How do you cancel plans by text without sounding flaky?

It is half past five and the table is booked for seven. You have decided you cannot go, and you have written the message three times. One draft stacks up apologies and exclamation marks. One promises to make it up to them without saying how. The shortest says “something came up”, which is true, and which you suspect will sound like the fifth time even though it is the first.

I co-founded Subtext, an app that reads a draft before you send it and shows how it may come across, so I have a stake in you trying it. Further down, Subtext reads a last-minute cancellation that gives a vague reason and promises to make it up, and I show the tags and versions it returned.

One cancellation is a change of plan, and I would not call it flaky. The one published survey I found built around cancelling plans asked people to imagine being cancelled on, so it records what they say they would want and not what happened between real friends¹.

Flaky is an everyday word, and the cancellation research I read does not use it, so I describe behaviour: how late the notice came, whether the message said plainly that the plan is off, whether it has happened before. Ghosting, ending a relationship by cutting off all contact², is something else, and why people ghost covers it.

How does a cancellation land?

Little research has tested how a cancellation lands, and the best evidence I found comes from people imagining one. Caron and colleagues asked 1,192 people how they would want a friend to cancel on them¹. Most were university undergraduates, the mean age was 22, and nobody was watched being cancelled on. Asked in their own words (1,134 usable answers), 59.1 per cent mentioned advance notice and 48.0 per cent being told quickly, by a call or a text. A truthful reason came up in 28.6 per cent of answers, a new date in 12.3 and an apology in 10.1. Those were spontaneous mentions, so a low share means few thought of it, not that few wanted it.

On the rating questions, people on average did not find being cancelled on very upsetting, though a best friend’s cancellation upset them most. The average was 2.85 out of 5 for a best friend and 1.85 for an acquaintance, and about six in ten rated a best friend’s cancellation at a moderate amount of upset or more¹. Most said it would not much affect the friendship, though the authors say that question may have been too vague¹. The survey asked about one imagined friend with no history, and only about annoyance, upset and offence. The survey does point at three things people objected to: very short notice, an invented excuse found out later, and a better offer as the reason.

In experiments on declined invitations, people deciding whether to say no overestimated how badly the inviter would react³, which saying no to a friend without over-explaining covers in full. A preprint posted before peer review, and labelled preliminary by its authors, reports the same pattern for cancelling, in scenario studies with 6,824 American and French participants⁴. Anticipated anger is a prediction, not knowledge. It is also no promise that your friend will not mind, because people can be understanding and disappointed at once.

What should a cancellation text say?

It should do five things, in this order, and the first four take about a sentence each. This is my own framework, built from what Caron’s respondents asked for and the repair research below, so use it to check a draft, not as a formula.

1. Say what is changing, and name the plan. “I need to cancel dinner tonight.” Not “I don’t think I can”, and not “maybe”.
2. Give the truth at the level you choose. One true sentence, or a plain statement that you would rather keep the detail private. Never an invented one.
3. Name the practical cost. “I’m sorry for the late notice”, plus any specific cost, such as a booking.
4. Offer only a real next step. A date you want and can keep, or when you will get back to them. If you do not want another date, offer none.
5. If it is urgent, check it arrived. If they may already be on their way, use the channel they will see and confirm. If a call goes unanswered, send the clear cancellation anyway.

I made up the texts here, and none of them is the single correct text. They are plain on purpose, so add a warm word of your own only if it is true. Each after keeps what is true in its before and drops anything invented. Where an after adds a date, an offer, a reason or an apology, keep that only if you mean it.

Before “Work might run over tonight, so can we see how the afternoon goes? Sorry, I know it’s a hassle.”

After “I need to cancel dinner tonight. Work may run over, so I can’t be sure of getting there. I’m sorry for the short notice. Could we try Thursday at seven?”

The before makes the friend wait on the afternoon and never says the plan is off. The after is a decision with a small true reason, names the short notice and proposes one date, the added offer here.

