How do you say no to a friend without over-explaining?

A friend has invited you to something on Saturday and you do not want to go. The reply you have typed runs to six lines. The no, such as it is, sits in the fifth, behind your week at work, a half promise to come for an hour and two sorrys. You delete it and start again, and the second version is longer.

I co-founded Subtext, an app that checks how a draft may read before you send it, so I would like you to try it. Further down, Subtext reads a reply to a friend’s birthday invitation that hedges about Saturday, and I show the three versions it offered and what it wrote once the sender said the answer was no.

I think a no does not need a defence. It needs an answer your friend can find in the first line. Two sets of experiments support part of that. People who imagined declining a friend’s invitation expected the friend to be angrier and more disappointed than the friend reported¹. People unsure whether they could go, or whether they wanted to, overestimated how likely the person inviting them was to prefer a maybe to a clear no². Both were about invitations, and nearly all of it was imagined.

Rewriting a no five times says little about you, and stop overthinking what to text covers the rewrite loop. If worry about disappointing people is constant, it is worth raising with a professional, and texting anxiety covers the texting side.

If you only want wording to copy, go to the example texts. If the person you are writing to has threatened or frightened you, go straight to the paragraph on threats and pressure.

Do friends take a no as badly as we expect?

In the experiments that tested it, people who said no expected a worse reaction than the people on the receiving end reported, though most of the scenarios were imagined. Givi and Kirk ran five studies with more than 2,000 online participants¹. In four of them, people imagined a friend’s invitation and either predicted the friend’s feelings after a decline or rated their own after being turned down. The decliners expected more anger and disappointment, a stronger sense that they did not care, and a lower chance of being asked again.

The other study used 80 real couples after exclusions, one partner writing a decline they had been told to write, and the decliners again expected more anger than their partners reported. They were romantic partners, mostly together for over five years. I could not read the full Givi and Kirk paper, so its sample sizes and design details come from the publisher’s release and the authors’ open data¹.

Every decline gave a short, ordinary reason, and in the imagined scenarios the friend could still go with other people¹. So the studies show how a short no was predicted and rated, and say nothing about a long one.

The authors’ account is that decliners assume the person who invited them will dwell on the refusal, while that person is more likely to think about the considerations that led to it¹³. Answering as the inviter first reduced the overestimate in one study¹³. I suspect that is also why we over-explain, a long reason being padding for a blow we expect, though none of these studies measured explanation. The authors add that declining every invitation would probably end the invitations³.

Requests show a similar gap. In Lu, Fang and Qiu’s seven studies in China (2,132 people), people who refused predicted worse consequences, such as anger and unwillingness to help in return, than the people they refused reported⁴. In one, 61 pairs of friends rated a real, recalled refusal, and the gap held on two of three measures, the exception being change in the relationship. The other studies were mostly imagined scenarios, and not all involved friends.

Most requests were small, and I found no study of a refusal that costs the friend something real, such as a friend left without a lift. On average the forecast was gloomier than the reaction, and that tells you nothing certain about your friend.

Is a “maybe” kinder than a no?

When you are unsure whether you can go, a maybe feels like the polite answer, and the experiments suggest it is less welcome than the person sending it expects. Givi, Kirk, Grossman and Sedikides ran six experiments, five of them preregistered, in which US online adults imagined a friend’s invitation to a concert or a restaurant’s soft launch². In the first (404 people), 65 per cent of invitees said the friend would prefer a maybe, against 34 per cent of inviters who said they would. Inviters more often wanted a clear no, partly because a maybe left them feeling more disrespected than the invitees expected. In the last (300 people), 80 per cent of invitees said they would reply maybe, while 48 per cent of inviters said they would prefer one. So a third to a half of inviters preferred a maybe, and the finding is the gap between that and what invitees expected.

All the participants were US online adults, the scenarios were imagined, and the invitations were to social events. None involved favours or loans. The authors suggest that an invitee could still say maybe while explicitly giving the inviter permission to treat it as a no if that better suits the situation². Real uncertainty is a different case from a hedged no.

It matters because an over-explained no can be built from maybes: “I’ll see how I feel”, “I might be able to”, “if nobody else can”. If you have decided, those phrases are a no that has not been said yet. Cut them. If you have not decided, say when you will.

How much reason does a no need?

No study I found varied how many reasons a refusal to a friend gives, or tested whether a long explanation makes a friend argue. The evidence is from nearby settings, and it points less clearly than the advice built on it.

