How do you set a boundary over text without starting drama?

The draft has been open for twenty minutes. Version one says “please stop texting me after midnight” and reads like a slap. Version two apologises three times and asks for nothing. Version three is a paragraph about how tired you are, which they will answer point by point. You close the chat and send “haha no worries”, which is not true.

I co-founded Subtext, an app that reads how a message is likely to land before you send it, so I think about these drafts a lot. I am not a psychologist, and I have a stake in you using the app, so weigh what follows accordingly. I wrote earlier about whether setting boundaries is good for you at all, and the answer was mostly yes, with caveats. That piece also covers where the word comes from. This piece takes the narrower question. How do you word one over text so it lands as a limit and not as the opening line of a fight? Further down, a real run of Subtext reads an irritated reply to Mum, who has asked what the new job pays, with what it flagged and the three versions it offered.

My working idea is that most of what people call drama is not a difficult person. It is a limit nobody said out loud, followed by a message that was read colder than it was meant. I cannot show you a study that measures that share, so treat it as an idea and not a statistic. What I can show you is which parts of the fix have evidence behind them. Assertiveness can be taught¹. An opening line that uses “I” and also names the other person’s side was rated least likely to provoke a defensive reaction². And people overrate how well their tone travels in writing³. The rest, including every script below, is craft.

If you are frightened of the person you are writing to, skip to When not to text.

A boundary is about what you will do

The word as it is used today draws on three streams, and only one of them comes with trial evidence. Family therapists in the 1970s used it for the invisible lines between parts of a family, from too blurred to too rigid⁴. Behaviour therapists taught people to say no and to speak in the first person, and called it assertiveness training; that is the stream with the research¹. Then the self-help shelf turned it into scripts, most usefully Nedra Glover Tawwab⁵ and Melissa Urban⁶. Neither book reports outcome data, and both make a point the scripts often forget. A boundary describes your own behaviour, not theirs.

That gives you a test for any draft. Which of these three is it?

A sentence built on “if” is not automatically an ultimatum. “If you message after eleven, I’ll reply in the morning” is a boundary, because the consequence is a thing you will do. “If you message after eleven, don’t bother messaging again” is a threat dressed as one. Urban’s version of this distinction is the most quotable, but the three columns are a writing tool, not a validated classification.

The rest of the vocabulary matters less than people think. Assertive means direct and respectful of the other person’s rights⁷. Passive means hinting, over-apologising, or sending “lol no worries 🙃” about something that is a worry. Aggressive means blame, insults and commands. Passive aggressive is the sarcasm, the pointed silence, the “k.” I have a separate piece on that one.

“Avoiding drama” usually means saying it earlier

Researchers have tried to measure drama as a trait. There is a Need for Drama scale, twelve questions covering manipulation, blurting and a standing sense of being wronged, built on American online and student samples⁸. There is Karpman’s drama triangle, the Victim, Rescuer and Persecutor roles that people slide between, which began as a 1968 reading of fairy tales and got a validated questionnaire only in 2020⁹. I mention both so you know they exist. Neither is a tool for diagnosing your sister or your flatmate, and if you catch yourself scoring someone on them, that is the drama.

The useful idea in the triangle is the switch. You answer the 2am messages for weeks because you are kind. You answer them tired, then irritated, then you send “You are so selfish, I have work” at 2:14am. They reply “I was in a crisis, I guess I’ll never reach out to you again.” Now there is a fight, and the limit still has not been said. The drama came from the gap between the limit you held privately and the moment it came out as an accusation.

Conflict runs on reciprocity. Hostility tends to get hostility back, and the authors of the rating study on opening lines call it a potential downward spiral². So the opening line carries most of the weight, and the sooner you say the limit, before resentment builds, the less weight the line has to carry.

Two more habits count as drama even though they feel like the opposite. Routing a problem between two people through a third person, which family therapists call triangulation, is the group chat move of posting “some people need to stop sharing photos of me” instead of messaging the one person. And going silent to make someone anxious is not a boundary; it is a punishment. “I’m going to stop replying to this tonight, we can talk tomorrow” is a boundary. The difference is whether the other person knows what is happening.

