Gaslighting phrases in text messages
The phrase is never the evidence. The effect on you is.
Last updated
Language forensics
No phrase, on its own, tells you that you are being gaslighted — and treating a phrase list as a diagnosis is how people end up more confused rather than less. Not a lie detector. Inconsistency is not proof of deception. The word names a sustained effect, not a vocabulary: the thing it describes is what happens over months when your account of your own experience is repeatedly overridden until you stop trusting it. That is not something a screenshot can establish, and it is not a puzzle to be solved by analysing text. If you have found yourself doubting your own memory of things you were present for, that is the part worth taking seriously, and the right next step is a conversation with a person you trust — not a better analysis of the messages. There are support numbers a few paragraphs down, and they are not there as a footnote.
What are the phrases people mean by “gaslighting in text messages”?
The phrases most commonly named are “that never happened”, “you’re remembering it wrong”, “you’re overreacting”, “you’re too sensitive”, “it was just a joke”, “you’re making things up”, “I’m sorry you feel that way”, and “you made me do this” — and every one of them is also something people say honestly in ordinary arguments.
Read them with the innocent version attached, because the innocent version is extremely common and the pages listing these phrases almost never supply it. “That never happened” is what a person says when they genuinely do not remember an exchange, which happens constantly and in both directions. “You’re overreacting” is a claim about proportion that is sometimes correct. “It was just a joke” is what someone says when a joke landed badly and they are embarrassed. “I’m sorry you feel that way” is a bad apology, and bad apologies are mostly produced by people who are defensive and clumsy rather than by people running a campaign. Any list that presents these as manipulation markers has quietly assumed the conclusion, and it has done so for a reader who arrived already unsettled.
What changes their meaning is not the wording but three things around it: how often, in response to what, and with what effect on you. A phrase said once in a bad week is a bad week. The same phrase arriving every single time you raise anything, never accompanied by engagement with what you actually said, over a period long enough that you have stopped raising things — that is a pattern, and the pattern is the thing the word was invented for. Notice that none of those three factors is visible in a screenshot. All of them require the history you already have and nobody else does.
Why the phrase lists are built the way they are
The pages ranking for this query are largely produced by companies selling gaslighting-detection software. The one currently at the top advertises a specific accuracy rate for detecting gaslighting behaviour by software, offers no innocent reading for any phrase it lists, and points readers to a subscription rather than to a helpline. We are not repeating that figure here and we will not produce one of our own, because there is no defensible way to compute an accuracy rate for detecting a months-long relational pattern from message text — and a number that sounds precise is worse than no number at all when someone is deciding what to believe about their own relationship.
What does gaslighting actually mean, as distinct from an argument?
Gaslighting names a pattern in which one person’s version of events is repeatedly overridden until the other stops trusting their own memory and judgement — the defining feature is the erosion, not any individual denial.
The distinction from an ordinary argument is worth being precise about, because the word has become loose enough to cover any disagreement about facts, and that looseness is expensive in both directions. Two people remembering an evening differently is the normal condition of human memory; memory is reconstructive and confident, and both parties routinely feel certain. An argument in which one of you turns out to be wrong is a Tuesday. What is being described by the word is different in structure: it is not that you disagree about one evening, it is that disagreement always resolves the same way, and that the resolution is not “we settled it” but “I gave up and assumed I was wrong again”.
So the question to ask yourself is not “was that message gaslighting”. It is: over the last few months, has my confidence in my own account of things gone down? Do I check my memory against the thread before I trust it? Do I write things down now because I have learned I will be told I am wrong? Have I stopped bringing things up because of how it goes? Those are questions about you, they do not require you to diagnose anyone, and they are answerable without a single screenshot.
What should you do if you have started doubting your own memory?
Talk to a person you trust — a friend, a relative, a doctor, a counsellor — and describe out loud what has been happening. Persistent doubt about your own memory of events you were present for is the finding that matters, and it calls for another human being rather than a better text analysis.
This is the part of the page that is not about method, so it is stated plainly. If things at home have you frightened, walking on eggshells, or unable to trust your own account of ordinary events, there are people whose entire job is to talk that through with you, free, confidentially, and without any requirement that you name what is happening or do anything about it afterwards.
- United States — the National Domestic Violence Hotline: 1-800-799-7233, text START to 88788, or chat at thehotline.org. Free, confidential, 24 hours.
