Overexplaining: why extra detail reads as a tell

Overexplaining does not mean someone is lying. It reliably indicates that the person anticipated disbelief, which is a fact about the situation they were answering in rather than about the truth of what they wrote.

Work through who produces long, detailed answers to simple questions. Someone who has been disbelieved before, by you or by somebody else, and has learned that a plain answer gets challenged. Someone anxious, for whom the gap between sending a short reply and receiving a response is unbearable, so they fill it. Someone who has been in trouble for under-explaining. Someone whose job or upbringing trained them to justify. Someone with ADHD or autism, for whom the full context is simply how an answer is constructed. Someone who is in the wrong about something adjacent and is padding around it. And, yes, someone assembling a cover story. That last group is one entry on a long list, and nothing about the length of the message distinguishes it from the others.

The intuition has a real basis and it is worth naming precisely rather than dismissing, because dismissing it will not stop you having it. A fabricated account does take more construction than a recalled one, and construction can show. The problem is that the reader has no access to the difference between a message that took effort because it was invented and one that took effort because the writer was scared. Both arrive as a long message. You are being asked to distinguish two causes from a single observable, and it cannot be done from the observable alone.

The lists treat both as tells because they were built to be comprehensive rather than testable, and a rule that fires on long answers and short answers has no failing condition.

This is not a subtle inconsistency buried across different sources. The most widely shared list of signs someone is lying over text contains “providing excessive details” and “using highly descriptive language” alongside “giving one-word answers” and “offering no explanations”. Both directions are on the same page, a few hundred words apart. Whatever length of message you are holding, one of those items covers it, and the reader — who arrived worried — will find the one that matches and stop reading.

Once you notice that structure you can apply it as a filter to everything in the genre. Any deception cue that has a mirror image also on the list is not a cue; it is a way of describing whatever happened after the fact. The honest version of the length observation has to specify in advance which direction counts and against what comparison, and the only comparison that makes that possible is the person’s own usual output. That is the whole of the method on this page, and it is why the tool below refuses to say anything at all about someone whose writing it has not seen enough of.

There is no universal answer, and the only usable one is relative: too much detail is detail well beyond what this specific person normally produces for a question of this size.

Get the number rather than the impression. Scroll back to an ordinary stretch of the thread — no argument in it, nothing pending — and look at how long their replies actually are. Some people answer everything in four words. Some people cannot buy milk without a paragraph. If their median message is thirty words, a sixty-word answer is not an outlier and your sense that it was unusually long is a sense about how you write, not how they do. If their median is five words and the answer to “what time did you leave?” ran to a hundred and forty, that is a genuine departure and it is worth holding on to.

The comparison stretch has to predate the current problem. Once a conversation has become tense, everybody writes differently, and measuring a tense message against other tense messages normalises exactly the state you were trying to detect. If the thread does not go back far enough, or things have been strained for its whole duration, you do not have a baseline and the length observation is not available to you. That is a real finding rather than a dead end: it means the answer has to come from a checkable claim or a direct question instead.

Ratio matters more than raw length, and it is the part people skip. A hundred-word answer to “where were you between nine and midnight?” is proportionate. A hundred-word answer to “did you get milk?” is not, from anybody. Compare the size of the answer to the size of the question, then compare that ratio to how this person normally handles questions of similar size — which is the calculation the baseline method exists to make possible.

If the detailed answer arrived immediately after you signalled disbelief, you produced it, and it carries almost no information about whether the disbelief was justified.

Questions are not neutral instruments. “What time did you get back?” and “what time did you actually get back?” are the same request with entirely different instructions attached, and the second one asks for a defence. So does a question asked twice. So does a question with a timestamp attached, a question that arrives after three hours of silence, and a question phrased so that a short answer would obviously be insufficient. Each of those tells the recipient that a plain answer will not be accepted, and a person who has been told that will supply supporting detail, because supplying supporting detail is the reasonable response to not being believed.

The uncomfortable consequence is that the situation in which people most want the overexplaining tell — a confrontation — is precisely the situation where it is least available. You cannot accuse someone and then treat their defensive answer as confirmation. The measurement was contaminated by the act of measuring, and both a guilty and an innocent person return the same reading.

