiator. In practice, it's roughly as effective as giving Claude a system prompt with examples of your writing and saying "write in this voice." Which is to say: it works about 70% of the time, drifts over long sessions, and requires correction. The feature isn't bad. It's just not $49/month better than the free equivalent.

Copy.ai has pivoted toward marketing workflow automation — generating multiple content types from a single brief, managing content calendars, team collaboration. If you're a marketing team that generates high-volume copy (product descriptions, ad variants, email sequences), the workflow features might justify the price. For individual writers, it's overhead without value. I used it for two weeks and spent more time learning the interface than I saved on the writing.

Writesonic and Rytr are in the "why does this exist" category at this point. They offer what ChatGPT Free offers, behind a paywall, with a worse interface. I suspect both are operating on the inertia of users who signed up in 2023 and haven't canceled yet [VERIFY — current state and user base of Writesonic and Rytr as of early 2026].

The Enhancement Tools: A Different Category Entirely

Grammarly, Wordtune, and Lex are doing something fundamentally different from the tools above. They're not writing for you — they're augmenting writing you've already done. Grammarly catches errors and suggests improvements. Wordtune rephrases sentences. Lex is a writing environment with AI assistance built in. The model isn't "AI writes, you publish." It's "you write, AI helps."

Grammarly's AI features have gotten significantly better over the past year. The tone detection, the clarity suggestions, the full-sentence rewrites — these are genuinely useful if you write a lot and want a second set of eyes. I use Grammarly Premium for final-pass editing, not for generation. The catch: Grammarly's AI suggestions tend to flatten voice. It makes everything clearer at the cost of making everything sound the same. If you have a distinctive writing voice, you need to selectively accept Grammarly's suggestions rather than accepting all of them — and that selective process requires knowing what your voice sounds like, which means you need to be a good-enough writer to use the tool effectively. Irony noted.

Wordtune is useful for one specific thing: sentence-level rephrasing. When you've written a sentence that almost works but doesn't quite land, Wordtune gives you five alternative phrasings. About one in five alternatives is better than what you wrote. The others are worse or equivalent. It's a roulette wheel of sentences, and occasionally it hits. Not worth a standalone subscription for most people, but the Chrome extension's free tier is fine for occasional use.

Lex is the most interesting tool in this category because it's trying to be a writing environment rather than a writing tool. The experience is: you write in Lex's editor, and when you pause or hit a keystroke, it suggests continuations. It's like Copilot for prose instead of code. The suggestions are sometimes useful, sometimes intrusive, and always distracting in a way that makes me wonder whether assisted writing actually slows down people who can already write. I used Lex for a month and my output was lower than my output in a plain text editor. The suggestions interrupted my thinking more than they extended it. But I've heard from other writers who love it — so this might be a workflow preference, not a tool quality issue.

What I Actually Use (The Boring Answer)

My daily writing stack is three things.

Claude Pro for generation and heavy editing. When I need to write something from scratch — or rewrite something substantially — Claude is the starting point. I give it context (system prompt with voice guidelines, relevant background information, structural requirements), tell it what I need, and iterate on the output. The first draft is usually 60-70% of the way there. Two rounds of editing — one for content accuracy, one for voice — gets it to publishable.

A plain text editor for writing I want to do myself. Some things I don't want AI help with. Pieces where the thinking is the point, where working through the ideas is how I figure out what I think. For those, AI is a hindrance because it resolves ambiguity too quickly. The value of writing — sometimes — is sitting with the ambiguity long enough for your own thinking to clarify. An AI that instantly produces a polished version of your half-formed thought robs you of the thinking process that makes the thought worth having.

Grammarly for final-pass error checking. Grammar, typos, clarity issues I missed. I ignore about 40% of its suggestions and accept 60%. The 60% I accept are almost always genuine errors or clarity improvements. The 40% I ignore are voice-flattening suggestions that would make the writing cleaner and also less mine.

That's it. Three tools. Total monthly cost: $20 for Claude Pro, $12 for Grammarly Premium, and $0 for the text editor. Thirty-two dollars a month. Less than a single month of Jasper's lowest tier.

I tried twelve tools and landed on three, and one of them isn't even an AI. The expensive dedicated writing platforms are gone. The novelty enhancement tools are gone. What remains is one strong LLM for generation, one quality-check layer for polishing, and a blank page for the writing that matters most.

The Real Lesson

The AI writing tool market is going through the same cycle every software category goes through: explosion, consolidation, commoditization. In 2023, there were meaningful differences between AI writing tools because the underlying models varied significantly and the interfaces mattered. In 2026, the underlying models have converged enough that the interface is a thin wrapper over roughly equivalent capability. The dedicated writing tools that survive will be the ones that offer genuine workflow value beyond "access to an LLM" — and most of them don't, which is why most of them won't.

If you're currently paying for a dedicated AI writing tool, open Claude or ChatGPT in a separate tab and try the same task. If the output is equivalent — and it probably is — cancel the dedicated tool. You're paying a premium for a brand name and a template library that constrains your output more than it helps it.

The best AI writing tool is a good LLM with a good prompt. Everything else is packaging.

This article is part of The Weekly Drop at CustomClanker — one topic, one honest take, every week.

Related reading: Claude vs ChatGPT for Daily Work, The Tool Collector's Guide to Owning Nothing, The Hex Constraint — Free Download