Am I overreacting? A method for checking your own read

“Am I overreacting” asks about the size of your response, but almost nobody who types it is worried about size — they are worried about accuracy, and those need to be answered separately.

Take the four combinations seriously, because three of them get ignored. You can be right about the fact and proportionate in your response, which is the case everyone imagines they are in. You can be right about the fact and larger in your response than the fact warrants: he did leave you on read for two days, and the two days do not support the conclusion you reached about the relationship. You can be wrong about the fact and calm about it, which is the quiet, common case that never generates a search query. And you can be wrong about the fact and enormous about it, which is what people fear when they ask the question and is rarer than the fear suggests.

Only the first half of each pair is checkable by anything outside your own head. Whether a message was read at 11:04 is a fact. Whether being upset for three days about it is too much is a judgement that depends on what the relationship is, what happened last month, and what you were promised. So this page checks accuracy, and leaves proportionality to you with one observation attached: a response sized to a conclusion you have not yet verified is always going to feel like an overreaction later, and the fix for that is not to feel less, it is to verify sooner.

You separate the two by writing them as two sentences, where the first contains only things a stranger with access to your thread could confirm and the second contains everything else.

Here is the split done properly. Observation: on Thursday he said he would call after work; no call came; the next message was at 11:40pm and said “sorry mad day”. Interpretation: he is losing interest and covering it with vagueness. The observation is three facts with times. The interpretation is a theory that explains all three facts, and so do at least four others — a genuinely bad day, an avoided conversation about something unrelated to you, a habit of over-promising calls he never makes, or a slipping memory about a plan he only half-registered. The theory does not become more likely because it explains the facts; every surviving theory explains the facts. That is what surviving means.

Do the split on paper, not in your head. This is not a mindfulness instruction; it has a mechanical reason. Held in memory, the observation and the interpretation fuse within about a day, and afterwards you remember having seen the interpretation. People reliably recall “he was being cagey on Thursday” when what happened on Thursday was a missed call and a late message. Once the fusion has happened, re-reading the thread feels like confirmation because the interpretation is now part of what you think you read. Writing the observation down at the time is the only cheap defence against that, and it costs about ninety seconds.

The split has a second use, later. When you eventually ask a question, the observation is what you quote and the interpretation is what you keep to yourself, for reasons that are tactical rather than polite: an observation can be answered with a fact, and an interpretation can only be argued with. That distinction is the whole subject of how to ask a question that actually gets answered.

The ordinary explanation has to cover every detail you noticed, and the honest test is to state it at its strongest and then find the one thing it cannot account for.

Most people do this backwards. They list the suspicious details and then check whether the innocent version can survive the pile, which guarantees the innocent version loses, because a pile of five ambiguous details always looks heavier than one explanation. Run it the other way. Say the ordinary version out loud in full — he had a bad week, he is bad at phones, he did not realise the call mattered — and then go through your observations one at a time, striking out every one that version already explains. Slow replies during work hours: struck. A missed call after a long day: struck. The eleven-day drought in initiating anything: not struck, because being tired explains being slow to reply and does not explain never starting.

What survives the strike-through is the case. Usually it is one item, sometimes none, occasionally two. One surviving item is a normal result and is not weak — it is the only thing you had that was ever going to carry weight, and it was previously buried under four items that proved nothing. None surviving is also a result, and it is the most common one. It does not mean you imagined the week. It means the week has an explanation you already believe is possible, and there is nothing left to investigate until something new happens.

Behaviour only counts as evidence when it departs from this specific person’s own baseline, because almost everything people cite as a red flag is somebody’s ordinary Tuesday.

Scroll back six weeks in the thread and read a stretch from before you started worrying. You are not looking for content; you are measuring habits. How long do their replies normally take in the evening? Do they usually initiate, or have you always opened the conversation and only noticed this month? Are the short answers new, or is this how they have texted since the start and the warmth you remember came from being together in person? Baseline reading is uncomfortable in a specific way, because it frequently shows that the thing you flagged has been happening constantly and cheerfully for a year.

The trap is the noticing threshold. Attention does not switch on evenly; it switches on after some trigger — a comment from a friend, a bad night, a photograph — and from that moment forward you are recording everything and comparing it against a period you were not recording at all. The comparison is rigged, and rigged in one direction. This is also why a fresh worry so often finds immediate supporting evidence within an hour of starting: not because the evidence appeared, but because the counting did.