Why a boundary text reads colder than you meant

You hear your own tone while you type. The reader does not. Justin Kruger and colleagues ran five experiments on this in 2005³. In the third, with 154 pairs of students, senders predicted their emails would be read the way they meant them 88.8 per cent of the time, and readers managed 70.4 per cent. In the second experiment, when the task was telling sarcasm from seriousness, email readers did no better than chance, while people who heard the same lines spoken got about three quarters right. These are American students and email, not you and your mum on WhatsApp, so hold the numbers loosely. Hold the direction tightly. I go through what this research does and does not show in what a text actually means.

It does not mean every text gets misread. When Pollmann and Roos asked both sides of real message threads between people who knew each other in 2025, readers on average took the tone the way the sender meant it¹⁰. Their messages were mostly friendly and the topics benign, though, and a boundary text is neither. It is short, it carries tension, and the reader brings a history the words do not show.

Kristin Byron’s model of workplace email predicts which way a misreading goes. Neutral messages get read as negative, and warm ones as flat¹¹. It is a theory paper, not an experiment, but it fits the boundary text problem well. You strip the warmth out to sound clear, and the reader fills the gap with something colder.

Small things carry the load. A one word reply ending in a full stop (“Okay.”) was rated less sincere than the same word without one, in texts but not in handwritten notes¹². A later study found the effect was no longer significant in longer messages, which the full stop piece goes through, so this is not a rule about punctuation, only a reminder that a short boundary text has very few cues and every one of them is read.

Then there is the reply that does not come. When researchers in Taiwan interviewed messaging users about reading messages without replying, the reasons people gave for not replying were mostly ordinary¹³. A read receipt is not an answer, and being left on read after a boundary message is the moment many people send a second, worse message.

So write for the coldest plausible reading. Put one warm clause in on purpose. Use whole sentences. Do not send it at 1am.

Three questions before you type

The drafts that go wrong usually skipped this part. Answer these first, in a note to yourself, not in the message.

• What do I actually need? Not “for them to be less annoying”. A time, a topic, an amount, a frequency. “No calls after ten.” “Not my relationship.” “Not money.” If you cannot name it, you have a feeling, not a limit yet, and the message will come out as a complaint.
• What will I do? This is the part that makes it a boundary. “I’ll reply in the morning.” “I’ll change the subject.” “I’ll leave the group for a while.” If the only consequence you can think of is something they must do, you are writing a request. Requests are fine. Just know which one you are sending, because a request can be declined and a boundary does not need their agreement.
• What am I asking of them, if anything? One thing. Not a list. Not the last four months.

Two checks on timing. If you are angry right now, the draft can wait; no study backs a particular number of minutes, but pausing before you send is common advice. And if the limit is new and the person is close, a sentence of reason helps. If you have said it before, it does not. Reasons are something to argue with.

A message that fits on one screen

The most direct test of opening lines for a conflict that I could find is a 2018 study at Edith Cowan University in Perth, in which 253 university students rated hypothetical openers that varied two things, “I” language against “you” language, and whether the speaker gave their own side, the other person’s side, both, or neither. Both mattered. The best rated statements did both at once, which the authors sum up as “I understand why you might feel that way, but I feel this way”². They were ratings of sentences on a screen, not real arguments, so the finding is about how a line reads, which, for a text message, is the whole problem. I go deeper into that study in how to tell someone they hurt you over text.

DBT skills training teaches a longer version of the same shape: describe the facts, say what you feel, ask for one thing, and keep both the relationship and your self-respect in view¹⁴. It was built for clinical settings and has barely been tested on its own, but I find the order useful.

Put together, a boundary text has four parts, and the whole thing should fit on one phone screen without scrolling:

1. Warmth, or their side, in one clause. “I know Sunday is the only day you can call.” Only if it is true. A fake one reads as fake.
2. The limit, as something about you. “I keep Sunday mornings for myself now.”
3. What you will do. “So I won’t pick up before twelve, and I’ll ring you back after lunch.”
4. Reassurance. “It’s not about you, I just need the slow start.” Then stop. No apology chain, no list of past grievances, no “ok?” at the end.

