You tried ChatGPT for a blog post. What came back read fine and said nothing. Generic. Safe. The kind of paragraph about how fast technology moves and how much security matters that every other MSP in your market published the same week.
So you went back to writing everything yourself, or you stopped publishing.
Here is what went wrong. You used a blank tool and expected it to know your voice. It does not. Out of the box, an AI writing tool gives you the average of everything it read on the internet, and the average of the internet is mush.
A writing engine fixes that. You spend a few hours up front teaching the tool how you talk, what you believe, and what you refuse to say. After that, first drafts come back sounding like you wrote them tired instead of like a press release wrote itself.
This page walks through how to build one.
What a writing engine actually is
Strip away the marketing and it is three things.
Source material. Real examples of your voice that the tool reads before it writes anything.
An instruction set. Written rules about voice, tone, structure, and banned language.
A repeatable prompt pattern. The way you ask, every time, so you stop rolling the dice.
Most tools support this. Custom GPTs in ChatGPT, Projects in Claude, and Copilot agents all let you attach documents and save instructions. Pick one and stay there. Switching tools every month costs you more than any feature you gain.
Step 1: Feed it your real voice
This is the step people skip, and skipping it is why the output stays generic.
The instinct is to upload your website copy. Do not start there. Your website was probably written by a marketing agency two years ago, and it sounds like a marketing agency. Feed that in and you get more of it back.
Feed it these instead, in rough order of value.
Sales call and discovery call transcripts. This is the highest value source you have and almost nobody uses it. It is you, explaining your business, to someone who needs to understand it. Your real vocabulary. Your real analogies. The way you handle the pricing objection. Pull five or six recordings, run them through a transcription tool, strip the client names, and upload them.
Recorded talks. Podcast appearances, webinars, conference sessions, internal all hands meetings. Same reason. You talk more like yourself than you write.
Emails you wrote to clients. Not templates. The ones you typed at 7pm explaining why the migration slipped. That is your voice under pressure.
Blog posts you actually liked. Five is plenty. Only include the ones where you read them back and thought yes, that sounds right. If you include mediocre posts, the tool learns to write mediocre posts.
Your positioning. Who you serve, what you sell, what you refuse to sell, what you charge for. The tool needs this to avoid writing advice that contradicts your own business.
Ten to fifteen documents beats fifty. Quality of source material sets the ceiling on quality of output.
Step 2: Write the instruction set
The source material teaches sound. The instruction set enforces rules. You need both.
Most instruction sets fail because they are vague. "Be professional and approachable" tells the tool nothing. Every model already thinks it is professional and approachable. Get specific enough that a new hire could follow the same rules.
Here is a working example. Adapt it, do not copy it.
Audience. Owners and operators of managed service providers with 10 to 20 employees, $3M to $5M in annual revenue. They are technical. They are busy. They have heard every pitch in this industry twice.
Voice. Direct and plain. Write like you are talking to a peer across a table, not presenting to a board.
Point of view. Address the reader as you. Never write in third person about "MSPs" when you mean the person reading.
Sentence structure. Vary the length. Short sentences for emphasis. Longer ones when the idea needs room. Never write three long sentences in a row.
Certainty. State things plainly. Do not hedge with "could," "might," or "may" when you know the answer.
Banned. No em dashes. No semicolons. No emojis. No exclamation points. No "in today's fast paced world." No "game changer," "cutting edge," "seamless," "robust," "leverage," "unlock," or "delve." No "In conclusion" or "Furthermore."
Structure. Open with the problem, not with context. Use subheads every 200 to 300 words. End with one clear next step.
Honesty. Say when something is hard, expensive, or does not work. Never claim a result without a reason behind it.
The banned list does the heaviest lifting. Every model has favorite tics, and once you name them, they mostly stop. Add to that list every time you catch a phrase you hate. Mine has grown for two years.
Step 3: Prompt with intent
A trained engine still needs a real brief. "Write a blog about cybersecurity" gets you a bad blog about cybersecurity.
Give it five things.
- The topic, narrowed. Not "backup," but "why your backup passes every test and still fails during a real recovery."
- The audience, if it differs from the default. Writing for your clients reads differently than writing for peers.
- The angle. What do you believe about this that most people do not? That is the piece worth publishing.
- The source. Point it at a specific document you uploaded. "Open with the story from the March discovery call transcript."
- The shape. Length, format, and what the reader should do at the end.
A brief that takes you four minutes to write saves you thirty minutes of editing. That trade is always worth it.
Step 4: Edit in passes, not all at once
The first draft lands around 70 percent. That last 30 percent is the difference between publishing and embarrassing yourself.
Do not tell the tool to "make it better." Give it one job at a time.
Pass one, structure. Does the piece open with the problem? Does every section earn its place? Cut anything that repeats a point you already made.
Pass two, language. Highlight the whole draft and ask it to strip every phrase on your banned list. Then read for hedging and make those sentences definite.
Pass three, truth. This one is on you. The tool invents statistics, misremembers vendor details, and states things about compliance that sound right and are wrong. Check every number and every claim about a regulation. It will not check itself.
Pass four, read it out loud. If you run out of breath in a sentence, break it. If a phrase feels like something you would never say to a client, rewrite it. This test catches more than the other three combined.
What this does not do
Worth being clear, because overpromising here wastes your money.
It does not know your client stories. You have to give it those.
It does not have opinions. The angle comes from you. An engine turns your thinking into clean prose. It does not do the thinking.
It does not remove the human read. Publish unreviewed AI content and you will eventually publish something wrong with your logo on it.
It does not replace a writer if you have a good one. It makes a good writer roughly twice as fast.
The mistakes that cost the most
Feeding it agency copy. You get agency copy back. Use your own words.
One engine for everything. Blog posts, proposals, and client emails need different rules. Build separate engines. It takes twenty minutes each after the first one.
Accepting the first draft. The first draft is raw material. Treat it that way.
Vague instructions. If your rules would not help a new marketing hire, they will not help the tool.
Never updating it. Add to the banned list monthly. Swap in better source material as you produce it. An engine you last touched a year ago writes like you did a year ago.
What to expect
Setup runs two to four hours if you have your source material ready. Gathering transcripts takes longer than writing the instructions.
After that, a solid 900 word draft takes about 20 minutes instead of 90. The quality holds because the rules hold. And you publish consistently, which matters more than any single post.
Where this goes wrong for MSPs specifically
Most MSP owners do not have a content problem. They have a positioning problem wearing a content costume.
If you cannot say in one sentence why a prospect should pick you over the MSP across town, no writing engine saves you. You will produce clean, well structured, on brand content that says the same thing as everyone else, just faster.
Sort the positioning first. Then build the engine.
Talk it through
Sort the positioning first. If you cannot say in one sentence why a prospect picks you over the shop across town, no writing engine saves you. That sentence is the work, and it is the part I do with owners.
Everything above is a template. Run it yourself this week and never talk to me. What I will not do is hand you a system and tell you your shop is wrong for not matching it.
Not sure this is your actual constraint? Take the MSP Owner Reality Check. Five questions, nine minutes, and it names the two or three things quietly capping your growth. https://themsphero.com/resources/msp-owner-reality-check-assessment
If you already know what is broken, book a 30 minute fit call at https://letschat.themsphero.com
Mike Kolb The MSP Hero