How to test a suspicion instead of feeding it

A test has a result you would accept in either direction; a feed has only one direction available, which is why it never ends.

Sort your last three hours into the two columns and the difference stops being abstract. Re-reading the thread: no new information enters, so nothing can be eliminated — feed. Scrolling their tagged photos: cannot exonerate anyone, since the absence of an incriminating photo is not evidence of anything — feed. Describing the situation to a friend at length: the friend has only your framing and your selection of details, so their agreement is a reflection rather than a check — feed, and the most convincing one, because it comes back in someone else’s voice. Comparing this week’s reply times against the same thread in June: could genuinely come back showing no change — test. Asking a question with a factual answer: could come back with an answer that fits — test.

The reason the distinction matters is that feeds raise confidence without adding evidence, and confidence is what you will act on. After three hours of feeding you will be substantially more certain than you were and will hold exactly the same facts. That is not a character flaw, it is what repetition does to anyone; it is also why the person who has spent all night on it is often the least reliable narrator of their own case by morning. The whole discipline is to notice which column an activity belongs in before you spend an hour on it.

You write the suspicion as one sentence containing a specific behaviour and a time, and if the sentence cannot be false, you do not yet have a suspicion — you have a mood about a person.

Work through the rewrite in the order it usually goes. First draft: “something is off with him.” Nothing in that can be checked, so it can never be settled, which means it will still be here next month in the same words. Second draft: “he has been distant lately.” Better, but “distant” is a summary of your own experience rather than an observable event, and “lately” has no boundary. Third draft: “since the 14th he has not started a conversation once, and before that he started most of them.” That sentence can be wrong. You can scroll and find that he started two, or that he never started them and you had simply forgotten. It has a date, a countable behaviour, and a comparison. It is the first version that is worth an hour of anyone’s night.

Precision does not mean triviality. The claim is allowed to be serious — “on the 9th he told me he was at his brother’s, and the thread shows him replying to a group at 1am with a photo of a bar” is precise and serious at once. What precision removes is the escape hatch, in both directions: a vague suspicion can never be disproved, which feels like safety and is actually the trap, because it means the worry has no exit. Sharpening the claim gives it a way to end.

Keep one sentence, not five. A suspicion split across five claims is unfalsifiable again, because when one is refuted the others carry the feeling forward and nothing has been resolved. Pick the strongest — usually the one with the sharpest timestamp — and test that. If it survives, take the next. Sequential testing is slower and is the only version that can ever finish.

You write down what would count as disconfirming before you look because afterwards every result gets absorbed into the theory, including the ones that should have ended it.

The mechanism is visible in miniature every time. You suspect he is avoiding you; you look; he replied within four minutes. Without a pre-written standard, that becomes “he is being extra attentive because he feels guilty”, and the check that should have cleared him has just been converted into further evidence against him. The same result would have counted as reassurance if you had written “a reply inside ten minutes on a weekday evening means the avoidance theory is dead” before opening the app. Nothing about the fact changed. What changed is that after the fact there was room to reinterpret it, and there is always room.

So write the standard as a sentence with a number in it, and write it somewhere you will not edit — a note with a timestamp, a message to yourself. It should name the result, not the feeling: “if he initiated at least twice in the last fourteen days, the claim is wrong” rather than “if it looks normal”. Anything phrased as looking normal will not look normal, because you will be reading it in the state that produced the question.

Four tests are available: your own thread against itself, your existing knowledge of their routine, one direct question, and time — and the list is short on purpose, because everything past it stops being a test.

The thread against itself is the cheapest and the most underused. Every claim about timing can be laid against the timestamps you already have, and every claim about a change can be laid against the same thread three months earlier. This is the check that most often comes back clean, and it comes back clean in a way you can look at: the reply times in June were the same as the reply times this week, and the graph in your head was wrong. Run this one first, always, because it costs nothing, involves nobody else, and removes a surprising share of suspicions before they reach anyone’s mouth.