How much notice, and text or call?

Send it as soon as you know, and by text if they are likely to see it. Both are etiquette, not findings.

In Caron’s survey, respondents mentioned advance notice most, and for plans a week away the typical answer to when a cancellation would be most annoying was a few minutes before¹. That gives no threshold. The authors say their own question about how much notice to give may have been too vague. The sooner they know, the more they can do about a booking or other plans.

The closest experiments on channel are about other things. In one, people reconnected with an old friend by phone or email, and in others they chatted with a stranger by video, voice or text. Conversations that included a voice built stronger bonds and were no more awkward than text, though miscalibrated expectations could lead people to prefer text⁵. None of that was about cancelling, so at most it gives me a reason to consider a call when a cancellation costs the other person something. A call does not suit everyone, since in a survey of 245 autistic adults phone calls were unpopular, especially with services and unfamiliar people, and for friends and family face-to-face and written contact were preferred⁶. AuDHD and communication goes through that survey. On voice notes, see why a voice note transcript is not a text message.

Group plans are etiquette too. Tell whoever organised it first, because they carry the booking and the headcount, then the group.

Before (to the group chat, thirty minutes ahead) “Think I’m out for tonight, have fun without me 🙂”

After (to Imogen, who booked the table, then the group) “Hi Imogen, I can’t make dinner tonight and I’m sorry for the late change. Could you take me off the booking? If the restaurant charges for my place, tell me how much and I’ll cover it.”

The after tells Imogen first, asks for the practical change and says what the sender will do about any charge. Offer to cover a cost only if you will pay it, and check the booking’s rules first.

Do you have to give a reason?

I would give one true sentence or say plainly that you would rather keep the detail private. In Caron’s survey, respondents most often named health and family as reasonable reasons, 49.4 per cent each¹. The reason most often called inappropriate was a better offer, at 53.3 per cent. Giving no reason was named as inappropriate by only 1.3 per cent, but the authors caution that how often something was named is not a ranking, and that giving none might be among the most inappropriate.

I sort reasons into three kinds. A short true reason explains the change without an intimate account: “I’m unwell and can’t make dinner tonight.” A truthful but private one sets a limit: “Something personal means I need to cancel tonight, and I’d rather keep the details to myself.” If the real reason is a better offer, say that you would rather keep the details private, and do not call it personal unless it is. The third kind is a vague reason with an unclear next step: “Something came up, maybe another time”. What needs editing in that message is the uncertain next step, not the reason. The survey’s authors leave open whether people would prefer a vague reason or an unwelcome true one¹.

In experiments on replying to invitations, people overestimated how much an inviter would prefer a “maybe” to a clear no, because they underestimated how disrespected the inviter would feel⁷. The authors list a “yes” that warns you might cancel as untested, so I would still say whether the plan is off.

Before (to your parents) “Not sure we can face the drive this weekend, the week’s been mad. We’ll let you know nearer the time.”

After “We won’t be coming this weekend. The week has been mad and we can’t face the drive. We’re sorry to leave you planning around us, and we’d like to ring you on Sunday afternoon.”

The before leaves your parents planning around a maybe. The after states the decision with a short true reason and offers a call, which is no replacement for the visit.

Is a made-up excuse safer?

I would not make one up, though the evidence is indirect.

In Caron’s survey, people were asked how offended they would be if they later found that a friend had made up an excuse, such as being ill when they were not. The average was 3.59 on a scale of 1 to 5, and 83.1 per cent answered a moderate amount or more¹. That is an imagined reaction to one hypothetical, and the measure is offence, not trust.

Trust has been studied in money games. In Schweitzer, Hershey and Bradlow’s game, 262 participants, almost all aged 19 to 22, played for money against a scripted partner. Trust damaged by untrustworthy moves recovered after a run of trustworthy ones, but trust damaged by the same moves plus lies about future behaviour never fully recovered within the experiment, even after a promise and an apology⁸. The lies were false promises about returning money, so this shows a pattern in a game and forecasts nothing about friends. In Levine and Schweitzer’s game experiments, lies told to help the other player raised trust in the liar’s intentions and lowered willingness to rely on their word⁹. An excuse to cancel mostly spares the sender, which is a different case.

