By Brett Kaufman. 8 years using relationships to build sales pipelines, from political fundraising to a 7 figure business to running matchmaking for two communities. Over 1,000 introductions facilitated.

AI can write texts that sound like you, but only after it has learned from how you actually text. Out of the box, every tool writes in the same voice: polite, complete, a little too warm, and obviously not you.

You have seen it. You ask a tool to draft a note to a broker you have not talked to since spring. It comes back: “Hey Mark! Hope you’re doing well. I came across your recent LinkedIn post about the HVAC listing and really loved your insights. Would love to reconnect and hear more about what you’re working on!” Every word is fine. Mark would know in two seconds you did not write it.

That is the real test for any tool that drafts messages for you. Not the feature list. Whether you would send the draft as written.

I have facilitated over 1,000 introductions between deal makers, operators and investors. An intro only lands when the message sounds like a person wrote it. That is the bar I hold every draft to.

What makes a text sound like a bot

The tells are consistent. Once you see them, you see them everywhere.

A greeting before the point. “Hey Mark!” then a pleasantry, then the reason. Real texts between busy people start with the thing.

Filler openers. “Hope you’re doing well.” “I wanted to reach out.” “I came across.” These are email habits. Nobody texts a peer that way.

Length. Four sentences where one would do. A text that needs scrolling reads like marketing no matter how personal the details are.

Generic compliments. “Great insight.” “Spot on.” “Really resonated.” Grading words make you sound like a critic reviewing their work, not a person who read it.

Reading their post back to them. “Your post argued that sellers underestimate working capital.” They know what they wrote. Summarizing it proves you read it the way a machine reads it.

Naming the platform or the date. “Your October 5th LinkedIn post.” People who actually saw something say “saw the post” and keep going.

The interview question. “How do you think about valuation multiples in a rising rate environment?” Nobody wants homework from a text.

Facts nobody gave it. The draft says “congrats on the Florida expansion” when the post only said they opened new stores. Invented detail is the fastest way to get caught.

A broker buried in messages every week has learned every one of these. The people who get the most messages are the best at spotting them.

What makes a text sound like you

Look at the last ten texts you sent to people who matter. They probably share a few things.

The specific thing comes first. “Mark, saw the HVAC listing.” The first five words tell him why you are writing.

Short. One to three sentences. Under 40 words. You stop when the point is made.

Your phrasing, not good phrasing. Maybe you open with “man” or never use a greeting at all. Maybe you write all lowercase, skip periods, or say “been a minute” every time. Those habits are the signature. A draft with perfect punctuation from someone who never uses it is a tell to everyone who knows you.

You read the relationship. If it has been six months, you say so lightly instead of pretending it has not. “Mark, been a minute. Saw the HVAC listing, that one should move fast. How’s the rest of the year looking?”

One easy question or an open door. Something a friend could answer from the parking lot.

Compare that to the bot version above. Same person, same reason. One sounds like a relationship. The other sounds like a sequence.

How to teach any tool your voice

This works in ChatGPT, Claude, or anything that drafts. What I tell people is simple: show it, do not describe it.

1. Give it real texts, not adjectives. “Casual but professional” produces nothing useful. Paste in ten or more messages you actually sent to peers. The model copies what it sees far better than what it is told.

2. Give it how you talk. Record yourself texting a friend out loud for half a minute, transcribe it, and paste it in. Spoken language carries your openers, your rhythm and your favorite words in a way polished writing does not.

3. Give it your edits. Every time you fix a draft, paste the before and after back in. “Here is what you wrote, here is what I sent.” Over a few weeks the pattern of what you cut becomes the rule.

4. Write your own banned list. The generic one is a start: “hope you’re doing well”, “just checking in”, “circle back”. Add yours. If you would never say “awesome”, tell it.

5. Set a hard word limit. Say 40 words. Without a number, every model drifts long.

6. Fence in the facts. Tell it to use only what is in the post or your notes, and never to state anything about you that you did not give it. A draft that says “with the operators I work with” when you do not work with operators is worse than a bot greeting.

That will get you drafts you can send. The catch is that you have to carry all of it into every chat, keep it current, and remember to paste the edits back. That is the gap covered in ChatGPT as a personal CRM: the writing is easy, the system around it is the work.

How FRONT OF MIND does it

I built the voice into FRONT OF MIND, and it keeps learning without you pasting anything.

Your texting voice. In Settings, you record about 30 seconds of yourself texting a friend out loud. It is transcribed, and every draft, rewrite and intro ask is written to match it: the same casing, the same kind of opener, your level of punctuation, and one or two of your actual phrases where they fit.

Your edits. When you change a draft before sending, it learns what you cut, add and reword.

Your words win. Your own voice wins on style. Every draft is still 40 words max, opens with something specific to them, never mentions research or AI, and never adds facts about the person that are not in what they posted or your notes. It never invents anything about you either.

Your own news, used sparingly. You can tell it what is new with you, and it can read your own feeds. A draft uses that only when it is a real bridge to what the other person just posted. Most messages do not use it at all.

And nothing sends on its own. Every draft waits for your tap, and you can edit it, hold the mic and say it your way, or skip the person for today. The full answer on that is in will an AI relationship app send messages without me.

The test you should run

Open the app on a real morning. Read the first three drafts. Ask one question of each: would I send this to this person, as written?

Here is how I would judge it. If the answer is no, change it. That edit is what teaches it. If the answer is yes, you just saved the hardest part of staying in touch, which is finding the words when you are between calls. For examples of where those words go after an event, see the follow up text after a conference.

FRONT OF MIND opens on your 10 minutes for today, tells you who to text and why, and writes the draft in your voice. Your contacts are encrypted, no one on our team looks at them, and you can export or delete anytime. Start free with your first 25 contacts and see whether the first draft sounds like you.

See how it works