How to Use AI for Writing Without Sounding Robotic

You asked an AI tool to write a client update, and it came back polished, polite and completely unlike you. It opens with “I hope this email finds you well,” uses 3 adjectives where 1 would do, and ends with a tidy summary nobody asked for. Your client will know a machine wrote it before they finish the first line.

The fix is not a better tool. It’s a better process. AI is very good at turning raw material into a first draft, and poor at knowing what you actually think, what your reader already knows and how you normally talk. This guide shows the prompt patterns that get a draft closer to your voice, and the editing pass that makes it yours.

Why AI writing sounds robotic

Large language models generate text by predicting likely next words based on patterns in their training data. “Likely” is the problem. The most probable phrasing is, by definition, the most common one, so unedited AI text drifts toward the average of everything ever written about a topic. If you want the mechanics, our explainer on what a large language model is covers it in plain English.

In practice, that average shows up as a handful of recognizable habits:

• Throat-clearing openers that restate the question before answering it.
• Stacked qualifiers such as “may potentially help to” instead of “helps.”
• Rule-of-three lists everywhere, even when there are really 2 points or 5.
• Generic praise words and buzzwords that carry no information.
• Uniform sentence length, which makes the rhythm feel flat.
• A summary at the end that repeats what the reader just read.
• No specifics: no names, numbers, dates, examples or opinions.

That last one matters most. Readers don’t spot AI because of one phrase. They spot it because the text could have been written by anyone, about any company, for any reader.

Feed it your thinking, not just a topic

The single biggest improvement comes before you type a prompt. A one-line request like “write a LinkedIn post about our new scheduling feature” forces the model to invent the content, so it invents the most generic version possible.

Instead, give it the raw material. Dump your notes, even if they’re messy:

• What happened, in your own words
• Who the reader is and what they already know
• The 1 point you want them to remember
• Specific details: numbers, customer feedback you’re allowed to share, a quick story
• What you want the reader to do next

Then ask the AI to organize and tighten your notes rather than write from scratch. The words and ideas stay yours; the model handles structure and grammar. This also cuts down on made-up facts, a risk covered in our guide to AI hallucinations.

Prompt patterns that keep your voice

Model makers’ own documentation points to the same few techniques. Both Anthropic and OpenAI describe examples (few-shot prompting) as one of the most reliable ways to steer tone and format, and Anthropic recommends telling the model what to do instead of what not to do. Here are 6 patterns you can reuse.

1. Show a sample of your writing

Paste 2 or 3 short pieces you wrote and are happy with, such as past emails or posts. Add: “Match the tone, sentence length and vocabulary of these examples. Don’t copy their content.” Pick examples that are similar to the task, not your most formal or most casual work.

2. Describe the voice in plain terms

Write a short voice note you can reuse: “Direct and warm. Short sentences. Plain words. Uses ‘we’ and ‘you’. Occasional dry humor. No exclamation marks. Never opens with a question.” Save it in a document or in your tool’s custom instructions so you don’t retype it.

3. Say what you want, not only what to avoid

“Don’t be wordy” is vague. “Keep it under 120 words, with sentences under 20 words, and lead with the decision” is something the model can act on. Anthropic’s guidance gives a similar example: instead of banning a format, describe the format you want.

4. Explain who the reader is and why

Context helps the model choose the right level of detail. “This goes to a busy property manager who skims on her phone and only cares whether the repair is done Friday” produces a very different draft than “write a repair update.”

5. Match your prompt to the output

Anthropic notes that the style of your prompt can influence the style of the response. If you write your prompt in heavy bullet points and headings, expect bullets and headings back. If you want flowing paragraphs, write your request in paragraphs.

6. Ask for options, not a final answer

“Give me 3 different opening lines, each under 15 words” lets you pick or combine. You stay the editor instead of accepting the first thing the model produces.

For a deeper look at building prompts, see what prompt engineering is and how it works.

The 10-minute editing pass

Even with a great prompt, treat the output as a draft. This editing pass takes about 10 minutes for an email or short post and makes the biggest difference.

1. Rewrite the first sentence yourself. AI openings are the most generic part. Start with the point, the news or the question your reader has.
2. Cut the last paragraph if it only summarizes. Most short pieces end better on the call to action.
3. Delete filler words and qualifiers. Search for “very,” “really,” “truly,” “potentially,” “in order to” and “it’s important to note.”
4. Replace every vague claim with a specific. “Improves efficiency” becomes “cut our invoice time from 3 days to 1.”
5. Add 1 thing only you know. A customer comment, a mistake you made, a local detail or an opinion.
6. Break the rhythm. Mix a few very short sentences with longer ones.
7. Check every fact, name, number and link. Don’t assume the model got them right.
8. Read it aloud. If you wouldn’t say a sentence to the reader’s face, change it.

Key takeaway: AI is fastest at the rough first draft. The finishing work, your opinion, your specifics and your edits, is what makes readers trust it. Don’t skip that part.

Robotic tells and how to fix them

Use this table as a checklist during your editing pass.

When to use AI for writing, and when not to

AI assistance works best for:

• First drafts of routine emails, updates, job posts and proposals
• Restructuring messy notes into a clear outline
• Shortening long text to a word limit
• Alternatives for headlines, subject lines and openings
• Tone checks, for example “Does this sound defensive?”
• Grammar and clarity passes on your own writing

Be more careful with:

• Personal messages such as condolences, apologies and thank-you notes. People can tell, and it matters to them.
• Expert or regulated content in health, legal or financial topics, where errors carry real risk.
• Anything with facts you haven’t checked, especially statistics and quotes.
• Confidential information. Check your company’s rules before pasting client or employee data into any AI tool. Our guide on whether it’s safe to put company data into AI covers what to look for.

If you publish on a website, Google’s guidance on generative AI content says AI can help with research and structure, but generating many pages without adding value for users may violate its spam policy on scaled content abuse. It also asks site owners to check AI content, including titles and meta descriptions, for accuracy before publishing. In other words, the editing pass is not optional for web content either. Our guide to writing SEO blog posts covers the rest of that process.

Where to start: build your personal voice kit

You can set this up in under an hour and reuse it for months.

1. Collect 5 samples of your best writing across email, social posts and longer documents.
2. Write a 5-line voice note describing your tone, sentence length, words you use and words you never use.
3. List your banned phrases, the ones you keep deleting from AI drafts.
4. Save a reusable prompt template: role, reader, goal, your notes, voice note, samples, length and format.
5. Store it in your AI tool’s custom instructions, a project or a saved document.
6. Run every draft through the 10-minute editing pass above, and add new tells to your banned list as you notice them.

If you’re rolling this out to a team, pair it with clear guidance on what employees may and may not use AI for. Our template for an AI acceptable use policy is a good starting point, and how to use AI at work covers other everyday tasks.

Conclusion

AI can save you real time on writing, but only if you keep the parts that make writing yours: what you think, what you know and how you talk. Give the model your notes and samples, ask for drafts and options, and always finish with your own editing pass. Your readers will get clearer messages, and they’ll still hear you in them.

Sources

• Anthropic: Prompting best practices — using examples to steer tone, adding context, saying what to do instead of what not to do, matching prompt style to output style
• OpenAI: Prompt engineering guide — few-shot examples, adding context and setting communication style in instructions
• Google Search Central: Guidance on using generative AI content on your website — accuracy checks, metadata review and scaled content abuse