My inference is that an invented excuse adds a fact someone can ask about later. “Feeling better?” is an easy question for a true reason and a hard one for an invented one.

Before (you said yes to something you never wanted) “So gutted, I think I’m coming down with something. Probably best I skip Sunday, have a great day without me.”

After “I’m going to drop out of Sunday. I said yes too quickly. I’m sorry to change my mind this late. I hope the day goes well.”

The before invents an illness that has to be kept up on Monday. The after states the decision, owns the late notice and the earlier yes, and offers no new date. It may still be awkward to send.

How sorry should it sound, and what about another day?

My advice is to say sorry once, for the specific cost, and to offer another day only if you want one and can keep it.

The apology study I draw on is Lewicki, Polin and Lount’s. In two scenario experiments, 333 US online workers and 422 business students rated how effective the parts of a job candidate’s apology would be after a client’s tax return was filed wrongly, and a request for forgiveness was rated lowest of the six parts in both¹⁰. That was words only, about a work error, not a missed dinner, so I take two things from it: say what you are sorry for, and do not ask to be forgiven. How to apologise over text goes through the research. “I’m the worst, please don’t hate me” asks them to reassure you before the practical question is answered, which is my reading.

On another day, the survey’s authors call their respondents relatively split. The average importance of rescheduling was 2.82 out of 5, and 12.3 per cent mentioned it unprompted¹. No study I found shows that offering a date makes a cancellation land better, so the test is whether the offer is true.

Before (you are unwell and want to see this person again) “Hi, so sorry, tonight’s not going to work. Can we maybe do another time? Hope that’s ok.”

After “Hi, I need to cancel tonight. I’m unwell and wouldn’t be good company. I’m sorry for the short notice. I’d still like to see you. Does Saturday afternoon work?”

The after says the plan is off, gives the reason, names the short notice and proposes one time. If you do not want to meet again, offer no date, since I think a cancellation is a poor place for a verdict on the person.

What if the reason is low energy, anxiety or illness?

You can say so plainly, and you do not need a diagnosis to do it. I think needing a quiet night is an ordinary reason, though people differ. In Caron’s survey, 17.9 per cent named feeling tired or unwell without an illness as a reasonable excuse, and 18.1 per cent named not feeling like it as an inappropriate one¹.

I would name the limit and leave out the label: “I need a quieter evening”, “I can’t manage the journey today”, “my symptoms have flared”. Whether to name a condition is yours to decide. If the worry is anxiety, texting anxiety research covers what is known, and why a short reply takes an hour covers autistic burnout.

Before “I’m so sorry, I’ve got nothing left tonight and I’d be useless company. I hate doing this. Rain check?”

After “I need to cancel tonight. I’ve run out of energy and wouldn’t be good company. I’m sorry for the short notice. I’ll message you on Sunday about another evening.”

The before talks the sender down and leaves “rain check?” without a time. The after names the limit without a diagnosis, apologises once, and says when the sender will come back about a new date.

When it is the third time

Say that it is the third time, and make the next plan smaller.

The imagined friend in Caron’s survey had no history, and its authors list whether the cancelling friend often cancels, and with what excuses, as unmeasured¹. The preprint says its results may differ for repeated cancellations⁴. In the money game above, trust recovered after a run of trustworthy moves⁸. My inference for friends, and it rests on scripted strangers in a lab, is that a plan you keep does more than a better message.

So the text for a third time names the pattern, takes responsibility for the arrangement as well as the evening, and offers something smaller.

Before (you have cancelled on Noor twice, and the real reason is that you keep agreeing to more than you can manage) “This week’s gone mad again and I can’t believe I’m doing this to you. Let’s rebook soon, I promise this time.”