- United Kingdom — the National Domestic Abuse Helpline, run by Refuge: 0808 2000 247, free and 24 hours, or nationaldahelpline.org.uk for live chat.
- Anywhere — a GP, a therapist, a trusted friend, or a family member. Being talked through it by someone who knows you is worth more than any tool on this site, and the point of saying it out loud is partly to hear how it sounds outside your own head.
Calling one of those numbers is not a claim that your situation is severe, and it does not commit you to anything. People use them to think out loud about relationships they are unsure about. If you are in immediate danger, contact your local emergency number instead — 999 in the UK, 911 in the US.
Why a trusted person beats another read of the thread
Someone who knows you can tell you whether you have changed, which is the one observation you cannot make about yourself and the one that matters most here. They saw what you were like a year ago. They can say “you never used to check with me before you trusted yourself on something like this”, and that sentence carries more information than any amount of re-reading, because the thing being measured is the erosion and you are standing inside it. A thread can only ever show you the messages. A friend can show you the difference.
Is it gaslighting if they say you are overreacting?
“You’re overreacting” is not gaslighting by itself. It is a claim about proportion, and people make that claim honestly in both directions — including, sometimes, correctly.
Keeping that honest is not a concession, it is the thing that makes the rest of this page usable. If every dismissal counts as manipulation, then you have no way to distinguish a partner who is systematically undermining you from a partner who thinks you were unfair about the dishes and said so badly. Both feel identical in the moment, because both involve someone telling you your reaction was wrong, and the feeling is not the evidence. Plenty of people have been told they were overreacting on an occasion when they were, and have been glad afterwards to have heard it.
What moves it from disagreement to pattern is substitution. Watch whether the proportion claim arrives instead of any engagement with what you actually said. “That’s a bigger deal to you than it is to me, but you’re right that I said I’d call” engages and disagrees. “You’re overreacting” full stop, every time, on every subject, engages with nothing — and the tell is not the phrase, it is that after twenty of those exchanges you cannot name a single thing that got resolved. If you scroll back through six months of raising things and cannot find one instance of the substance being addressed, you have learned something real, and you learned it by counting outcomes rather than by identifying phrases.
The same distinction runs through our page on the words people use when a story is thin: no phrase is a signal, and the only readings that survive are the ones you can check against something other than your interpretation.
What about “I never said that” when you have the messages?
“I never said that” when the message exists in your thread is the one case in this area with a checkable answer, and the check is worth doing carefully — because the most common explanation is genuine misremembering, and the second most common is that the message says something slightly different from what you remember it saying.
Before treating a denial as manipulation, read the original message again in full, with the messages either side of it. People hold on to a compressed version of what was said — the gist, plus the feeling — and the compressed version is often stronger and cleaner than the actual sentence, which turns out to have contained a “probably” or to have been about a different week. That is not a failure on your part; it is how memory works for everybody, and it is exactly why the messages are useful. Discovering that they said something more hedged than you remembered is a real result and it is better to find it yourself than in the middle of an argument.
Where the message genuinely does say what you remember, you are holding a factual disagreement with a documented answer, and the questions become practical rather than diagnostic: does this person accept correction when it is in front of them, or does the denial survive the evidence? Someone who reads it and says “God, I did, I’d completely forgotten” is a person with a normal memory. A denial that continues after the message has been read aloud is a different situation, and it is the one worth taking to someone you trust. The full method for that comparison is on “I never said that” — when the receipts disagree.
What can a record of your messages actually show — and what can it not?
A message record can show exactly what was written and when, and whether two accounts of the same event differ. It cannot show intent, it cannot show what was said out loud, and it cannot establish that a pattern of denial was deliberate rather than forgetful.
That limit is worth internalising before you start building a case, because the gap between “here is what was written” and “here is what they were doing to me” is the gap where people lose arguments they were right about. The record is strongest on the narrow, checkable things: this message exists, it was sent at this time, it says this, and the later account of it says something else. It is silent on everything that happened in the room, on tone, on what was meant, and on why. Most of what makes a coercive dynamic recognisable happens outside the text, which is one reason a text analyser is the wrong instrument for it.