The study people cite is about a photocopier. Langer, Blank and Chanowitz asked 120 adults at a university library whether they could use the machine first, and an empty reason (“because I have to make copies”) helped for a small favour and did nothing for a large one⁵. It tested whether a reason made strangers more likely to agree to a request, so no one was giving a reason for a refusal. The piece on saying no to extra work tells it in full. A 1985 partial replication did not reproduce it in either follow-up copier study and found that a reason the person could control (“because I don’t want to wait”) drew less compliance than one they could not (“because I feel really sick”)⁶.

Donnelly and colleagues’ six studies of how people judge excuses for declining invitations found that saying you lacked the time left people feeling less close than saying you lacked the money, a gap the authors trace to time seeming more under your control⁷. They tested how excuses are perceived, say nothing about how much detail a reason needs, and do not license inventing one.

My view is to say the no first, give one true piece of context if you want to, and stop. A long explanation can bury the answer, hand your friend something to solve (“you can stay at mine”), give away more of your money, health or week than you meant to, and carry a maybe in with it. A longer message can be right when the situation is complicated or your friend has already paid a cost.

A no in five steps

My working framework has five steps. It is my own judgement, and nobody has tested it as a package.

1. Decide what you are declining. This occasion, this kind of request, or an ongoing role. “I can’t come on Saturday” and “I can’t be your regular lift” are different decisions.
2. Say the answer in plain words. “I won’t be there”, “I’m not lending money.” A Western Australian self-help module advises a brief, polite no that owns the decision (“I don’t want to” in place of “I can’t”) without elaborate reasons⁸. That is expert guidance without a trial behind it, and “I can’t afford it” is fine where it is true.
3. Add one true piece of context, only if you want to. “I can’t afford the trip” is a true limit and needs no balance sheet. The nearest evidence on invented excuses is economic games between strangers, where a person who lied to help someone else was trusted more for goodwill than a person who told a truth that hurt someone, but people were less willing to rely on the liar’s word⁹. That does not make an invented excuse safe. Cancelling plans goes into it.
4. End on something human. Thanks, a good wish, or an alternative you would really carry out. “We should do something soon” can sound like a promise.
5. Check what is left. No accidental maybe, no invented excuse, no offer you do not intend to keep.

Then check the draft against eight questions.

• Do they have to read a story to reach the answer?
• Is there a detail that does not help them understand or plan?
• Am I giving personal information I would rather keep private?
• Did I write “maybe”, “I’ll try” or “we’ll see” after deciding no?
• Am I asking them to reassure me that I am still a good friend?
• Have I offered a favour or a date I do not mean?
• Is any part of the reason not true?
• Does a recurring expectation need a different conversation?

Each yes is a place to look again, and several suggest the no is buried.

What does a short no look like?

These are made up for the post, and a different writer would word each differently. Keep any thanks, sympathy or good wish an after adds only if you mean it.

A party. A friend is having a housewarming on Saturday, and you have decided not to go.

Before “I’m so sorry, I’d love to see the flat but Saturday is looking tricky, my sister might be in town, though that isn’t certain, so I’ll see how the day goes and let you know?”

After “I won’t be there on Saturday, but thank you for inviting me. I hope it goes well.”

The before hides the no inside two maybes, “my sister might be in town” and “I’ll see how the day goes”. The after keeps the one fact that matters.

A favour. A friend asks you to look after their cat for two weeks, and you cannot.

Before “I’d love to, I adore her, it’s just that I’m not sure about those weeks, my parents might visit and work might send me away, and I might get it wrong, but I could maybe do a few days if nobody else can?”

After “I can’t look after her for those two weeks. I adore her, so I hope you find someone good.”

“Might” appears three times, and “if nobody else can” makes you the fallback. The offer of a few days lived inside a maybe, so it goes. If you would really do three days, say that plainly.

A trip. A group of friends is booking a weekend away that you cannot afford.

Before “This sounds amazing and I’m gutted, but with the flights and the hotel and the car insurance coming up I’d be stretching myself, although prices might drop, so can I let you know in a few weeks?”

After “I can’t afford the trip, so I won’t be coming. Thank you for including me, and have a great time.”

The before lists costs nobody asked about and keeps a yes alive with “prices might drop”. The after gives the true limit and no figures.

What about money, and a request that keeps coming back?

Two examples follow, a loan and a lift.

A loan. A friend has asked to borrow money for the second time, and you are not going to lend it.