Existing knowledge of routine means the shifts, classes, gym nights and commutes you already know about, not a routine you set out to reconstruct. The distinction is sharp and worth holding: using what you were already told is checking; building a schedule of someone’s movements you were not told is something else, and it also degrades your judgement, because a person keeping a log will find patterns in random noise within a week.

The direct question is the strongest test available and the one most people skip, partly because it risks getting an answer. It is the only check that can produce genuinely new information rather than reprocessing what you already hold — and it is the only one whose result the other person gets to participate in. How to ask a question that actually gets answered covers the phrasing that makes an answer possible instead of a fight.

Time is the fourth, and it is a real test rather than a euphemism for doing nothing. Some claims can only be settled by waiting a defined period and counting: whether the initiating comes back, whether the plan actually happens, whether the pattern repeats next month. Set the window, write the count, and stop looking until it ends. An unbounded wait is not this test; it is the loop with a longer stride.

Yes — sustained checking changes the other person’s behaviour, and the changes it produces look exactly like the ones you are watching for.

Monitoring is rarely as invisible as the person doing it believes. It shows up in the questions that get asked, in the speed of the reply to “where are you”, in a tone that arrives half a second late. Most people register it within a couple of weeks even if they cannot name it, and they adapt: they start explaining their whereabouts before being asked, they mention who they were with unprompted, they answer faster than usual, they get careful about the wording of ordinary messages. Every item on that list is what a guilty person does. Every item is also what an innocent person under audit does. Once the watching has been running long enough to change the behaviour, those two readings are no longer distinguishable, which means heavy checking does not merely fail to settle the question — it destroys the instrument.

There is a second-order version that hits your own evidence. Vigilance changes what you record: a night spent refreshing a profile stores a much denser memory of that night’s activity than any ordinary night, so the comparison you make later — this week versus a month ago — is comparing a monitored week against an unmonitored one and will always find more in the monitored one. The asymmetry is invisible from inside and it reliably produces the feeling that things have got worse.

The practical consequence is the same in both cases: front-load your checking, do it once, and stop. A single careful pass through the thread with a written claim and a written disconfirming standard is worth more than three weeks of ambient watching, and it leaves both the relationship and the evidence intact enough to act on.

If a clean result does not move the worry, the problem has changed address — the question is no longer about the other person’s behaviour, and no further check on them can resolve it.

This is not a way of telling you that you are wrong. It is a statement about what instruments can do. You ran a check that could have gone either way, you agreed in advance what a clean result would mean, the result came back clean, and the worry continued unchanged. Whatever is producing the worry is therefore not being fed by the evidence, because the evidence just went the other way and nothing happened. Running the same check again will produce the same clean result and the same lack of relief, which is the loop, described precisely.

Set the number before you start rather than after. Two clean tests in a row on the same claim is a reasonable stopping rule and you should write it down while you are still calm enough to mean it, because the number chosen mid-loop is always one more than the number of checks you have run. If you pass the stopping rule and the worry is still here, the honest next move is a conversation about the pattern rather than a fourth examination of one evening: “I keep landing back here even when the answer is fine, and I do not want to be the person auditing you” is a true sentence and a very different conversation from an accusation.

Whether the worry moves after being answered is also the most useful signal you have about what kind of worry it is, and it is a procedural signal rather than a diagnosis of you. That distinction — a worry that stays on one target versus one that relocates each time the last is cleared — is worked through in anxious attachment vs actual evidence, which is careful to avoid telling anyone what they have.

You stop testing and say it out loud as soon as the cheap checks are exhausted, which is usually far earlier than it feels — most people run six weeks of silent checking on a question they could have asked in one sentence on day two.

The delay has a reason worth naming: an unasked question stays at whatever probability you find tolerable, while an asked one produces an answer you have to live with. Silent checking is comfortable precisely because it cannot conclude. But the cost accrues in two places at once. It accrues in you, in sleep and attention. And it accrues in the relationship, because six weeks of watching changes how you behave toward someone, and they will experience that long before they know why — so the conversation you eventually have starts from a worse place than the one you could have had at the beginning.