Where a record earns its keep is for you, and for the people helping you. Writing things down close to when they happen — what was said, when, what you understood it to mean — gives you something that does not get rewritten by the next conversation, and that is valuable precisely because your confidence in your own memory is the thing under pressure. Keep it somewhere only you can reach. If you later talk to a helpline, a solicitor, a doctor or a support worker, a dated, contemporaneous set of notes is the single most useful thing you can bring, and it is far more useful than screenshots assembled afterwards from memory.
Producing the receipts is a decision, not a move
Do not assume that showing someone the proof is the natural next step. Evidence settles disagreements between people who are both trying to get the facts right. In a relationship where someone is controlling, being confronted with proof can escalate matters rather than end them, because the disagreement was never really about the facts — and the person holding the receipts is not usually the person best placed to judge how that will go, because they are inside it. If any part of you is worried about how the other person will react, that worry is itself information, and it is worth talking through with the helpline or the friend before you decide, not after.
Why should you not diagnose the other person?
Diagnosing the other person converts a solvable situation into an unwinnable argument about a label, and it moves the question away from the only thing you actually need to decide: whether you can live like this.
Consider what happens when the word gets deployed. You say the word; they reject the word; the next two hours are spent on definitions, on what the term technically means, on examples that do or do not qualify, and on whether you have been reading things online. Nothing about your actual experience is discussed. Meanwhile the thing you walked in with — that you are exhausted, that you have stopped raising things, that you do not trust your own memory any more — has not been said, and it is the only part of this that is not arguable, because it is a report about you.
There is a second reason, and it is about accuracy. Gaslighting as a deliberate campaign and forgetfulness combined with defensiveness look nearly identical from inside the relationship, and they produce almost the same thread. You are not in a position to distinguish them, and neither is any tool. What you are in a position to establish is the effect on you and the count of things that never got resolved, and both of those support a decision without requiring a verdict on anyone’s character or intent. “I do not want to keep living like this” needs no diagnosis to be valid.
The same restraint applies to the tool on this page. It compares two tellings and shows what changed between them. It does not know who wrote them, what happened before, or what anyone meant, and it will never tell you that someone is manipulating you — because a piece of software that offered that verdict would be lying to you about what it can see.
What if someone has accused you of gaslighting them?
Treat the accusation as information about how the conversation is landing for them, and set aside the question of whether the label is technically accurate — because arguing the definition guarantees the underlying complaint never gets addressed.
The complaint underneath is nearly always specific: they raised something, and what came back dismissed it instead of engaging with it. That is answerable. Ask which conversation they mean, go and read it, and respond to the substance of what they raised. If you genuinely remember the event differently, say so in exactly those terms — “I remember that differently, and I might be wrong” is a sentence that leaves both accounts standing. What does the damage is the upgrade from “I remember it differently” to “that never happened” to “you’re making things up”, because the last two are claims about their mind rather than about the event.
It is also worth checking honestly whether the pattern they are describing exists, independently of the word. Look at the last several times they raised something. Count how many ended with the substance addressed and how many ended with them apologising for having brought it up. That count is not flattering to look at and it is not conclusive, but it is a fact about the relationship rather than a debate about terminology, and it is available to you right now without anyone having to win.
When does this stop being a question about texts?
It stops being a question about texts the moment the thing you are trying to establish is whether you are safe, whether you are being controlled, or whether you can still trust your own judgement — because none of those is answerable from a message thread, and continuing to look for the answer there delays getting an answer at all.
The signs that you have crossed that line are recognisable from the inside. You are keeping evidence in case you need to prove something. You rehearse conversations before having them. You check the thread before you trust your memory of a conversation you were part of. You have stopped mentioning things that bother you because of what follows. Your friends have started saying things about how you seem. None of those is about anyone’s phrasing, and no amount of textual analysis touches any of them.
What helps at that point is talking to somebody: the helplines above, a GP, a therapist, or the friend you have been not quite telling. Bring your notes if you have them. If someone is monitoring your devices or your accounts, use a phone or computer that they cannot reach — a friend’s, a library’s, a workplace one — and ask the helpline about that specifically, because they deal with it constantly and can talk you through it properly. That is a conversation for a trained person, not for a scenario page.