Before “You know I’d help if I could, and it’s not that I don’t trust you, it’s just that things are tight for me too, though I might be able to move some things around next month, so can I think about it?”

After, warm “That sounds hard. I won’t be lending you money, and I know that’s not what you hoped to hear.”

After, brief “I won’t be lending money.”

After, firm “I’m not lending money, and that won’t change. Please don’t ask again.”

All three state the decision. The before offers a maybe and asks for time, which leaves the question open. In an interview study of 57 clients of a California nonprofit, some people described turning down friends’ and relatives’ loan requests indirectly, for example by saying they had no money, which the authors describe without recommending it¹⁰. I would use warm for a close friend who is struggling, brief for a friend who wants an answer and nothing more, and firm for a third ask or a friend who argues. The firm one also promises it will not change and asks them not to ask again, so use it only if you mean both.

A lift. A friend who has shared your lift to work for a month keeps asking for more, and you want to stop.

Before “I can do tomorrow, and probably Thursday, but next week’s up in the air and the mornings have been hard, so let me know and I’ll try, I’m happy to do the odd one.”

After “I’m going to stop driving you to work. I’m happy to do the odd one when you’re stuck, but you’ll need another way in most mornings.”

The question is what you are declining, one lift or being the lift, so the after names what changes. “The odd one” is there only because the sender offered it in the before. Add a start date only if it is real.

What if they push back, go quiet or keep asking?

I found no study of how to answer a friend who pushes back or of what their silence means, so what follows is my own judgement. Set boundaries over text has the full pushback playbook, guilt trips included.

They answer your reason with a fix. Decide whether it changes your answer. “Share a room with me and it’s half the price” solves the cost and does nothing for not wanting to go. Say so: “That’s kind, and it doesn’t change my answer.” Offering a fix is a fair reply to the reason you gave.

They ask why. One true sentence can be enough, such as “It’s just not for me this time”, and a second reason can hand them a second thing to answer.

They ask again. Repeat the decision in nearly the same words and add no fresh reason: “I know it’s not what you wanted. It’s still a no from me.” The Western Australian module calls this the broken record⁸. If your first message had a maybe in it, the follow-up may be answering your maybe, so clear that up first.

It keeps coming. Name the pattern once, when neither of you is waiting for an answer: “I’m glad to help now and then, but I can’t be the person you rely on every week.” I would try a call for that conversation, if a call suits you both. In a field experiment, calling an old friend built stronger bonds than emailing, with no more awkwardness¹¹, but that was about catching up, so the call is my inference. Cancelling plans by text says more about voice.

They go quiet. Silence alone does not establish anger, rejection or consent, and the invitation studies did not measure it. Leave it, and send an ordinary message about something else later. Left on read covers silence.

If a friend threatens you, punishes you for refusing, monitors you or keeps pushing after you have said no, then wording is not the issue any more. If you are in immediate danger, call your local emergency number. The section on when not to text in Set boundaries 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. Talk to someone trained for situations like this, and keep your conversation with that friend out of Subtext and every other AI tool.

Is there an app that helps me write a no?

Yes, there are several kinds, and the cheapest need no app. A saved script you adjust each time, a pause so that deciding comes before wording, and saying the reply aloud to hear how long it runs are ordinary options. A long pause can leave the friend waiting.

Apps come in a few types. A CBT thought diary is built to help you examine a prediction such as “they will stop liking me”, and it does not write the message. Rewriters on phones and in chat apps restyle a draft you give them. General chatbots draft whatever you ask for, including alternatives. Tone checkers, which include Subtext, read your draft and say how it may land.

No study I found tested any of these tools, Subtext included, for refusing a friend. What they offer is speed and a second reading, and some add things, such as a reason, an apology or a date you never gave. If you use one, tell it to keep your decision, to give one short warm version and one firmer one, and not to add reasons, apologies, promises or another date.

What a full AI rewrite does to your voice, and whether to say you had help, is in does AI make your texts sound like a robot?.

Where Subtext fits, and what it cannot do

Subtext reads the no you wrote, marks words that could read as vague, over-apologetic or like a promise you did not mean to make, names the feeling under it in one word, and offers up to three versions, each marked safe to send or heads-up. You can also ask for a shorter, warmer or more confident take on any version.