What survives is unprompted elaboration. Detail nobody asked for, arriving before any challenge, in a message that could have been three words — a full itinerary attached to “running late”, an explanation of who else was present when nobody asked who was present — is an observation about them rather than about your question. Even that is not evidence of deception; it is evidence that the person felt a need to justify, which some people feel constantly and about everything. But it is at least an observation you did not manufacture, which is more than the confrontation version can claim.

The checkable parts: a named place, a specific time, a person who exists, a journey that either happened or did not. Everything unverifiable — how tired they were, what they were thinking, how loud it was — cannot be tested and adds nothing to what you know however much of it there is.

Sort the message into those two piles before you react to it. The exercise takes a minute and it does something valuable: it converts a wall of text that feels overwhelming into a short list of claims and a large pile of nothing. Very often the second pile is nearly the whole message. A four-paragraph answer that contains no place, no time and no third party has told you a great deal about the writer’s state and nothing about the evening, and no amount of re-reading will extract more, because there is nothing in there to extract.

When the checkable pile has something in it, that is the useful outcome, and it points both ways. A person who volunteers checkable detail has handed you something that could exonerate them, and it usually does — the place was real, the friend confirms, the timings work. Treat that as a result and stop, because a check you refuse to be satisfied by was not a check. If the checkable detail fails, you have a specific discrepancy rather than a feeling, and a specific discrepancy is something you can raise in one sentence.

Most of it does not transfer, because it was developed for live questioning and it depends on conditions a text message does not have.

The best-known interview techniques rely on real-time pressure. Asking someone to tell the story in reverse order works because reversing a narrative under time pressure is cognitively demanding and cannot be prepared. Withholding what you already know and releasing it gradually works because the person must commit to an account before learning what you can contradict. Both assume an interviewer, a clock, and no opportunity to review. A text message has none of that: it is composed at leisure, revised before sending, answered whenever the writer chooses, and frequently written with the entire prior conversation visible on screen above the reply box.

Two things do survive the move to text, and they are both structural rather than behavioural. Sequence survives: the thread records what was known at the time each message was sent, so an account that gained its supporting detail only after you revealed what you already knew is visibly different from one that had the detail all along. Repetition survives: a written record lets you compare two tellings word for word, which no live interview permits, and word-for-word identity across days is itself notable, because people normally re-word what they recall. Both are comparisons between records rather than readings of a single message, which is the pattern everything reliable in this cluster follows.

What does not survive is the entire behavioural half — hesitation, eye contact, filler, the pause before answering. None of it exists in text. A three-minute delay before a reply means the phone was in a pocket, and the composed message that arrives afterwards looks the same whether it took eight seconds or eight drafts.

If you explain at length by default and are being told it looks like guilt, the useful move is to answer the question first in one line and put the context underneath — because the problem is not the amount of detail, it is that the answer is buried inside it.

Notice what the other person is actually experiencing. They asked what time you left. Four paragraphs arrived. Somewhere in the third one is “about eleven”, and to find it they had to read a description of the traffic, an account of who suggested going on somewhere, and a note about how tired you were. Reading that feels like being handed a document instead of an answer, and that feeling — not any theory about deception — is what most accusations of overexplaining are reporting. Leading with the direct answer fixes it almost entirely, and it costs you nothing, because the context can still follow.

It is worth knowing why you do it, because the reason is usually not about this conversation. People who over-explain have very often learned it: from a parent who demanded justification, a job where a short answer got picked apart, an ex who treated brevity as evasion. That history makes a plain answer feel dangerously exposed, and the extra detail feels like insurance. It is not insurance. In practice it does the opposite, because volume is the thing being misread, and every additional paragraph adds to the pile without adding to the answer.

Do not go to the other extreme in response. Someone told their long answers look suspicious will sometimes switch to two-word replies, and two-word replies are on the same lists as suspicious too — which is the whole problem with those lists, and it is not a reason to police your own writing into some imagined neutral register that does not exist. Answer the question that was asked, add the context you want to add, and if you are with somebody for whom no length of reply is acceptable, the difficulty is not your prose style.

