Is it okay to use AI to help you write a text to a friend?

You have not spoken to this friend in a year. You open the message thread, type a sentence, delete it, type another, delete that one too, and eventually give up and open a chatbot instead. "Help me write a text to an old friend, we have not talked in a year, I do not want it to sound like an apology or a guilt trip." It hands you back three sentences that are warmer and better-put than anything you wrote yourself. You copy one in, adjust a word, and just before you hit send, a new and oddly specific discomfort arrives: if a machine helped you find the words, is this still actually from you?

What the discomfort is actually about

It is worth being precise about the worry, because "using AI feels fake" is not quite what is being objected to. Nobody feels this way about a spell-checker, or about asking a partner to read a draft before you send it, or about copying a line from a book because it says something you meant but could not phrase. The unease is narrower than "getting help." It is the suspicion that the caring itself — the part that decided this person was worth the effort of reaching out — somehow got outsourced along with the sentence structure, and that what arrives on your friend's phone is closer to a form letter than to you.

That suspicion deserves to be taken seriously rather than dismissed, because it is sometimes right. It is just not right for the reason people usually assume.

Two things that get bundled into one worry

The confusion mostly comes from treating "who wrote the words" and "who decided to send them" as the same question, when they are not.

The decision to reach out — that this particular person has been on your mind, that a year of silence has started to feel wrong, that today is the day you do something about it — is a decision no tool made for you. You brought that to the blank message box before you ever opened a chatbot. It is the part that was actually costly, and it is entirely yours.

The phrasing — finding a sentence that says "I have thought about you and I am sorry it has been this long" without sounding stiff, or apologetic in a way that puts the burden on your friend to reassure you — is a separate, much smaller skill, and it is the one that is genuinely hard for a lot of people under pressure, at 11pm, on the fourth deleted draft. Getting help with that part does not reach back and undo the first part. The caring came first. The tool only helped it get out the door.

Where it works well

The cases where this genuinely helps are the ones where the obstacle was never emotional, it was mechanical: you knew exactly what you wanted to say and could not find a shape for it that did not sound like an accusation, an apology tour, or a greeting card. A chatbot is good at exactly this narrow thing — offering phrasing, softening a sentence that reads harsher than you meant, suggesting an opener that is not "hey stranger 😅" for the fifth time this year. Used this way, it is doing the same job a friend does when you show them a draft and ask "does this sound weird," except it is available at midnight and does not require explaining the whole backstory first.

Where it goes wrong

It goes wrong in one specific case: when the message is asked to invent a feeling rather than phrase one that already exists. If you type "write something that sounds like I've missed her" while genuinely unsure whether you have, the tool cannot tell the difference, and it will happily produce warmth on request. Your friend, on the other hand, often can tell the difference, not because AI writing has some telltale style, but because a message performing warmth reads slightly differently from one describing it — it reaches for a stronger adjective where the true version would have reached for a specific memory. The failure is not the tool. It is asking it to supply something that was not there to begin with, which was never a problem you could outsource your way around, with or without AI.

There is a second, quieter failure worth naming: pasting the output in unedited. Even a good draft was written by something that does not know this friend, has not seen the year that passed, and cannot know which detail would actually land. A message that could have been sent to anyone tends to read that way, however well-constructed the sentences are. The fix is small — swap in one real, specific detail only the two of you would recognize — and it matters more than any amount of polishing the tone.

A version that keeps it honest

• Start from something true, even if it is one sentence. Give the tool your actual reason for reaching out, not a request to sound like you have one. "I keep meaning to check in and never do it, want to fix that" produces something you can stand behind. "Make me sound like I never stopped thinking about her" produces something you might not.
• Use it for structure, not for content you do not have. Phrasing, tone, trimming something too long — all fair game. Inventing the memory, the apology, or the affection is not, because those are the only parts your friend actually came for.
• Put back one thing only you would know. A shared reference, an actual detail from the last time you spoke, the real reason you thought of them this week. This is what turns a well-written message into your message.
• Read it once as if you were the one receiving it. Not for grammar — for whether it sounds like the friend they remember, or like anyone.

The honest caveat

None of this means a tool can manufacture a friendship that has actually ended, or turn a message you do not mean into one that lands as sincere. If the discomfort you feel is really about whether you want to reconnect at all, better phrasing will not resolve it, and it is worth sitting with that question honestly before you open any chatbot. But if the answer is yes — you do want to reach out, you just could not find the words on your own tonight — getting help with the sentence is not a smaller or less real gesture than struggling through one alone. The part that mattered happened before you opened the app at all.