After “I have to cancel Friday supper, and I know it’s the third time. I’m sorry. I’ve been saying yes to more than I can manage. Could I come to you on Wednesday after work for an hour instead?”

The after names the pattern, says the real reason and offers an hour on a weekday, which you should offer only if you mean it. If the true answer is that you cannot see them for a while, say that, and say when you will get back in touch. If you have gone quiet since cancelling, why you can’t reply to messages you want to answer has more.

When they are upset, press for details or go quiet

What follows is editorial. It says the practical thing, adds nothing untrue and does not argue them out of being disappointed.

They are disappointed. Name the specific effect once and do not argue with it: “You’d set the evening aside and I changed it late. I’m sorry.” Then deal with any booking, cost or replacement plan.

They press for details. You can hold the limit without adding evidence: “I can’t make it, and I’d rather keep the reason private.” If the argument is really about the practical result, talk about that separately from the private reason.

They go quiet. A read receipt is not an answer, and left on read goes through what is known about it. If you need them to have seen it, one more message about logistics is enough: “Just checking you saw that I can’t make it tonight.” Do not ask for reassurance while they decide how to reply.

It turns into a row. If the reply is a guilt trip or an insult, restate the decision and the next practical step, and do not turn it into a contest over who is worse. Setting a limit over text has wording for pushback.

All of this assumes someone who is annoyed, not someone you are afraid of. If the reaction frightens you, if they threaten you or keep contacting you after you have said no, or if you are cancelling because of how they treat you, this is no longer a wording problem. You do not owe a frightening person an explanation, in person or in writing. If you are in immediate danger, call your local emergency number. The section on when not to text in Setting a limit over text covers threats, monitoring and stalking, and the section on when raising it isn’t safe in asking a partner to share the mental load lists helplines. Please keep your conversation with that person out of any AI tool, Subtext included.

Can an app help you cancel plans, and what can’t it do?

An app such as Subtext can help with the wording, and only the wording. Some websites and apps produce an excuse once you choose a situation and a style, and some compose a cancellation from lines written in advance. Phones now come with writing tools that can rewrite a draft in a chosen tone on supported models. Another kind reads the draft you wrote. I found no study of whether a generated excuse is believed or found out, or of whether any writing tool, mine included, changes how cancellations are received. In two experiments with online workers chatting in pairs about a policy question, people suspected of using smart replies (the suggested replies some apps let you send with a tap) were rated more negatively, which is an association, since suspicion was not randomised¹¹. People who used them were rated as closer and more cooperative¹¹. So my preference is to write the sorry myself and use a tool for a second reading. The experiments did not test that, and it describes the kind of tool I co-founded, so weigh it accordingly. How to sound better over text goes through the studies.

Subtext is that kind. It marks words in your draft that could read as vague, over-apologetic, guilt-tripping or an overpromise, with a short reason for each, and offers up to three versions, the first staying close to your wording, each marked safe to send or heads-up. The versions made from your draft carry an estimate of how much of your wording changed, and you can ask for any of them shorter, warmer or more casual. A reason, a date or an offer you did not give is meant to stay out of the versions, and the app says so under them when it notices one, though it can miss one. You paste the draft or share a screenshot of the conversation, and you copy the version you pick and send it yourself.

It cannot supply a believable reason, decide whether to cancel, or tell you whether the plan matters to the other person. It does not guess their mood, and the site’s FAQ says so. It has no link to your calendar, so it moves, cancels and books nothing, and it cannot make anyone reply. A safe to send badge is Subtext’s own estimate from the words in front of it, and a heads-up is a reason to read that version again, not proof of a problem.

What does Subtext do with a last-minute cancellation?