A workable trigger: once you have a written claim with a time in it, and the thread check has not cleared it, ask. Do not wait for a second instance, because waiting for a second instance is a decision to let it happen again. Do not wait until you are certain, because certainty is not available from this evidence and waiting for it means never asking. Ask while you still have one specific thing to point at, which is also when the question is easiest to answer.

Then run the check below before you do, honestly rather than protectively. It is entirely capable of coming back with “there is not much here” — that outcome is written into it deliberately, because a self-check that can only escalate is not a check. And then answer the question it does not ask: if the answer you get tomorrow is a good one, are you going to accept it?

• Am I overreacting? A method for checking your own read — the method this page uses, step by step.
• Anxious attachment vs actual evidence — Not a diagnosis. The usable distinction is what the worry attaches to, and whether it moves.
• How to ask a question that actually gets answered — A question that gets answered has a right answer they can give you in one message.

How do you stop being paranoid in a relationship?

You stop the loop by converting the suspicion into one specific claim that could turn out to be false, testing that claim once, and deciding in advance what result you would accept — because a worry with no possible negative result has nothing to stop it. Advice to journal, meditate or breathe through it assumes the suspicion is wrong before checking, which is why it fails for the people whose suspicion happens to be right. The check comes first; if it comes back clean and the worry continues anyway, that is when the worry itself becomes the thing to work on.

Why does re-reading old messages make you more suspicious?

Re-reading old messages adds no new information and raises felt certainty anyway, because each pass makes the words more familiar and familiarity is easy to mistake for corroboration. By the fifth reading you are no longer reading the message, you are reading your memory of the last four readings plus the interpretation you attached along the way. The practical fix is to write the message down verbatim once, with its timestamp, and treat that written line as the evidence — a written line does not get more damning overnight, and a remembered one does.

What is the difference between checking something and obsessing over it?

A check is an action that could come back negative and that you have agreed in advance to accept; obsessing is an activity that can only ever return more of what you already feel. Scrolling someone's followers, re-reading a thread, or asking a friend who only knows your version of events cannot exonerate anyone, so they are not checks no matter how much time they consume. If you cannot say what result would end the activity, it is not a check, and doing it more thoroughly will not turn it into one.

Does checking up on your partner change their behaviour?

Yes — visible monitoring changes the behaviour you are trying to measure, which is why heavy checking makes the evidence worse rather than better. Someone who senses they are being watched gets careful about ordinary things: they explain their whereabouts unprompted, they answer faster, they mention who they were with before being asked. Those changes look exactly like guilt from the outside, and they are equally consistent with an innocent person adapting to being audited, which means that once the watching starts you can no longer tell those two readings apart.

How can you tell if a suspicion is worth taking seriously?

A suspicion is worth taking seriously when it can be written as a sentence with a time in it that a stranger could confirm or refute from your own thread. “He has not initiated a conversation since the 14th” is testable and can be wrong. “He has been distant” cannot be wrong, so it can never be resolved and no amount of checking will settle it. The test is not whether the suspicion feels strong — strength is not a measure of accuracy — but whether it makes a claim precise enough to fail.

Is it normal to want to check your partner's phone?

Wanting to is common and acting on it does not work, because a phone search has no result you would accept as clearing them. A clean phone proves nothing about a deleted conversation or one held somewhere else, so the relief lasts days at most, and anything ambiguous you do find becomes an argument about how you got it rather than about what it was. It is worth naming the want out loud to yourself as a sign of how much the uncertainty is costing you, and then spending that energy on the one question you have not yet asked.

Can a suspicion be proved right or wrong by messages alone?

No, messages alone cannot prove a suspicion right or wrong, and this page is not a lie detector — inconsistency is not proof of deception. Messages can show that a specific claim does or does not fit the timestamps around it, which narrows the number of explanations still standing without eliminating them all. The realistic best outcome of a careful check is not certainty but a much shorter list: one dated, specific thing that the ordinary explanation does not cover, which you can then ask about directly.