Don’t get bot hurt. Get bot even.

This month, Bot Hurt has been looking at the new rules of living with AI without making everything weird.

So far: Do you need to tell people when the bot helped? When does someone deserve to know they are dealing with AI? And when does efficiency stop being helpful and start becoming rude?

This week: When does AI stop supporting the human and start replacing the effort?

The robot should slowly back away from the greeting card

AI can write a thank-you note.

It can smooth out an apology.

It can help you respond to the message that has been sitting unanswered for three days because your brain opened it, panicked and quietly left the building.

That can be useful.

Sometimes the bot helps you find the words.

Sometimes finding the words is the whole point.

The bot may actually help

Here is the inconvenient little twist.

AI may be pretty good at this.

A 2025 iScience study found that AI-assisted messages built trust just as well as unaided messages, but in less time. In the experiments, some participants could use predictive text assistance while writing trust-building messages.

So, yes. The bot may help people sound clearer, warmer and more trustworthy.

The study also found that AI-assisted messages were slightly less authentic, but showed more warmth and complexity.

That is the weird little trade-off.

The bot can make the message better.

But better is not always the only thing people are measuring.

^{Source: Steven Bartlett, @thediaryofaceopodcast, Instagram}

The polished sentence problem

There is a reason people ask AI to help with emotional writing.

The blank page can be brutal.

What do you say when someone is grieving? How do you apologize without making it worse? How do you thank someone in a way that does not sound like a scented candle label?

The bot is good at removing the wobble.

It can turn “I’m sorry this got weird and I don’t know what to say” into something that sounds like it attended a conflict-resolution seminar and brought handouts.

Sometimes that is exactly what you need.

Sometimes the wobble is the human part.

The effort is part of the signal

Personal messages are not only judged by their final wording.

They also carry a little invisible receipt.

Did this person think about me?

Did they mean it?

Did they just toss this into the feelings microwave and wait for the ding?

Nobody wants a forensic audit of a birthday text. Please do not make your loved ones attach an AI disclosure statement to “Hope your day is amazing.”

But if the message is supposed to carry care, effort still matters.

A bot can help with the envelope but it probably shouldn’t become the whole card.

Put yourself back in

This is not an argument for suffering over every sentence.

People freeze. People avoid hard messages. People stare at the blinking cursor and suddenly decide the junk drawer has become a moral emergency.

A bot can help someone get unstuck.

It can suggest a structure. It can make a clumsy sentence clearer. It can remind you not to make the apology about yourself, which, frankly, some humans could use taped to the fridge.

The trick is to put yourself back in before you send it.

Add the specific memory. Keep the phrase that sounds like you. Delete the line that sounds like it was approved by the Department of Appropriate eelings.

A better test

Before using AI on a personal message, ask one question: Is the bot helping me say the thing, or helping me avoid the thing?

If it is helping you say the thing, fine. Let the bot hand you a flashlight.

If it is helping you avoid the thing, maybe the robot should slowly back away from the greeting card.

Final bot thought

Use the bot to help find the words.

Just make sure the words still have your fingerprints on them.

Bot Talk: The chatbot needs a passport

AI companies spent years telling us the technology was borderless.

Governments appear to be reading the fine print.

Anthropic said the U.S. government issued an export control directive requiring the company to suspend access to two advanced models, Fable 5 and Mythos 5, for any foreign national. That included Anthropic employees who are foreign nationals.

The company said the directive’s practical effect was that it had to disable those models for all customers to ensure compliance. Access to its other models was not affected.

Which is how the chatbot went from “ask me anything” to “please present your documents.”

Once an AI model is treated as sensitive enough for export-control-style restrictions, it starts looking less like a normal app and more like controlled infrastructure.

The bot used to need better manners.

Now it may need paperwork.

Bot Field Report

Thanks to reader Crystal for sending a clip of a robot trying to dance to Michael Jackson and losing a very public argument with stairs.

^{Source: @zo__! YouTube}

After last week’s humanoid robot story, this felt like an important reminder: The robots may be coming, but some of them still need knee pads.

🚀 Coming up next week …

Bot Hurt is taking next week off for the Fourth of July holiday.

The bots, obviously, are not.

We’ll be back July 9 with a slightly tweaked format: more Bot Talk up top, more timely AI weirdness and the same Bot Hurt attitude.

Until then, enjoy the long weekend, hydrate and feel free to let the chatbot help with the potato salad. Just maybe check the measurements.

Don’t get bot hurt. Get bot even.