If what you actually have is a factual dispute about something that was said, and the relationship is not the frightening kind, the tool below does the narrow job honestly: paste two tellings of the same account and it marks exactly what changed between them. When you want the whole thread read rather than two paragraphs compared, you can run the conversation through the analyser, which will show you what does not line up and will still not tell you what anyone intended.
Check the neighbouring tell the same way
- The words people use when a story is thin — the method this page uses, step by step.
- “I never said that” — when the receipts disagree — The message proves what was written. Not what was remembered.
- Overexplaining: why extra detail reads as a tell — It shows they expected not to be believed. Often because of you.
Free tool · runs in your browser
Not a lie detector. Inconsistency is not proof of deception. This counts what is in your thread — timestamps, gaps, wording. It cannot tell you whether someone is lying.
Nothing you paste leaves this browser. No account, no upload, no model call.
Paste both from the thread rather than from memory. A remembered first telling is a third version, and it will be the one that moved.
This counted the gaps. It cannot read the messages. Run the actual conversation →
Common questions
- What are the most common gaslighting phrases in text messages?
- The phrases most often named are "that never happened", "you're remembering it wrong", "you're overreacting", "you're too sensitive", "it was just a joke", "you're making things up", "I'm sorry you feel that way", and "you made me do this". Every one of them is also a thing people say in ordinary disagreements, honestly and sometimes correctly, which is why a phrase list cannot identify gaslighting. What identifies it is repetition over time, and its effect: whether you have started to doubt your own memory of events you were present for.
- Can a phrase tell you that you are being gaslighted?
- No single phrase can tell you that you are being gaslighted, because gaslighting names a sustained effect rather than a vocabulary. Someone saying "you're overreacting" once, in an argument, is someone disagreeing with you about proportion — and they may be right. The thing the word actually describes is a pattern in which your account of reality is repeatedly overridden until you stop trusting it, and that pattern is measured in months and in how you feel about your own memory, not in a screenshot.
- Can a text analyser detect gaslighting?
- No text analyser can detect gaslighting, including this one. Not a lie detector — inconsistency is not proof of deception, and software cannot see intent, history, or what happened in the room. Tools that report a gaslighting percentage from message text are reporting a number they have no way to justify, and treating that number as a finding about a real person in your life is a worse position than having no number at all. What software can honestly do is put two versions of an account side by side and show which details changed.
- What should you do if you keep doubting your own memory of what was said?
- Talk to someone you trust — a friend, a family member, a GP or doctor, a counsellor — and describe what has been happening, out loud, to a person rather than to a text analyser. Persistently doubting your own memory of events you were present for is the part that matters, and it is a reason to bring another human being into the situation rather than a puzzle to solve alone at one in the morning. In the US, the National Domestic Violence Hotline is on 1-800-799-7233, text START to 88788, or thehotline.org. In the UK, the National Domestic Abuse Helpline is on 0808 2000 247, free and 24 hours, at nationaldahelpline.org.uk.
- Is it gaslighting if they say you are overreacting?
- "You're overreacting" is not gaslighting by itself, and treating it as a diagnosis will cost you more than it gains. It is a claim about proportion, and people make that claim honestly in both directions — sometimes wrongly and dismissively, sometimes correctly about a reaction that was genuinely out of scale with what happened. It becomes part of a serious pattern when it is the standing response to every concern you raise, when it arrives instead of any engagement with the substance, and when its cumulative effect is that you have stopped raising things at all.
- Should you show someone the texts that prove they said something?
- Not automatically, and not as a strategy for winning. Presenting evidence works when the other person is capable of saying "you're right, I forgot" — and in a relationship where someone is being controlling or coercive, producing proof can escalate things rather than settle them, because the disagreement was never really about the facts. Keep the record for your own clarity and for anyone helping you. Decide separately, and with someone you trust, whether showing it to the other person is safe and likely to help.
- What if someone has accused you of gaslighting them?
- Take the accusation as information about how the other person is experiencing the conversation, and separate it from the question of whether the label is technically correct. Arguing about the definition guarantees the underlying complaint goes unaddressed, and the underlying complaint is usually specific and answerable: they raised something, and what came back dismissed it rather than engaged with it. Ask which conversation they mean, read it again, and respond to the substance — including saying plainly when you do remember it differently, without telling them their memory is broken.
A suspicion score, not a lie detector. It measures what does not line up — nothing more. Everything else is your call, which is the only place it can honestly sit.