Editing, Shortening and Proofreading with AI
Writing a first draft is only half the job, and it is often the easier half. The gap between a mediocre email and a genuinely strong one is usually closed in editing: cutting what does not need to be there, catching errors, and tightening loose sentences into ones that land. AI tools are as useful for this second pass as they are for the first draft, and this lesson covers the specific editing prompts that reliably improve any email, whether you wrote the original yourself or an AI drafted it.
Shortening without losing meaning
Most professional writing, AI-generated or human, is longer than it needs to be on the first pass. A simple, powerful habit is to always ask for a shortened version after the first draft, even one you are happy with. A prompt like “Cut this email by a third without losing any of the key facts or the call to action” forces a genuine tightening rather than superficial trims, and it is worth doing on nearly every email longer than a hundred words. If the result still feels padded, iterate again: “Cut it by another 20 percent, be more aggressive about removing unnecessary words.”
It helps to know what specifically gets cut in good shortening. Redundant phrases such as “I just wanted to reach out to” or “in order to” collapse to nothing or to “to.” Throat-clearing openers like “I hope this finds you well” often disappear entirely without loss. Qualifying phrases like “I think that perhaps” weaken a sentence and can usually be removed outright. Ask the AI directly to hunt for these: “Remove any hedging language, redundant phrases, and unnecessary qualifiers, keep the meaning exactly the same.”
Proofreading: catching what you cannot see anymore
After reading your own draft several times, your brain starts autocorrecting errors without you noticing, which is exactly why a fresh pass, human or AI, catches mistakes you have stopped seeing. A simple, reliable prompt: “Proofread this email for grammar, spelling, and punctuation only. Do not change the tone, wording choices, or structure, just fix errors and list what you changed.” That last instruction, asking the tool to list what it changed, matters because it lets you confirm the AI has not quietly altered your meaning or softened a firm request while “just fixing grammar,” something that occasionally happens if you do not constrain the task tightly.
A structured editing checklist
| Pass | What to check | Prompt to use |
|---|---|---|
| 1. Clarity | Is the main point in the first two sentences? | “Is the core message clear within the first two sentences? If not, suggest a fix” |
| 2. Length | Is every sentence earning its place? | “Cut this by a third without losing key facts” |
| 3. Tone | Does it sound like the intended relationship and mood? | “Does this tone match writing to [describe relationship]? Flag anything off” |
| 4. Accuracy | Are all names, dates and figures correct? | Manual check only, never delegate this step to AI |
| 5. Grammar and typos | Final mechanical check | “Proofread for grammar and spelling only, list any changes” |
Notice that pass four, accuracy, is explicitly marked as a manual step. This is the most important caution in this entire lesson: AI tools cannot verify facts they were not given, and they will proofread confidently around an incorrect date or a misspelled client name without flagging it, because from the model’s perspective, “Thursday the 15th” reads as perfectly grammatical whether or not the 15th is actually a Thursday. Always do a final human pass focused purely on facts, names, dates and numbers, separate from any AI-assisted language pass.
Getting a second opinion on persuasiveness
Beyond mechanical editing, AI tools can act as a useful critical reader for anything meant to persuade or influence, sales emails, proposals, requests for a favour. A prompt like “Read this as if you were the busy recipient with no reason yet to care. What would make you stop reading or ignore this?” often surfaces real weaknesses, an unclear ask buried too late, an opening that does not establish relevance fast enough, that are hard to see from inside your own draft. This works particularly well with Claude and ChatGPT, which can hold a genuinely critical, outside perspective when explicitly asked to role-play the recipient rather than simply asked whether the email is good.
Reading level and audience fit
A final useful editing pass is checking whether the email is pitched at the right complexity for its audience. Internal technical emails to specialists can carry jargon and assume background knowledge; emails to a general client or a non-specialist stakeholder cannot. A prompt such as “Rewrite this so someone with no background in our industry could follow it easily, explain any technical terms in plain language” is a quick, reliable way to adjust reading level without a full rewrite, and it is worth running any time you are unsure whether your usual audience matches the one actually receiving this particular email.
Editing turns a good draft into a sendable one. The next lesson looks at how to stop rebuilding good emails from scratch every time, by turning your best AI-assisted drafts and prompts into a reusable personal library of templates.