Where the baseline check genuinely bites is on changes with a date. “He has replied within ten minutes for eight months and this week it is four hours” is a real observation with a boundary on it, and it is worth a question. “He does not text like a boyfriend should” has no boundary and is not an observation at all. The first can be confirmed or refuted by scrolling. The second can only be agreed with.

A worry that stays on one target until it is answered is behaving like a question; a worry that jumps to a new target within a day of each answer is behaving like a state, and the difference tells you which of the two you are dealing with.

The test is procedural, and deliberately so. It does not require you to diagnose yourself, decide whether you are an anxious person, or work out whether a feeling is intuition or fear — none of which can be settled from inside the feeling at 1am. It requires you to look at the last three things that bothered you and what happened after each was resolved. If the answer about Thursday landed and Thursday genuinely stopped mattering, the worry tracks facts. If the answer about Thursday landed and by Friday afternoon the concern had reattached to a photograph from March, the worry is not tracking facts, because facts do not behave that way.

The migrating version is not a verdict on you and it is not proof that nothing is wrong. Someone can be in a genuinely unreliable relationship and also be running a worry that moves — the two happen together often, since unreliability is what teaches the pattern. What migration tells you is narrower and more useful: your sense of alarm has stopped being a measuring instrument for this person’s behaviour, so anything you conclude for as long as it is moving has to come from the thread rather than from how strongly you feel. The full version of that distinction, including what it does and does not have to do with attachment styles, is in anxious attachment vs actual evidence.

No — going through someone’s phone damages the relationship whether or not it finds anything, and it does not resolve the uncertainty in either direction.

Start with why it fails on its own terms, before any question of ethics. A phone shows you what is currently on a phone. It does not show you what was deleted an hour ago, what was said on a call, what lives in an app you did not think to open, or what happens next week. So a clean phone buys a few days of relief and then stops working, because the mind that produced the original doubt can produce the same doubt about the parts of the phone you did not see. People who have done this describe the same arc almost without variation: relief, then the return of the doubt in a slightly different shape, then a second look. It is a test with no negative result, which is another way of saying it is not a test.

The failure when it does find something is worse. Say you find an ambiguous message. You now hold a fragment with no context, no timestamps you can place, and no way to ask about it without disclosing how you got it — which converts the conversation from “what happened on Thursday” into “you went through my phone”, a subject on which you are the one with something to answer. Whatever you found is now permanently entangled with how you found it, and it will be used that way in every argument you have about it, fairly or not.

There is also a straightforward practical cost that gets left out of the moral version of this advice: it spends the plain question. Before you looked, you could ask about Thursday and watch what an unprepared answer looked like — which is genuinely informative, and free. After you looked, you already know the answer, so the question is theatre and you learn nothing from it. The tool below is built for the situation before that point, and this page does not have a version for after it.

Being told you are overreacting is a claim about you, and the way to evaluate it is to check whether it arrived with an answer to your question or instead of one.

The phrase can be entirely fair. People do sometimes bring a small thing at high volume on a bad night, and a partner saying “I think this is bigger in your head than it was in reality” may be both right and kind. The tell is what follows the sentence. If it is followed by an account of what actually happened — where he was, why the call did not come, what the message meant — then the label is packaging and the account is the substance, and you should evaluate the account and ignore the packaging entirely. Fair criticism of your reaction and a real answer travel together comfortably.

The other version is where the label is the whole reply. The question was about a specific evening; the response is about your character, your history, your ex, or your tendency to do this. Notice the structure without leaping to a conclusion about motive: what has happened is that the subject changed, and the original question is still sitting there unanswered. That is worth naming out loud, once, in the least loaded language available — “maybe I am, but I still do not know what happened on Thursday” — which returns the conversation to something answerable and does not require you to litigate whether your reaction was proportionate.

Two guardrails. A single deflection is weak evidence; people deflect when tired, cornered or embarrassed about something unrelated, and one dodged question is not a pattern. And a pattern where every concern you raise is met with a comment about your reaction and never with an answer is a finding in itself, independent of whether any particular suspicion was correct — but it is a finding about how the relationship handles questions, which is a different thing from the thing you were originally worried about, and worth keeping separate in your own head.

You decide in advance what each possible answer would mean, then ask once, and then hold yourself to what you decided — because an answer you would not have accepted before you heard it is not an answer you are going to accept afterwards.

Write the two outcomes down before the conversation. If he says he was at his brother’s and the timings fit, this is closed and I stop re-reading the thread. If the answer moves the times around or does not touch the detail I asked about, I have a second question and I ask it calmly next week rather than tonight. Pre-committing is not a trick to make yourself feel better; it is what stops a resolved question from being reopened at 2am by a mood, which is the failure mode that turns one Thursday into a three-week project.