“I know Sunday is the only day you can call. I keep Sunday mornings for myself now, so I won’t pick up before twelve, and I’ll ring you back after lunch. It’s not about you, I just need the slow start.”

That is about forty words. There is no “always” in it, no “you need to”, and no account of how tired you are. Short usually beats long, not because a study proved it (I found no study comparing short and long boundary texts) but because every extra sentence is a thing to argue with. The rule I use is minimum sufficient context, not brevity for its own sake. A parent getting this for the first time can have one sentence of reason. A friend who has heard it before cannot.

Six boundary texts, three temperatures each

Warm, direct and firm are three ways to send the same limit, not three levels of virtue. Warm fits a close relationship hearing this for the first time. Direct fits most people most of the time. Firm is for the third round. None of these sentences has been tested, and of the ingredients only the opening line has, in the rating study above.

Flaky plans. Before: “Wow ok. Guess I’ll take the silence as an answer.”

• Warm: “I know things come up. Last minute changes are hard for me, so next time I’ll only head out once we’ve confirmed.”
• Direct: “If plans aren’t confirmed before I leave, I’ll assume they’re off.”
• Firm: “I’m going to stop planning around maybes. Send me a time and I’ll tell you if I’m free.”
• What changed: frustration becomes a rule you can keep, and the passive aggressive sign-off goes.

Late night messages. Before: “can u PLEASE stop texting me at 2am”

• Warm: “I love hearing from you. My phone goes on Do Not Disturb at eleven, so I’ll reply in the morning.”
• Direct: “I don’t reply after eleven. I’ll get back to you the next day.”
• Firm: “I’ll mute the chat after eleven and reply the next day. Call if it’s an emergency.”
• What changed: a demand about their behaviour becomes a description of yours, which is what makes it a boundary.

Money. Before: “I can’t believe you’re asking me again after last time.”

• Warm: “I’m sorry things are tight. I can’t lend money, but I can help you look at options this weekend.”
• Direct: “I’m not able to lend money right now.”
• Firm: “My answer is no, and I’m not going to revisit it.”
• What changed: the decision stays apart from any argument about character, and there is no reason offered that can be bargained with. A standing rule (“I don’t lend to friends”) works the same way; it takes the personal rejection out.

A parent who asks too much. Before: “Why do you always need to know everything about my life??”

• Warm: “I know you ask because you care. I’d rather keep my finances private, but I’m happy to tell you about the new job.”
• Direct: “I’m keeping money private. Ask me anything about the job itself.”
• Firm: “That’s something I keep private. If it comes up again I’ll change the subject.”
• What changed: “always” and the accusation go, the warm version acknowledges their motive, close to the opener that rated best in the study above, and the follow-through is yours.

The group chat. Before, posted to everyone: “Some people need to stop posting stuff about me without asking.”

• Warm, by private message: “Hey, could you check with me before posting photos of me? I’d feel better about it.”
• Direct, in private: “Please don’t post photos of me without asking first.”
• Firm, in private: “I’ve asked to be asked first. I’m muting the chat for a while.”
• What changed: the problem leaves the audience. One person, in private, by name, instead of “some people” in front of twelve.

A friend who only vents. Before: “You only ever message to complain. I can’t be your therapist.”

• Warm: “I care about what you’re going through. I’m drained tonight. Can we talk tomorrow, and do something fun this weekend?”
• Direct: “I can listen for about ten minutes, then I need to switch topics.”
• Firm: “I’m not able to keep going over this one. Have you thought about talking to someone who can help more than I can?”
• What changed: a verdict on the friendship becomes a limit on tonight. The firm version points somewhere instead of leaving. Venting is not a crime. Research on co-rumination, going over the same problem together again and again, links it to closer friendships and, for some young people, mostly girls, to more depressive symptoms; a 2023 Swedish study found one group who got the closeness without the cost¹⁵. You are allowed to be full. Whether venting helps at all is its own question.

Three of these, the warm and direct ones for the group chat and the warm one for the friend who vents, are requests, and that is fine. A request is the friendlier first step, and it turns into a boundary when you add what you will do if it is ignored. For work, see how to say no to extra work.