Pull out the one checkable claim inside it and ask about that, in a single message, without commenting on the length.

Responding to the length is the move that guarantees a worse conversation. “Why are you telling me all this?” or “that’s a lot of detail for a simple question” tells the other person that explaining did not work, and the reliable human response to being told that explaining did not work is to explain more. You will get a longer message back, and you will read it as further confirmation, and the loop has no exit. Say nothing about the volume. Take the fact you can test and ask about the fact. “Which place did you end up at?” is a whole message.

Then let the answer land. If the checkable thing checks out, that is the finding, and it is a good one — this is the case the incumbent pages never account for, the one where the reader turns out to have been wrong and deserves to be able to stop. If it does not check out, you have a specific discrepancy with a time and a place attached, which is raisable in one sentence and much harder to argue away than an impression about somebody’s writing style.

The tool below measures each person’s messages against their own usual length and their own usual rate of qualifiers, and reports nothing at all when it has fewer than six of their messages to work from — because a baseline built from three messages is not a baseline, and a number produced from one would be exactly the kind of false precision this page exists to argue against. When you want the whole conversation read rather than one person’s writing measured, you can run the thread through the analyser, and it will still decline to tell you what anyone intended.

• The words people use when a story is thin — the method this page uses, step by step.
• Gaslighting phrases in text messages — The phrase is never the evidence. The effect on you is.
• “I never said that” — when the receipts disagree — The message proves what was written. Not what was remembered.

Does overexplaining mean someone is lying?

No, overexplaining does not mean someone is lying. Long, detailed answers are produced by people who expect not to be believed, and expecting not to be believed is caused by being accused, by having been disbelieved before, by anxiety, and by simply being the kind of person who explains things — as well as by having something to hide. Because the same lists that treat long answers as a tell also treat one-word answers as a tell, the reading cannot fail either way, which means it is not a test at all.

Why do people give too much detail when they are being questioned?

People give extra detail when they think a plain answer will not be accepted, which is a reaction to the question rather than a property of the answer. A question that arrives with an edge — repeated, timed, or clearly already decided — invites a defence rather than an answer, and a defence is longer than an answer by design. This is why the tone in which you asked is part of the evidence: if the extra detail appeared immediately after you signalled disbelief, you produced it.

Is a short answer or a long answer more suspicious in a text message?

Neither is suspicious on its own, and the only meaningful comparison is against how that specific person normally answers. A two-word reply from someone who writes in two-word replies is nothing; the same reply from someone who normally writes paragraphs is a change worth noticing. Length carries information only as a departure from a person's own baseline, and even then a departure indicates that the moment was difficult or effortful, not that the content was false.

Can the length of a text message show that someone is lying?

The length of a text message cannot show that someone is lying. Not a lie detector — inconsistency is not proof of deception. What message length can show is that a particular answer cost more effort than that person's usual output, which is a real observation about the moment and a silent one about the reason. A long answer from someone who has just been accused is the expected result whether the accusation was right or wrong, so it carries almost no information in the situation where people most want it to.

What kind of detail in an explanation is actually worth checking?

Checkable detail is worth checking: a named place, a time, a person, a receipt, a journey that either happened or did not. Unverifiable detail — how tired someone was, what they were thinking, how loud the bar was — cannot be tested at all and adds nothing to what you know regardless of how much of it there is. Sorting an explanation into those two piles is more useful than measuring its length, because the checkable pile is the only part that can either exonerate or fail.

Does the advice about spotting lies in interviews work on text messages?

Most of it does not transfer, because it was developed for live questioning. Techniques such as asking for an account in reverse order, or holding back what you already know and releasing it gradually, depend on real-time pressure and on the other person not being able to re-read and edit before answering. A text message is composed, revisable, and answered at whatever pace the writer chooses, which removes exactly the conditions those methods rely on.

What should you do when you get a very long answer to a simple question?

Find the checkable claim inside it and ask about that one thing. A long answer usually contains one or two factual assertions surrounded by context, and the context cannot be tested no matter how carefully you read it. Pull out the specific claim — the place, the time, the person — and ask a single question that could resolve it, rather than responding to the length itself, which will only tell the other person that explaining more did not work and invite them to explain more.