Which tool to choose
There's no single tool: it depends on what you're doing.
If you want a reviewer that reasons about your text (pointed comments, proposed rewrites, analysis of tone and structure): use Claude. It tends to produce the most natural prose, handles voice and tone well and holds quality on long texts. The free version gives access to Claude Sonnet with a daily message limit, and on long responses the free plan behaves better than ChatGPT's.
If you already work inside Google Docs or Word: stay in your environment. Word with Copilot or Google Docs with Gemini save you the copy-paste back and forth, and this matters more than a few extra points of quality.
If you only need to clean up typos, grammar and tone (emails, short posts): Grammarly. It focuses on correcting the writing, not generating it: it checks grammar, spelling, clarity and tone, and this is where Hemingway doesn't reach.
If the problem is that the text is too dense and tiring to read: Hemingway Editor. The web version covers the readability check without needing the paid plan. Note: it measures how long and convoluted the sentences are, not the typos.
The combination I use and recommend: Claude (or ChatGPT) for reasoned feedback + Hemingway for the final readability check. They cover the two different jobs you really need.
How to do it
The critical point: if you ask "what do you think?" you get useless compliments. You have to give context and ask for criticism.
- Open Claude (claude.ai) or ChatGPT (chatgpt.com). From browser or app the path doesn't change: go to the message box.
- Paste the prompt below and replace the part between quotes with your text.
- Always state purpose, recipient and register (formal/informal): without these, the AI doesn't know what to judge against.
- Read the points first, then the proposed rewrites. Don't accept everything wholesale.
- For sentences that only half convince you, ask for two or three alternatives and choose yourself.
The working syntax for critical feedback:
You are an expert editor. Analyze the text below.
Context:
- Purpose of the text: [e.g. convince a client to book a call]
- Recipient: [e.g. SME owner, little time, skeptical]
- Intended register: [e.g. professional but direct, no jargon]
What I want from you, in this order:
1. The three most serious problems in the text, from the most serious. For each one: where it is, why it's a problem, how you'd fix it.
2. What already works and should be kept.
3. A rewritten version of only the weak points, leaving the rest.
Don't rewrite everything. Be blunt: if a sentence is weak, say so.
Text:
"[paste your text here]"
After this step you should have a list of concrete problems, not a generic grade. If the AI gives you back only praise, reply: "Be stricter, find the three weakest points even if the text seems good to you."
For the final check: paste the text into Hemingway Editor (hemingwayapp.com). It highlights in yellow and red the sentences that are too long or complex and gives you a reading grade. For web texts aim for a grade between 6 and 8: below that threshold the sentence flows.
Concrete example
Marta needs to send an email proposing a collaboration to a shop. Her first version: "Hello, my name is Marta and I would be interested in understanding whether there might be the possibility of evaluating a potential collaboration between our businesses, should you be available."
She pastes the text into Claude with the prompt above, stating: purpose = get a meeting, recipient = busy owner, register = courteous but direct.
The feedback in a few seconds: the sentence is full of conditionals that make it insecure ("I would be interested", "there might be the possibility", "a potential", "should you"), it lacks a concrete proposal and there's no clear call to action (the CTA, Call to Action, that is, the explicit request for an action).
Proposed and adopted rewritten version: "Hello, I'm Marta. I have a collaboration proposal that would bring new customers to your shop: how about a 15-minute call this week?"
Result: three conditionals eliminated, a concrete benefit and a precise request with timing. Marta got the reply the same day.
When it does NOT work (and how to fix it)
If the AI just fills you with compliments
Free versions tend to be accommodating. Rewrite the request imposing a fixed number: "Find exactly three weaknesses, even if the text is good. For each one explain why a demanding reader would notice it." Forced to a number, the AI is obliged to look for problems.
If it rewrites everything and erases your style
It's the most common risk: the text comes back correct but depersonalized. In the prompt write "don't rewrite everything, intervene only on the weak points and keep my voice". If it already happened, paste your original version and the AI's and ask: "Merge the two keeping my tone and only the necessary corrections."
If you hit the free message limit
Claude's free plan offers Sonnet with daily limits that vary based on server load. When you exhaust it, you have two real options: switch to another tool (ChatGPT free for general tasks, Claude free for writing: two free accounts cover your day), or, for very long texts, split the document into two or three parts and do one session per part.
If you need to publish and you're worried about transparency
The AI helps you improve, but the text stays yours. Using it for editing is like using a thesaurus or hiring a proofreader: the line you cross is when you pass off as entirely yours a text the AI generated from scratch. If you publish a book or enter contests, check the rules: Amazon, traditional publishers and most contests now ask you to declare whether AI generated significant portions of the manuscript. Read the contract before submitting.
A tip from someone who actually uses it
Don't ask for feedback just once. Do two rounds with two different lenses: the first on structure (does the order of ideas hold? is something out of place?), the second on the sentence (weak words, repetitions, rhythm). Mixing the two levels in a single request produces confused feedback. And always keep your original version aside: sometimes the rewrite is smoother but less yours, and the right version is a crossing of the two that you decide.
Frequently asked questions
What's the best prompt to have a text corrected?
The one that gives context. The difference isn't in the magic formula but in the three pieces of information you provide: purpose, recipient, register. Use the copyable block above and customize those three lines: the rest of the work the AI does.
Do I have to pay to get decent feedback?
No, for normal use. On each of these platforms you do real work for free: the difference with the paid plan is how much you can work before hitting the limits, not the quality of the individual feedback. You pay only if you review many texts a day.
Can I have a whole document reviewed, not just a paragraph?
Yes, but with method. To condense a 50-page document into a summary Gemini does well; Claude is preferable when the result itself has to be well written. Upload it as a file (all the free versions accept uploads) instead of pasting it, ask first for a structure analysis and then drill down into the individual chapters.
Doesn't the AI risk making my text the same as everyone else's?
It's the real risk, and it depends on how you use it. If you ask it "you write it", yes: you get the average of the internet. If you use it the way a reviewer does with an author — it flags the problems, you decide — the text stays yours. The AI recognizes patterns well, but it's bad precisely at the things readers pay for: a precise point of view, lived experience, the detail you know only if you were there. That value you put in; the AI only helps you express it better.