How do I write with AI without sounding like AI?

The short version

• Start from your notes: the facts, the customer, and what you want them to do.
• Ask for one short draft in plain words. Then edit it like a new hire’s work.
• Cut filler, add one detail only you know, and read it aloud.
• Check every name, number, date, price, law and link against the source.
• Google allows AI-assisted content but says to fact-check it and not mass-produce pages.

How do I get a draft that sounds like me?

1. Write your notes first

Two minutes of your own notes beat any clever prompt. Jot down:

• The facts: what, when, where, price if any, and who to call.
• The reader: a first-time customer, a regular, a supplier.
• The one action you want: call, book, reply, come in.
• One real detail: something that happened on a job, a question customers ask, a mistake you see people make.

2. Ask for something small and specific

For example: “Using only these notes, write a 120-word email to existing customers about our new Saturday hours. Plain words, short sentences, no exclamation marks. If anything is missing, ask me.” Paste a paragraph you’ve written before and say “match this tone.”

3. Edit hard

AI drafts tend to share the same habits. Cut them:

• Openers that say nothing, like “In today’s busy world…” Start with the news.
• Stacked adjectives and big promises. “Comprehensive, tailored solutions” means nothing to a customer.
• Lists of three for rhythm when one point is enough.
• A closing paragraph that repeats the email. End with the action and your name.

Then add back what only you know: the name of the street you’re on, the question Mrs. Jones always asks, the reason you changed the hours.

4. Read it aloud

If you wouldn’t say it to a customer at the counter, change it. This is the quickest test there is.

5. Check the facts

This is the step people skip. Google’s guidance on generative AI content explains that these models predict likely words rather than look up facts, so output can contain errors. Its advice is to “manually factcheck and review all AI-generated content” before publishing. Go line by line:

1. Names, addresses, phone numbers and hours: against your own records.
2. Prices and dates: against your price list and calendar.
3. Numbers and statistics: find the original source, or delete the number.
4. Laws, rules and deadlines: against the government or official page, never the AI’s summary.
5. Links: click every one. AI tools sometimes write links that look right and go nowhere.
6. Quotes: if it says someone said something, find where they said it.

Google says the same review applies to page titles, meta descriptions, structured data and image alt text, since these can show in search results.

Why does this matter?

Customers notice writing that could have come from anyone, and they stop reading. A wrong fact costs more: a customer drives over on the wrong day, or a page promises something you don’t do.

There’s a search side too. Google’s helpful content guidance says using automation “to produce content for the primary purpose of manipulating search rankings” breaks its spam policies. The using-AI page warns that generating many pages without adding value may count as scaled content abuse. AI-assisted pages that help real people are fine. Dozens of thin pages written to rank are not.

What are the common mistakes?

• Publishing the first draft. It reads like a template because it is one.
• Asking for “an SEO blog post about plumbing.” You’ll get a generic article that dozens of other sites already have.
• Fake people. Google’s helpful content page warns against “AI-generated headshots, made-up names, or false credentials.” Pages should be by real people.
• Invented numbers. “Most homeowners agree” or a precise-sounding percentage with no source. If you can’t link the source, cut it.
• Letting it guess your details instead of pasting them in.
• Copying another business’s page into the tool and asking for a rewrite. That’s still their work.

Should I say that I used AI?

There’s no general rule that you must label every AI-assisted email. Google suggests explaining how content was made where a reader “might think” about it, such as an article built mostly with automation. For product listings, Google Merchant Center has its own labeling rules for AI-generated images and text. When in doubt, the honest test is simple: would the customer feel misled if they knew? Our guide to what the FTC says about AI covers the cases where honesty is the law, such as reviews.

What if the drafts still sound robotic?

• Give it more of your own writing as an example, and fewer instructions.
• Ask for half the length.
• Ask it to list the facts it used. If you see one you didn’t give it, delete that line.
• Write the first and last sentence yourself, and let it fill the middle.
• For anything personal, such as an apology or a thank-you to a long-time customer, write it yourself. AI is best for the routine.

What does it cost, and how long does it take?

Free tiers handle this kind of writing. The time saved is in the blank-page stage. Plan on spending at least as long checking and editing as the tool spent writing.

How does E-3 handle this?

At E-3, people make the changes to client sites, and updates and bigger changes are tested on a staging copy first. On this Learning Center, every page lists its sources and a last-reviewed date so you can check them yourself.

Related questions

• What are AI assistants good and bad at?
• Can I use AI to answer customer emails and reviews?
• Can I use AI images, and who owns them?
• Which pages are worth writing for AI answers?
• Back to the complete guide to AI tools for small business

Sources

1. Using generative AI content on your website (Google Search Central, updated October 1, 2026)
2. Creating helpful, reliable, people-first content (Google Search Central, updated October 5, 2026)