A last-minute cancellation to a friend called Ravi, made up for the run, reads “Hey Ravi!! so sorry, something’s come up and I don’t think I can do tonight 😬 can we rain check? I’ll make it up to you, I promise!!!” Subtext named the feeling Apologetic and returned three tags. Excess exclamation noted that the stacked exclamation marks “can feel anxious or over-the-top for a cancellation”. Overpromise called the promise “a big, vague promise you may not want to be held to”. Clarity said “Something’s come up” and “rain check” “give no sense of when you could reschedule”. All three versions were marked safe to send, with 25, 50 and 70 per cent of the draft’s wording edited, by the app’s own estimate, and all three keep the vague reason, so the app did not invent one.

The first, Close to yours, shortens the promise to “I’ll make it up to you” but keeps “I don’t think I can do tonight”, the hedge that step one warns against. The second, labelled “Warmer, with a next step”, ends “I’ll make it happen”, which I read as another open promise, and adds “I was looking forward to it”, a feeling the draft never stated. The third, Natural rewrite, proposes “sometime next week” and ends “Dinner’s on me.” Under the versions, Subtext’s closing line names “next week” and “dinner’s on me” as things “you didn’t say”, but not the second version’s added feeling, which that version’s note counts as a plus. Two version notes predict how the text will land, “Should land well with a friend” and “makes the cancellation easier to take”. Read them as the app’s estimate from your words, since it cannot know Ravi. That is the app’s real output, unedited.

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

A chat holds your friend’s words. Paste the draft and the one line it replies to, not the whole conversation, crop names and numbers out of a screenshot, and keep their health, money and private life out. Subtext’s privacy policy says nothing you paste in is used to train an AI model and that a conversation is deleted five days after you last use it, or 90 days if you pin it, and that notes you save to Memory stay until you delete them¹², though the text does go to AI providers. Which writing tools train on your messages has the long version.

The order to do it in

Decide that you are cancelling and say so, because a “maybe” leaves them holding the plan. Send it as soon as you know and keep every reason and promise true.

Before you send, ask the draft six questions. They are a prompt to look again if you worry it will sound flaky.

• Does the first sentence say the plan is off, and which plan?
• Am I sending it as soon as I know, and not after finding the perfect reason?
• Is every reason, promise and warm word in it true?
• Have I named the practical cost, such as the late notice or a booking?
• If I offer another day, do I want it and can I keep it?
• If this has happened before, have I said so and changed the plan?

If you have rewritten it more than twice, I would check it against these and send it.

Questions people ask

Is it rude to cancel plans by text? In itself, I would say not. In Caron’s survey, respondents most often asked for notice and a quick call or text, and only 8.3 per cent asked for a longer conversation¹.

How late is too late to cancel? The survey gives no threshold, and its authors say their own question about notice may have been too vague¹. Send it as soon as you know.

Should I say I am ill when I am not? I would not. In Caron’s survey, 83.1 per cent said they would be at least moderately offended to find out a friend had made up an excuse, an imagined reaction¹. I think an invented illness has to be kept up afterwards.

Is there an app that helps you cancel plans by text without sounding flaky? Some write an excuse for you and some read the message you wrote. Subtext reads your own draft, marks vague wording and overpromises, and offers up to three versions, but it cannot supply a reason, and sending the text is up to you.

What if I am the one being cancelled on? I would reply to the practical part first, and say once, plainly, if it hurt. In Caron’s survey, a best friend cancelling upset people most¹. If it keeps happening, the limit is yours to state.

Sources

Numbered in the order they appear above. Where a figure could not be verified against the primary source, the text says so. Checked 6 November 2026.

