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.

Can people tell AI wrote a message? Strangers mostly cannot. The people in your phone who matter most usually can, because they are not checking the grammar. They are checking whether it sounds like you.

Here is the scene. A banker you have known for six years gets a text from you. Four sentences. A greeting, a compliment on his “insightful post,” a line about “reconnecting,” and a question that sounds like a survey. Every word is correct. He reads it twice, and something is off. He does not reply. He does not tell you why. You never find out it was the text.

That is the fear, and it is the right fear. One text that sounds like a bot can cost you a relationship worth six figures, and you will never see the bill.

What the research actually says

Two studies are worth knowing, because they point in a direction most people miss.

The first, published in PNAS in 2023 by Maurice Jakesch, Jeffrey Hancock and Mor Naaman, ran six experiments with 4,600 participants. People were unable to detect self descriptions written by AI in professional, hospitality and dating settings. Their gut rules were wrong. They assumed first person pronouns, contractions and family topics meant a human wrote it.

So strangers are bad at spotting AI by the words alone.

The second, published in Scientific Reports in 2023, looked at AI smart replies in conversations. When people believed their partner was using AI suggestions, they rated that partner as less cooperative and felt less affiliation toward them. When partners actually used more of the suggested replies, the other side rated them more positively.

Put those together and the risk is clear. The tool is not the problem. Suspicion is. The moment someone thinks you outsourced the message, you lose warmth with them, even if the message was fine.

My read: the studies tested strangers. Your contacts are not strangers. They have years of your texts to compare against. They do not need to detect AI. They only need to notice that this does not sound like you.

What gives an AI text away

The tells are not technical. They are about fit.

Length. You text a banker in one line. The draft runs four sentences. Length alone tells him someone, or something, else wrote it.

Formality. “I hope this message finds you well.” “I wanted to reach out.” Nobody you have shared a dinner table with talks to you that way. A text that reads like an email reads like a template.

Generic praise. “Great insight.” “Really resonated with me.” “Love what you are building.” These words fit anyone, so they prove nothing. Praise that could go to 200 people tells the reader it did.

No specific detail. This is the big one. A real text from you carries something only you would know: the deal you talked about at the dinner, his partner’s name, the town he was moving to. A draft built only from a public post has nothing private in it. It is attention without memory.

There is a longer list of tells, with examples, in will AI drafted texts sound like me. The short version: a person who knows you notices when the voice is wrong, not when the words are wrong.

How a draft becomes yours

I treat every draft as a starting point, never a finished message. Here is the ten second pass.

1. Cut the opener. If it starts with a greeting and a pleasantry, delete both. Start with the thing.
2. Cut it in half. Then see if it still works. It usually does.
3. Add one detail only you know. Something from your last real conversation. One clause is enough. “Did the Tampa office ever open?” beats any compliment.
4. Write it the way you write. If you never use exclamation points, take them out. If you write in lowercase, write in lowercase. Your habits are your signature.
5. Read it as them. If you would not say it out loud across a table, change it.

Compare the two versions of the same note to that banker.

The bot version: “Hi Dan! Hope you’re doing well. I really enjoyed your insightful post about the lower middle market. Would love to reconnect and hear what you’re working on!”

The edited version: “Dan, saw the post on the lower middle market. You called that at dinner in March. Still seeing it on your side?”

Same reason to reach out. One sounds like a sequence. The other sounds like a person who remembers.

The draft is not the risk. Sending blind is.

Most people who get burned by AI text did one thing wrong. They let the tool send, or they sent without reading. The words were never the problem. The missing step was the human one.

So the rule I would hold any tool to is simple. It drafts. You decide. Nothing goes out until you have read it, changed what needs changing and tapped send yourself. If a tool can send on its own, it can burn a relationship on its own. More on that in will an AI relationship app send messages without me.

How FRONT OF MIND handles it

The app was built around the fear in this article.

Every draft is a text style message, 40 words max. Short enough to sound like you, short enough to edit in seconds.

Nothing sends without your tap. You read it, change it or rewrite it, and you send it yourself.

It learns from your edits. What you cut and what you add shapes the next draft, so it moves toward how you actually text instead of how a model thinks people text.

Each draft is built from a real, dated reason: they posted, it is their birthday, you met at an event and never followed up, a deal in your pipeline has been quiet 7 days. Posts are only shown if they are confirmed to belong to that person, so you never congratulate the wrong Dan.

Your contacts are encrypted. No one on our team looks at them. Export or delete anytime.

Start free with your first 25 contacts and read the first drafts yourself. If they do not sound like something you would send, you will know in a minute. See how it works

See how it works