The draft, made up for the run, answers a birthday invitation from a friend called Hannah and reads “Ahh Hannah I’d love to come!! Not sure I can make Saturday though, work has been mental and big groups are a lot at the moment.” It goes on “I’ll let you know closer to the time, maybe I can pop in for an hour? Sorry!!” Subtext returned the tags Excess exclamation and Clarity and named the feeling as Apologetic. The note on the first tag said the double exclamation marks “read as anxious”, a comment on the punctuation. The note on the second named “Not sure I can make Saturday” and “maybe I can pop in”, which “leave Hannah unsure whether you’re coming”.

All three versions were marked safe to send, and all three kept the maybe. The first (Close to yours) kept “maybe I can pop in for an hour?”, the second (Warmer, less sorry) said “I can’t promise Saturday yet”, and the third (Clear and calm) asked “Can I tell you nearer the day?”. The note under each version says something about how Hannah will take it: “Hannah should take it well”, “It should feel kind and low pressure” and “Hannah knows where she stands”. Subtext cannot know any of that, and the second tag’s note speaks for Hannah in the same way. Its closing line said it had “made it clearer that Saturday is a maybe.” The draft reads as a real maybe, and nothing in it told the app the sender had decided no.

So the sender told it, in one follow-up line: “I’ve decided not to go. Make it a clear no, please.” Subtext answered “Good on you for deciding” and offered one version, Clear no, marked safe to send: “Ahh Hannah, I’d love to come, but I’ve decided I can’t make Saturday. Work has been mental and big groups are a lot at the moment. Sorry! I hope it’s a brilliant night.” It dropped the hour and the “I’ll let you know”. It kept both reasons and the sorry, which this article would cut, added a good wish the draft did not have, and its note again said “Hannah knows where she stands”. So in this run the app wrote the no once it was told the answer, and the trimming was left to the sender. That is the app’s real output, unedited.

Subtext cannot decide whether you should say no, know how your friend feels or will react, or look after the friendship once you have sent it. 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. Neither is a prediction about your friend. It sends nothing, and a version can add a detail you did not give, so read each one before sending.

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

Questions people ask

How do I say no to a friend without over-explaining? Put the no in the first line, in the words you mean, add one true piece of context only if you want to, end with thanks or a good wish, and stop. That method is mine, built on two findings about invitations and on my own view of how a long no can go wrong.

Should I say sorry? One sorry is not a rule either way. In studies of social rejections, rejections that came with an apology were rated as more hurtful, and in a face-to-face task were followed by more aggressive behaviour¹². Those studies covered rejections in general, and the evidence on apologies points different ways by context, as how to apologise over text explains. I would use thanks and save sorry for something you did.

Do these rules work in every friendship and language? Linguists describe a refusal as parts, such as a direct no, regret, a reason, an alternative and thanks, and one study using written scenarios found that Japanese speakers writing in English carried habits from Japanese into the order, frequency and content of those parts¹³. It had 60 participants answering imagined scenarios, so ask how you two normally talk and do not guess from a nationality. Second language writers can start with do my English texts sound polite?.

What if I already said yes and now want out? That is a cancellation, a different problem from a decline, and how to cancel plans by text without sounding flaky has its own research.

What should I paste of a friend’s messages? Only what the reply needs. Crop names and numbers out, and leave out anything about their health, money or private life. OpenAI’s help page says that on personal ChatGPT accounts, sharing chats to improve its models is on by default and can be turned off¹⁴, and which AI writing assistants train on your messages compares thirteen tools. Nothing you send to Subtext is used to train an AI model, though the text still goes to the AI providers its privacy policy names, and its terms ask you to confirm you may submit other people’s words¹⁵. A conversation deletes itself five days after you last use it, or 90 days if you pin it, and notes you save to Memory stay until you delete them¹⁵.

The order to do it in

Decide what you are declining: this occasion, this kind of request, or a role. Write the no in the first line, in the words you mean. Add one true piece of context if you want to, and nothing you would not say to their face. End with thanks or a good wish, and with an alternative only if you would carry it out. Take out every maybe you do not mean and every offer you do not intend to keep.

Then read it once as the friend who asked. In one study, answering as the inviter first shrank the overestimate¹, though that was a rated scenario, and a real decline may differ.

If they push back, repeat the decision and add nothing new. If the same ask keeps coming, talk about the pattern when neither of you is waiting for an answer. If it stops being about wording, stop writing and get support.

In the experiments, the fear ran ahead of the reported reaction on average. They cannot tell you how your friend will take yours, and they cannot tell you what a long no costs. Give your friend the answer in the first line, keep the rest true, and stop there.

