How we draft this blog with AI, and what we check

This is the first post on the LMG Labs blog, so it starts with how the blog itself gets made. AI writes the first draft. A person at LMG Labs reads it, fixes what needs fixing, and approves it before it is published.

That is also our short answer on using AI to write business blog posts. Let AI do the first draft. Keep a person in charge of the facts, the tone and the final yes.

How this blog gets made

Every post starts with a topic. AI turns that into a draft that has to follow rules we wrote down.

The draft then goes into a pull request in GitHub. A pull request is a proposed change that sits and waits for a person. Nothing in it is on the blog yet.

One of us reads the draft, edits it if needed, and merges it. Merging is what publishes the post. If nobody merges it, it never goes live.

The rules every draft has to follow:

• No invented clients, results, quotes, statistics or prices.
• Anything said about LMG Labs has to come from a short list of approved facts.
• Any number needs a link to its source.
• Posts on legal or tax topics carry a note that they are not advice.

We are also setting up a routine where one draft is prepared each morning from a topic list that includes our own notes. That routine is still being set up. When it runs, a person will still read and approve every post.

Using AI to write business blog posts: what works and what goes wrong

AI is a useful writing helper. It is not a reliable source of facts about your business.

Where it helps:

• Getting a rough first draft on the page, so nobody starts from a blank screen.
• Turning messy notes into clear paragraphs.
• Suggesting headings, a better order, or plainer words.
• Rewriting the same idea for a different reader.

Where it goes wrong:

• It makes things up. AI can write a number, a quote or a "fact" that sounds right and is not, with no warning.
• It gets bland. Left alone, it writes filler that could sit on any company's website.
• It only knows what you tell it. Unless you give it your prices, your past work and your customers' real questions, it will guess.
• It drifts in tone. Drafts can sound stiff, salesy or oddly cheerful.

That goes for the drafts on this blog too. It is the reason a person approves every post.

What a person should check every time

This is the part you cannot hand off. A quick read for typos is not enough. The reviewer's job is to catch anything that is untrue, unhelpful or not in your voice.

A simple review checklist

1. Claims about you. Is every statement about your business true, and would you put it in writing to a customer?
2. Numbers. Does each number have a source you opened and read yourself? If not, cut it.
3. Names and quotes. Are the customers, partners and quotes real, and do you have permission to use them? If in doubt, take them out.
4. Risky advice. Legal, tax, safety or HR topics need a not-advice note and, for anything specific, a check from the right professional.
5. Promises. Does the post promise a price, a timeline or a result you cannot always deliver?
6. Usefulness. Would a customer learn something they can act on? Delete lines that only fill space.
7. Voice. Read it out loud. Does it sound like your team talking to a customer?
8. Links. Click every link and make sure it goes where it should.

How to set this up in your own business

You do not need special software. You need a few written rules and one person who owns the final yes.

• Start from your own notes. The questions customers ask, the advice your team gives on the phone, the mistakes you see people make. That is what keeps a post from sounding generic.
• Write a short facts list. What you do, where you work, who you serve. Tell the AI tool to use only those facts about your business.
• Write your rules down. Borrow ours if they help: no made-up numbers, names or results.
• Pick one approver. Nothing goes live until that person says yes.
• Keep drafts in one holding spot. A shared folder or a document marked "draft" works fine. The point is that nothing gets published by accident.
• Keep private details out. Do not paste customer names, contact details or anything confidential into an AI tool unless you know how that tool stores it.

This fits how we like to work in general: one useful thing, then the next. Get one post through the process before you plan a whole calendar.

Example (hypothetical): A small heating and cooling company in central Alberta keeps a list of questions customers ask on the phone. The office manager picks one each week, adds the team's usual answer as notes, and asks a writing tool for a draft. The owner reads it, removes a line promising same-day service they cannot always offer, adds a tip from one of the technicians, and posts it.

If blog posts are not your biggest time drain, start with our post on what to automate first.

What to do this week

• Write a one-page facts list about your business.
• Collect the questions customers ask most, from email, phone calls and the front counter.
• Pick one question and have an AI tool draft a short post from your notes.
• Run the draft through the checklist above and note what you had to fix.
• Decide who gives the final yes before anything is published.

If you want help setting up AI drafting with a review step your team can trust, talk to us.