Three prompts to improve your writing
Supercharge your writing skills with a little help from AI
If you read our previous article on a pragmatic workflow for incorporating generative AI into your writing, you’ll know we love AI, but we also believe the only genuine way to use it is by keeping the human at the center. If you haven’t read it, check it out; it’s pretty cool.
This post complements that article: we share three prompts we’ve been experimenting with lately to streamline this process even more.
We use these prompts as templates for Bearly.ai because it makes it extremely convenient to simply fill in some fields and hit a button, but you can use them directly with ChatGPT as well. We haven’t tried these with other LLMs so they might not work with Llama, Claude, or anyone else. If you do try them, we’d love to know!
Let’s go!
Proofreader
The first prompt acts as a proofreader. The most important feature of this prompt is that it makes the least possible number of changes to fix grammar, orthography, and punctuation. Thus, it tends to keep the same paragraph and sentence structure unless you have very long sentences. If your text is flawless, it will probably make no changes.
To use it, paste your paragraph, substituting the {{ input text }}
fragment.
You are a proofreader. Your task is to rewrite the input text below to improve the grammar, orthography, and punctuation. Keep as much of the original text as possible, only rewrite where there is a mistake. Remove repeated words and split long sentences. Do not add any content that is not in the original text.
"{{ input text }}"
Editor
The second prompt is more flexible. It allows the model to rewrite the entire text to fit it into a new style and audience. So, always review the end result carefully, because it will change almost all your words, and probably reorder sentences as well.
In this prompt, you must also specify the tone —e.g., “formal”, “casual”, or even “slightly drunk with a bit of sadness”— and the audience —e.g., “little kids”, “professional computer scientists”, etc.
If the text is too long (like, several paragraphs) or too short (like a single sentence), this prompt can hallucinate and add its own content. However, we’ve seen when the text is one or two paragraphs, it tends to behave correctly. Do experiment with different text lengths.
You are an editor. Your task is to rewrite the text below to improve its style. Keep as much of the original content as possible, but rewrite sentences in active voice, and split the text into meaningful paragraphs. Consider the following tone: {{ tone }}. Assume the reader is {{ reader }}, and choose words appropriate to their level of comprehension. Do not add any content that is not in the original text below.
"{{ input text }}"
Structured Outline
The final prompt is different in that, instead of changing the wording, it generates complementary content in the form of an ordered list of short sentences. These are the main claims of your input text. We use this prompt to extract key insights from very messy audio transcripts, for example, so we can reorder, mix, and refine them before doing a second round of writing.
Again, it is sensible to the length of the input text. Too long or too short, it will tend to divagate or hallucinate, but it works pretty well with chunks of ~1000 words. If your text is longer, consider splitting it into sections.
Your task is to extract the main claims of the following text and create an structured outline. Identify the main claims made through-out the text, and summarize each in a single sentence, in active voice. Enumerate all the claims in the same order in which they appear in the original text. Answer only with the enumerated list of claims.
"{{ input text }}"
Needless to say, these prompts can fail spectacularly, so always double-check the output. They are meant to help you, not to substitute you. Do play with them, try some modifications, and let us know what works and what doesn’t.
Consider sharing this article with all your writer friends if you find it useful.
Great place to get folks started thinking about this stuff. I absolutely love the ability to check for grammar and spelling errors instantly, and I further appreciate the ability to ask what the takeaway is. If an LLM gives me a very different summary than I think the piece is about, I know I have a lot of work to do.
A change that I always tell chatGPT to do is show me the text that they’ve changed in bold with the previous content in parenthesis alongside it.