When they push back

I could not find a study of what to text after someone argues with a boundary. What follows is the assertiveness training playbook applied to a phone, and it comes down to one sentence. Acknowledge once, restate once, add nothing new, then do what you said you would do.

• Defensive or hurt. Acknowledge the reaction without taking the limit back. “I get that this feels abrupt. I’m still keeping evenings offline.”
• Guilt trip. Name the feeling, keep the answer. “I hear that you’re disappointed. I still can’t come tonight.” Do not debate whether you are selfish; that is a different conversation and you did not start it.
• Bargaining. If there is new information and you want a compromise, offer one alternative. If not, use the broken record: the same limit in nearly the same words. “I understand it would only be an hour. I’m still not available.” The temptation is to supply a better reason each round. Every reason is a new handle.
• The same argument again. “I’ve answered this, so I’m going to leave it here for now.” Then leave it there.
• Silence, or read and not answered. Do not chase. In the interview study above, most reasons for not replying were ordinary¹³. Act on the limit you stated. If the relationship is close and days have passed, send one light check-in, and if that goes unanswered too, call.

Switch channels after three rounds of misreading, when the topic is grief, health or money, or when you cannot understand them without hearing their voice. Voice carries the tone that text strips out, and in Kruger’s sarcasm experiment listeners did far better than readers³. That does not prove a phone call fixes a real argument. It does mean a fourth text rarely will. What to text after an argument covers the repair side.

When not to text

Everything above assumes two adults who are safe with each other and can both walk away. If the person monitors you, threatens you, cuts you off from other people, or keeps contacting you after you have asked them to stop, this is no longer a wording problem, and a better phrased boundary will not solve it. In England and Wales, repeated controlling or coercive behaviour in an intimate or family relationship is a criminal offence¹⁶, and the Home Office guidance on it is built around risk and the victim’s safety, not around the victim communicating more clearly¹⁷. The Suzy Lamplugh Trust recommends not engaging with a stalker at all¹⁸; the US National Domestic Violence Hotline advises saving texts, emails and voicemails and warns that replying is likely to be read as the tactic working¹⁹.

You do not owe anyone a confrontation, in person or in writing. Talk to someone trained in this first. In an emergency, call 999 in the UK, 911 in the US, 110 for the police in Germany, or 112 anywhere in the EU.

• UK: National Domestic Abuse Helpline, 0808 2000 247, free and open 24 hours. National Stalking Helpline, 0808 802 0300, free, open on weekdays.
• US: National Domestic Violence Hotline, 800-799-7233, or text START to 88788.
• Germany: Hilfetelefon Gewalt gegen Frauen, 116 016, free, 24 hours, with interpreters.
• Anywhere else: Find A Helpline lists local services.

Please do not paste these conversations into Subtext or any other AI tool. A pasted draft or screenshot leaves your phone, and anyone who can open your phone can open the app. Notes saved to Memory, such as a profile for the person, stay until you delete them, and anyone who opens the app can read them.

There is a gentler version of “don’t text” too. Some conversations need a voice: the ones about grief or illness, the ones with a parent in a family where deference matters, the ones that have already gone three rounds. Directness is not a virtue in itself. A meta-analysis of studies on face and conflict style found that the preferred way of handling conflict differs across cultures²⁰, and indirectness is not the same as weakness. “I know you’re asking because you care; I’m keeping that private for now” carries the same limit as “I won’t discuss that”. One of them keeps Sunday lunch.

Where Subtext fits, and when you shouldn’t use it

Subtext helps with a narrower part of this than the rest of the article covers, which is how the draft you have written will read. You paste the draft, or a screenshot of the thread so it reads both sides, and it marks the words doing the damage, names the feeling under the draft, and offers up to three versions, each marked safe to send or heads-up.

I ran the parent example from the list above through it, with a message from Mum, made up for the run, as the context: “So how much are they paying you at the new job? 😊”. On the draft “Why do you always need to know everything about my life??” Subtext returned the tags Generalization and Defensiveness and named the feeling as Irritation. The note on the first tag read “"always" and "everything" leave no room for nuance and can make her feel attacked.” After the versions it said what it had assumed: “I assumed you’d rather not share the pay but still want to be kind about it.”

