Grammarly Pro feels way too expensive in a world where affordable LLMs like ChatGPT exist and can handle grammar just fine. The only reason you’d still pay for it is the experience—inline editing that shows you exactly what’s wrong without rewriting your voice. Well, here’s a prompt that makes ChatGPT behave like Grammarly Pro, offering inline edits for seamless proofreading.
All you need is a specialized prompt
This was version one
If you’ve ever asked ChatGPT to proofread something, you already know the frustration. You paste in a paragraph, ask it to find all the errors and issues, and instead of marking what’s wrong, it hands you back a rewritten version—completely stripped of your voice.
And if you ask it to just share the places that have errors in them, it creates this bulleted list of all the errors with no easy way to find those instances in your original paragraph. You’re left doing a line-by-line comparison between two versions of your own writing—trying to figure out what changed and whether you even agree with it.
As such, even though modern chatbots are more powerful and affordable than Grammarly Pro—we still need to use it because it offers a more intuitive and streamlined editing experience. You see your original text, you see exactly what’s flagged, and you decide whether to accept it. If only ChatGPT could behave like this.
Well, with the right prompt, you can make ChatGPT function like Grammarly. You can force it to create a Canvas and return your original document with strikethroughs on the errors and bolded corrections immediately after. Punctuation fixes appear in brackets, like [,] or [?], so nothing gets buried. The result reads like a proofreader went through your draft with a pencil. You follow along, accept what makes sense by editing the document, and move on.
Here’s the prompt:
You are my personal proofreading assistant. I write in a flow state, so my drafts will have grammar mistakes, spelling errors, punctuation issues, and occasionally awkward or unclear phrasing.
Your job: Fix what's broken while preserving my voice and tone exactly.
Priorities (in order): Grammar > Spelling > Punctuation > Awkward phrasing > Overly long sentences
Style rules:
- NYT publication style guidelines
- US English (not UK)
- Use natural contractions (I'm, you're, it's) to avoid wordiness
- Target High School reading level — flag any word or phrase that reads as unnecessarily complex and suggest a simpler alternative
Output format:
- Place the full draft in a Canvas/Artifact
- Mark errors with ~~strikethrough~~ and corrections immediately after in **bold**
- If corrections are punctuations then put them in brackets like [,] or [?] and so on.
- For long sentences, add a note below that sentence suggesting how to split or trim it
Instead of pasting this prompt every time you start a new session, save it as a Custom GPT. This way, all you need to do is open the Custom GPT and paste your document and it’ll start proofreading—without any additional instructions.
Take it further with color-coded edits
It actually starts to feel like Grammarly now
The strikethrough-and-bold method works well, but there’s an even cleaner way to visualize edits—and it only takes one extra step. Instead of asking ChatGPT to mark up the text inline, you ask it to rewrite the document as an HTML file with color-coded edits: deleted text in red, corrections in green. Then you use ChatGPT’s Canvas feature to preview the rendered HTML directly inside the chat.
Here’s the prompt I use for this:
You are my personal proofreading assistant. I write in a flow state, so my drafts will have grammar mistakes, spelling errors, punctuation issues, and occasionally awkward or unclear phrasing.
Your job: Fix what's broken while preserving my voice and tone exactly.
Priorities (in order): Grammar > Spelling > Punctuation > Awkward phrasing > Overly long sentences
Style rules:
- NYT publication style guidelines
- US English (not UK)
- Use natural contractions (I'm, you're, it's) to avoid wordiness
- Target High School reading level — flag any word or phrase that reads as unnecessarily complex and suggest a simpler alternative
Output format:
- Start a Canvas/Artifact for HTML rendering
- Place the full draft that needs editing into the Canvas/Artifact with proper HTML formatting. Use responsive design with fluid typography and spacing. Optimize layout for mobile, tablet, and desktop. Avoid fixed widths; prefer adaptive containers
- Use HTML tags styled red for all the errors and issues, immediately followed by the correction in green — apply this for both words and punctuation
- For long sentences, add an inline note below that sentence suggesting how to split or trim it
- Never return the corrected passage in chat only — it must always be in the Canvas/Artifact
Feel free to edit this prompt based on the type of errors you want to catch most often. You can also optimize it to fine-tune how everything is rendered.
The advantage of this approach over the basic strikethrough method is readability. However, the downside is that ChatGPT now needs to wrap the entire output inside HTML code for proper rendering. That’s a lot of output, and if you’re on the free plan using one of the weaker models, it can mess up the task. As such, I would recommend that you use the color-coded edits only for documents that are less than 2000 words, and the simple strikethrough edits for anything longer.
I transformed ChatGPT into a full fledged project management system
No Notion, no Trello, no Asana—just ChatGPT, a custom prompt, and a workflow that actually sticks.
Jump between multiple AI models for increased limits
The same prompt works on Gemini and Claude too
If you’re on the free tier of ChatGPT and hit your usage limit, you don’t have to stop. The same prompt works for Gemini and Claude, and both cover the two ChatGPT features we used here. To save your prompt, Gemini has Gems and Claude has Projects—both work like Custom GPTs. For the color-coded canvas preview, Gemini has its own Canvas feature, and Claude has Artifacts, which renders the HTML output in a side panel.
Performance is also comparable across all three. Proofreading isn’t a task that demands frontier-level reasoning—it’s pattern recognition and style adherence. The current free models are more than capable of handling it well. That said, if I had to pick one, I find Claude is better at both HTML rendering and error capture, whereas Gemini is better at handling longer documents.
You’re using Claude wrong if you’re treating it like ChatGPT
Many people are fleeing ChatGPT for Claude but making the same mistakes.
One less subscription to justify
Grammarly’s real advantage was never the grammar engine—it was the editing experience. Now that a simple prompt replicates that experience on any AI chatbot, the value proposition quietly falls apart. Yes, Grammarly’s yearly plan is cheaper than ChatGPT Plus—but ChatGPT does a lot more than fix grammar. So for me at least, I’m canceling my Grammarly subscription.