1. Caron, S., Thomas, J., Torres, A., Oh, J., and Chopik, W. J. (2023). How to Cancel Plans With Friends: A Mixed Methods Study of Strategy and Experience. Collabra: Psychology, 9(1), 57549. Cross-sectional survey of 1,192 people, 82.6% university undergraduates and 17.4% Amazon Mechanical Turk workers, ages 18 to 80 (mean 22); open-ended and rating questions about how respondents would want a friend to cancel; not preregistered, no funding, convenience samples. Imagined preferences, not observed cancellations. Chopik is an associate editor of the journal and took no part in the review of this article.
2. Freedman, G., Powell, D. N., Le, B., and Williams, K. D. (2019). Ghosting and destiny: Implicit theories of relationships predict beliefs about ghosting. Journal of Social and Personal Relationships, 36(3), 905 to 924. Two online surveys of US adults (554 on Mechanical Turk, 747 on Prolific), the second including friendships; self-reported experience.
3. Givi, J., and Kirk, C. P. (2024). Saying no: The negative ramifications from invitation declines are less severe than we think. Journal of Personality and Social Psychology, 126(6), 1103 to 1115. Experiments on how people predict an inviter’s reaction to a declined invitation; only the abstract was read (paywalled), so sample sizes are not given here. About invitations, not cancellations.
4. Majumder, R., Aslan, E., and Martuza, J. (2026). People Underestimate the Acceptability of Canceling Plans with Others. PsyArXiv preprint, version 2, posted 12 June 2026. A working paper posted before peer review; scenario-based online studies with American and French participants (N = 6,824); version 1 (December 2025) counted 11 studies, version 2 counts ten. Preliminary.
5. Kumar, A., and Epley, N. (2021). It’s surprisingly nice to hear you: Misunderstanding the impact of communication media can lead to suboptimal choices of how to connect with others. Journal of Experimental Psychology: General, 150(3), 595 to 607. A field experiment in which 200 University of Chicago students were asked to reconnect with an old friend by phone or email (103 completed the follow-up) and laboratory experiments in which people chatted with a stranger by video, voice or text. About reconnecting and getting acquainted, not cancelling.
6. Howard, P. L., and Sedgewick, F. (2021). “Anything but the phone!”: Communication mode preferences in the autism community. Autism, 25(8), 2265 to 2278. 245 autistic adults ranked six communication modes across seven scenarios; self-reported preferences.
7. Givi, J., Kirk, C. P., Grossman, D. M., and Sedikides, C. (2025). Maybe don’t say “maybe”: How and why invitees fail to realize that they should not respond to invitations with a “maybe”. Journal of Experimental Social Psychology, 121, 104814. Six experiments, five preregistered, each with about 200 to 410 US online (Prolific) participants. About replies to invitations, not cancellations.
8. Schweitzer, M. E., Hershey, J. C., and Bradlow, E. T. (2006). Promises and lies: Restoring violated trust. Organizational Behavior and Human Decision Processes, 101(1), 1 to 19. Laboratory trust game with 262 participants, almost all aged 19 to 22, playing against a scripted partner with money at stake; not friends. Supported by the Zicklin Center for Business Ethics Research.
9. Levine, E. E., and Schweitzer, M. E. (2015). Prosocial lies: When deception breeds trust. Organizational Behavior and Human Decision Processes, 126, 88 to 106. Four studies using economic games with online workers and participants from a US city; the last analysed 457 of 500 online workers. The lies helped the other player; an excuse that spares the sender is a different case. Supported by the Wharton Behavioral Laboratory and the Wharton Risk Center.
10. Lewicki, R. J., Polin, B., and Lount, R. B. Jr. (2016). An Exploration of the Structure of Effective Apologies. Negotiation and Conflict Management Research, 9(2), 177 to 196. Two scenario experiments, 333 US online workers and 422 undergraduate business students, rating apology parts for a job candidate’s wrongly filed client tax return, as descriptions in Study 1 and as statements in Study 2; words only, no tone.
11. Hohenstein, J., Kizilcec, R. F., DiFranzo, D., Aghajari, Z., Mieczkowski, H., Levy, K., Naaman, M., Hancock, J. T., and Jung, M. F. (2023). Artificial intelligence in communication impacts language and social relationships. Scientific Reports, 13, 5487. Two randomised experiments with Mechanical Turk workers chatting in pairs (219 and 291 pairs); a publisher correction added the omitted funding statement only.
12. Subtext. Terms and privacy. Privacy policy and terms last updated 22 October 2026. First party.