The version labelled Close to yours, “Mum, why do you want to know so much about my life? 😊”, carried a heads-up, with the note “Softer than the original, but it still sounds like a complaint and she may feel brushed off.” By the test at the top of this article it is still a complaint dressed as a question, with no limit in it. Subtext flagged it for sounding like a complaint. It does not check whether a draft contains a limit, so that test stays yours. The second, labelled Light boundary, was marked safe to send: “Haha Mum, I’d rather keep the money side to myself. The job is going well though 😊”. That one says what you will do. The third, Warm and open, was close to the warm script in the list above, which I had written before the run. Every version added a smiley like the one in Mum’s message, though the draft had none, which you would want to check before sending. That is the app’s real output, unedited.

You can ask for any version shorter, warmer or more confident. Each one is a draft in your voice, not a template, and sending it is still your job; Subtext never sends anything.

What it cannot do is most of what this piece covers. It does not know the four months of history. It cannot tell you whether this person belongs in your life, or who is right, and it does not try. It reads your draft and how it may land, and it cannot know what the other person meant. A safe to send mark is a second reading of your words, not a promise about their reply. It is not therapy. And nobody independent has tested whether people who use it have fewer arguments. I would like that study to exist. It does not yet.

A boundary text often contains another person’s words, so a note on privacy. Nothing you paste into Subtext is used to train an AI model. A conversation deletes itself five days after you last use it, or 90 days if you pin it, and you can delete one straight away²¹. Notes saved to Memory stay until you delete them. Text still leaves your phone to be processed, so crop out names and numbers you do not need, paraphrase the other person where you can, and delete the thread when you are done. I read the training policies of thirteen writing tools if you want the long version.

Try Subtext in your browserDownload on the App StoreGet it on Google PlayGet the app

Questions people ask

Is it rude to set a boundary over text instead of in person? Not in itself. Text is a reasonable medium for a time limit, a declined invitation, a topic you will not discuss, or anything you want on the record. It is the wrong medium once the thread has gone three rounds, or when the subject is grief, illness or money and you need to hear them.

Should I explain why? Once, if the limit is new and the person is close. Not at all if you have said it before. Reasons are things to argue with, and a boundary is not a case you have to win.

They read it and didn’t reply. What now? Nothing. A read receipt is not an answer, and the reasons people give for not replying are mostly dull ones¹³. Do what you said you would do. If it is a close relationship and days have passed, send one light check-in, and if that goes unanswered too, call.

Is “I’ll stop replying tonight” the silent treatment? No, because you told them. The silent treatment is withholding contact to make someone anxious without saying so. A boundary says what is happening and when it ends.

Sources

Numbered in the order they appear above. Where a sample size, funding statement or replication could not be verified, the entry says so. Checked 24 October 2026.

