Which tool to choose
Any AI assistant does the job; what changes is what you ask it depending on how angry you are.
- You just want an opinion before sending: the general AI assistant. You paste the message and ask how it sounds and what you risk. The fastest filter, to keep handy for delicate messages.
- You also want a calmer rewritten version: the same assistant, asking it to rephrase while keeping the point but removing the anger. Handy when the message has to be sent anyway, but not like that.
- You're very fired up and risk a lot: use the AI to dump your outburst (write everything to it, not to the recipient) and then build together the version you'll actually send. The outburst goes nowhere, the message leaves cold.
How to do it
- Don't send. This is the step that matters most of all: between your finger and the send button, put the AI.
- Paste the message as it is, anger included, and ask for an outside reading. The operational syntax:
I'm about to send this message to [who: boss, client, loved one] while I'm angry. Before I send it: tell me how it sounds to whoever receives it, what consequences it risks having on the relationship or on the job, and which sentences I'd be better off cutting. Then rewrite it for me, keeping my point but in a way I won't regret tomorrow. The message is: [paste it].
- Read how it "sounds from the other side": that's the thing you can't see in the heat of the moment, because you only hear your own reasons.
- Choose: don't send at all, send the rewritten version, or wait until tomorrow. Often, reread with a cool head after thirty seconds, you don't send the message at all.
- Check: before pressing send on the new version, ask yourself whether you'd send it an hour from now too. If yes, go. If you hesitate, let it settle: important messages rarely get worse for an hour of waiting.
If you don't feel like opening the AI in the moment of anger, at least give yourself the postponement rule: save the message in drafts and commit to not sending it until you've had the AI read it. The filter only works if you give yourself the time to use it; haste is exactly what you want to defuse.
A concrete example
A client writes accusing you of a delay that isn't your fault. On impulse you write a cutting reply that makes clear who was wrong. Before sending it you paste it into the AI with the prompt above. The AI tells you the message sounds defensive and accusatory, that it risks ending the relationship even though you're right, and that three sentences in particular sound like an attack. Then it rewrites it for you: same content, no accusations, a proposed solution instead of a defense. You send the cold version. The client replies with an apology. The one you'd written in the heat of the moment would have lost the client even though you were right.
When it does NOT work (and how to fix it)
If you're so angry that you ignore the AI's opinion
The filter is of little use if you decide to send the venomous message anyway. Fix: impose a physical rule, not a willpower one. Move the message to drafts and give yourself a deadline ("not before tomorrow morning"). Anger drops on its own with time; the filter gives time, and time does the rest.
If the rewritten version loses your point and becomes too soft
The AI sometimes softens until it empties the message. Fix: tell it "keep the firmness on the substance, remove only the personal aggression". You can be clear and decisive without being offensive: ask for exactly that middle path, not a message that apologizes for existing.
If the message concerns a serious situation (threats, abuse, legal matters)
Here it's not a question of tone but of protection. Fix: for serious situations don't improvise, neither in the heat of the moment nor with the AI alone; keep the evidence, don't delete anything, and turn to someone competent (a lawyer, the authorities, a trusted person). The AI can help you put the facts in order, not handle on your own something that's beyond you.
A tip from someone who really uses it
Use the filter especially when you're convinced you're right. It's precisely when you feel you're on the right side that you write the most damaging messages, because being right makes you feel entitled to be harsh. Being right on the substance and losing the relationship over the form is the most common and most avoidable disaster: the filter works best when you think you don't need it.
Frequently asked questions
Isn't it excessive to have the AI read every message I write?
Not every message: only the ones written under strong emotion and aimed at people who matter. For ninety percent of messages it's not needed. The filter is for that ten percent where one less hour of clarity can cost you a client, a relationship or your job. There, thirty seconds are worth the potential damage.
Can the AI understand the tone and social consequences of a message?
It picks up well on register, aggression, sentences that will sound like an attack, and the typical consequences of certain tones in relationships. It doesn't know your personal history or the unspoken things between you two: for that, you remain the judge. The AI gives you the outside reading you don't have in the heat of the moment; you add the intimate context.
Isn't sending the "AI-rewritten" version dishonest toward whoever receives it?
You choose and approve the words: the AI only helped you say with a cool head what you'd have said badly while heated. Communicating with care isn't deceiving, it's respecting whoever reads. The message stays yours in content and in the decision to send it; the only thing that changes is that you won't regret it.