Then accept a clean result properly. This is the hardest instruction on the page and the one that most determines whether any of this was worth doing, because a method you only believe when it agrees with you is not a method. If the answer covers the leftover detail, the case is closed until something new happens — not permanently, not naively, but closed with the option of reopening on new evidence rather than on a feeling at midnight.

And if the answer does not cover it, you are holding one specific, dated, unexplained thing. That is a small file and a real one, and it is worth more than the fifty impressions you had before you started. What you do with it is not a question this page can answer, because it depends on things no thread contains. But you will be deciding with the thread as a witness rather than with a memory that has been re-read forty times — which was the entire point of the exercise. So run the check below, and notice which of the two answers you were hoping it would give you.

• How to test a suspicion instead of feeding it — The difference between checking a suspicion and feeding it: write the claim so it could be false, decide what would disconfirm it before you look, run one of the four tests actually available, and set the stopping rule in advance. Does not assume the suspicion is wrong.
• Anxious attachment vs actual evidence — Why the “calm intuition vs loud anxiety” test is unusable by the person who needs it, and what replaces it: a procedural check on whether a worry stays on one specific target until answered or relocates each time the last one clears. Diagnoses nobody, and does not dismiss real evidence as anxiety.
• How to ask a question that actually gets answered — How to build a question with a factual answer: quote the observation instead of naming the conclusion, ask one thing rather than three, choose text or in person for the right reason, decide beforehand what each answer would mean — and then accept a good one.

How do you know if you are overreacting or if something is actually wrong?

You separate the observation from the interpretation and check them separately, because they fail in different ways. Write down what actually happened as a sentence someone else could verify — the message, the time, the gap — and write what you concluded as a second, separate sentence. Most of the time the observation survives and the interpretation is one of several that fit, which is not the same as being wrong and not the same as being right; it means the question is still open and there is a specific thing to ask about.

Is a gut feeling enough to act on in a relationship?

A gut feeling is enough to look with, and not enough to conclude with. A feeling is a summary your brain produced from things it noticed and did not label, so it often points at something real, but it arrives without the underlying detail and cannot tell you whether the detail was a changed routine, a bad mood, or nothing at all. The usable move is to convert the feeling into one specific claim that could turn out false, then check that claim — a feeling that cannot be turned into a checkable sentence is not yet ready to be acted on.

Can a text thread prove someone is lying?

No, a text thread cannot prove someone is lying, and any method claiming otherwise is overselling. Not a lie detector: inconsistency is not proof of deception, because people misremember times, describe evenings loosely, and hide small embarrassments that have nothing to do with the thing you are worried about. What a thread can show is whether a specific claim fits the timestamps around it, and a claim that does not fit is a reason to ask one direct question rather than a conclusion about anyone's character.

My partner says I am overreacting — how do I know if that is true?

Being told you are overreacting is a claim about you, not evidence about the thing you raised, and it can be tested by whether it arrives with an answer or instead of one. If they say you are overreacting and then explain what actually happened, the explanation is the part to evaluate and the label is just packaging. If the label arrives on its own, and the original question is still unanswered afterwards, then nothing has been resolved regardless of whether your reaction was proportionate.

Should I check my partner's phone to find out the truth?

No — going through someone's phone damages the relationship whether or not it finds anything, and it does not resolve the uncertainty either way. A phone with nothing on it proves nothing about a conversation that was deleted or held elsewhere, so the reassurance evaporates within days, and a phone with something ambiguous on it produces a fight about how you know rather than about what you found. It also spends the one move that was still available, which is asking a plain question and seeing what the answer looks like.

How many times should you check something before you let it go?

Decide the number before you start checking, because the number chosen afterwards is always one more. A useful rule is that a check you would not accept a clean result from is not a check, so if three clean results in a row would not settle it, running a fourth is not going to either. When checks keep coming back clean and the worry does not move, the thing that needs attention has stopped being the other person's behaviour and has become the process you are running on it.

What is the difference between a concern and an accusation?

A concern names something you observed and asks what it was; an accusation names what you have decided it means and asks the other person to agree. The practical difference shows up in what a reply can do: a concern quoting a time and a message can be answered with a fact that either fits or does not, while an accusation can only be admitted or denied, and both of those produce another sentence to interpret rather than a fact to check. Phrasing matters here for mechanical reasons, not for politeness — one version can be settled and the other cannot.