1. Speed, B. C., Goldstein, B. L., and Goldfried, M. R. (2018). Assertiveness Training: A Forgotten Evidence-Based Treatment. Clinical Psychology: Science and Practice, 25(1), e12216. Narrative review of older clinical trials, mostly in clinical and student samples; no pooled effect sizes; nothing on texting. A 2026 meta-analysis of 12 trials found a moderate benefit for social anxiety; see whether setting boundaries is good for you.
2. Rogers, S. L., Howieson, J., and Neame, C. (2018). I understand you feel that way, but I feel this way: the benefits of I-language and communicating perspective during conflict. PeerJ, 6, e4831. 253 university students rating hypothetical openers, ethics approval from Edith Cowan University; no funding declared. Ratings, not observed behaviour. Earlier rating studies of I and you statements (Kubany and colleagues, 1992 to 1995) used smaller samples.
3. Kruger, J., Epley, N., Parker, J., and Ng, Z.-W. (2005). Egocentrism over e-mail: Can we communicate as well as we think? Journal of Personality and Social Psychology, 89(6), 925 to 936. Five experiments with US undergraduates. The 88.8 and 70.4 figures are Study 3, with 154 University of Illinois pairs; the chance-level sarcasm result (56 per cent, not reliably above the 50 per cent chance level, against the 78 per cent senders expected) and the voice comparison (73 per cent) are Study 2, with 29 pairs analysed. Funded by the University of Illinois and NSF grant SES-0241544. Email tasks, not personal texting.
4. Minuchin, S. (1974). Families and Family Therapy. Harvard University Press. Clinical theory, not an experiment.
5. Tawwab, N. G. (2021). Set Boundaries, Find Peace. TarcherPerigee. Practitioner book; no outcome data.
6. Urban, M. (2022). The Book of Boundaries. Dial Press. Practitioner book; no outcome data.
7. Mayo Clinic. Being assertive: Reduce stress, communicate better. Clinical education, not a trial.
8. Frankowski, S., Lupo, A. K., Smith, B. A., Dane’El, M., Ramos, C., and Morera, O. F. (2016). Developing and testing a scale to measure Need for Drama. Personality and Individual Differences, 89, 192 to 201. Twelve items, three factors; MTurk samples (69, 120, 245, 359, 494) and a University of Texas at El Paso student sample (261); self-report, no funding statement found. Not a diagnostic tool.
9. Karpman, S. B. (1968). Fairy Tales and Script Drama Analysis. Transactional Analysis Bulletin, 7(26), 39 to 43. Clinical heuristic. The validated Drama Triangle Scale is Lac and Donaldson, published online in 2020, Journal of Interpersonal Violence, 37(7-8), NP4057 to NP4081, 2022.
10. Pollmann, M. M. H., and Roos, C. A. (2025). “I get u”. People correctly interpret the tone of text messages and emails. Computers in Human Behavior Reports, 18, article 100689. Mostly positive messages between people who knew each other; recent and not yet replicated.
11. Byron, K. (2008). Carrying too heavy a load? The communication and miscommunication of emotion by email. Academy of Management Review, 33(2), 309 to 327. A theoretical framework about workplace email, not an experiment.
12. Gunraj, D. N., Drumm-Hewitt, A. M., Dashow, E. M., Upadhyay, S. S. N., and Klin, C. M. (2016). Texting insincerely: The role of the period in text messaging. Computers in Human Behavior, 55, 1067 to 1075. Undergraduate lab ratings.
13. Chou, Y.-L., Lin, Y.-H., Lin, T.-Y., You, H.-Y., and Chang, Y.-J. (2022). Why Did You/I Read but Not Reply? IM Users’ Unresponded-to Read-receipt Practices and Explanations of Them. CHI 2022. 25 interviews in Taiwan, June to July 2020; funded by the Higher Education Sprout Project and Taiwan’s Ministry of Science and Technology. Qualitative; no pair was interviewed from both sides, so it cannot say how often guesses were wrong.
14. Linehan, M. M. DBT Skills Training Handouts and Worksheets. Guilford Press. Clinical teaching materials; the interpersonal effectiveness skills have rarely been tested on their own. https://www.guilford.com/add/linehan7_old/lin-p-2teaching.pdf?t=1
15. Tilton-Weaver, L., and Rose, A. (2023). The trade-offs of co-ruminating with friends: A profile analysis. Journal of Research on Adolescence, 33(3), 957 to 972. 2,767 Swedish adolescents aged 12 to 16; measured depressive symptoms. Consistent with Rose, Carlson and Waller (2007), who found the costs in girls. Youth samples.
16. Serious Crime Act 2015, section 76. legislation.gov.uk.
17. UK Home Office (updated 27 July 2023). Controlling or coercive behaviour: statutory guidance framework.
18. Suzy Lamplugh Trust (updated 7 September 2026). Stalking help and advice.
19. US National Domestic Violence Hotline. Stalking Safety Planning.
20. Cardon, P. W., and Okoro, E. A. (2010). A Meta-Analysis of the Cultural Propositions about Conflict Management Styles in Face-Negotiation Theory. Journal of Rhetoric, Professional Communication, and Globalization, 1(1), 35 to 59. Tests individualist against collectivist preferences; averages across studies, not scripts.
21. Subtext. Terms and privacy. Privacy policy last updated 22 October 2026. First party.