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. 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. Five studies with more than 2,000 online participants in total (the publisher’s release and the authors’ open data), mostly imagined invitations from a friend, one with real couples; every decline gave a short stock reason. The full text sits behind a paywall and was not read.
2. 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, with US online adults reading vignettes about a friend’s invitation; Experiment 1 had 408 recruited and 404 analysed. No specific funding and no competing interests declared.
3. Givi, J., and Kirk, C. P. (2024). Are Our Fears of Saying “No” Overblown? The Conversation, 21 February 2024. Written by the study’s authors. Cited for their account of why the gap arises, the effect of answering as the inviter first and their caveat that declining every invitation would probably end the invitations.
4. Lu, J., Fang, Q., and Qiu, T. (2023). Rejecters Overestimate the Negative Consequences They Will Face From Refusal. Journal of Experimental Psychology: Applied, 29(2), 280 to 291. Seven studies (N = 2,132), with Chinese online and university samples, and four supplementary studies (N = 1,470); Study 1 had 209 adults imagining refusing a friend’s request for help designing a logo, and Study 2 recalled real refusals between 61 pairs of friends. Funded by the National Natural Science Foundation of China.
5. Langer, E., Blank, A., and Chanowitz, B. (1978). The Mindlessness of Ostensibly Thoughtful Action: The Role of “Placebic” Information in Interpersonal Interaction. Journal of Personality and Social Psychology, 36(6), 635 to 642. Three field experiments; the copier experiment had 120 adults at the Graduate Center of the City University of New York, with cells of 15 to 25. It tested whether a reason in a request changed agreement, not how a refusal is worded.
6. Folkes, V. S. (1985). Mindlessness or Mindfulness: A Partial Replication and Extension of Langer, Blank, and Chanowitz. Journal of Personality and Social Psychology, 48(3), 600 to 604. Four studies: three at copying machines (82, 42 and 96 people) and one with 69 undergraduates; Studies 2 and 3 did not reproduce Langer and colleagues’ results.
7. Donnelly, G. E., Wilson, A. V., Whillans, A. V., and Norton, M. I. (2021). Communicating Resource Scarcity and Interpersonal Connection. Journal of Consumer Psychology, 31(4), 726 to 745. Six studies of excuses for declining invitations, including perceptions of couples planning a wedding; sample sizes in the published text could not be checked.
8. Centre for Clinical Interventions (2008). Assert Yourself, Module 6: How to Say “No” Assertively. Department of Health, Government of Western Australia (contributors F. Michel and A. Fursland). A cognitive behavioural self-help handout, not a trial; the module is a PDF listed on the Centre’s assertiveness resources page.
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 (the authors’ count) using trust games with strangers, online and in a university laboratory, with no social context; funded by the Wharton Behavioral Laboratory and a Wharton Risk Center doctoral fellowship.
10. Wherry, F. F., Seefeldt, K. S., and Alvarez, A. S. (2019). To Lend or Not to Lend to Friends and Kin: Awkwardness, Obfuscation, and Negative Reciprocity. Social Forces, 98(2), 753 to 775. 57 interviews in 2015 with clients of a California nonprofit’s lending circles; qualitative, and the authors say it cannot be generalised to a population.
11. 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 refusing.
12. Freedman, G., Burgoon, E. M., Ferrell, J. D., Pennebaker, J. W., and Beer, J. S. (2017). When Saying Sorry May Not Help: The Impact of Apologies on Social Rejections. Frontiers in Psychology, 8, 1375. Three sets of studies (N = 1,096, 355 and 426) of social rejections; combined N approximately 1,880; findings specific to rejections.
13. Beebe, L. M., Takahashi, T., and Uliss-Weltz, R. (1990). Pragmatic Transfer in ESL Refusals. In R. C. Scarcella, E. S. Andersen, and S. D. Krashen (eds.), Developing Communicative Competence in a Second Language (pp. 55 to 73). Newbury House. Written discourse completion task with 60 participants (20 Japanese speakers answering in Japanese, 20 in English, 20 Americans in English); design as described in the CARLA speech act bibliography, as the chapter itself was not read. That page has a certificate fault in some browsers, and an archived copy is available.
14. OpenAI. Data controls in ChatGPT. Help Center page, checked 6 November 2026; the page shows that it was updated the same day.
15. Subtext. Terms and privacy. Privacy policy and terms last updated 22 October 2026. First party.