A narrator tells you something. You believe it, because believing the narrator is the default condition of reading. Forty pages later you realise they were wrong — or worse, that they knew.
Done well, this is one of the most powerful things fiction can do, because it makes the reader complicit in a misreading and then shows them the machinery of their own credulity. Done badly it feels like being cheated at cards. The difference between the two is almost entirely a question of what was on the page before the reveal.
Unreliability is a property of the character, not the plot
The most common mistake is to treat unreliability as a structural device — a twist mechanism bolted onto a story to create a late surprise. That is where the cheated feeling comes from. The reader senses, correctly, that the withholding served the author rather than the character.
Unreliability that works is a symptom. It comes from something true about the person telling the story, and it would be there whether or not there was a twist coming.
Consider the range of causes, because they produce very different books. A narrator who simply does not have the information — a child describing an adult catastrophe, a servant who is not told things — misleads without any distortion of character. A narrator whose judgment is bent by what they need to believe is a different creature: they are not lying, they have arranged the facts so that they can continue to live with themselves. A narrator actively constructing a false account for a reader they know is there is different again, and the most technically demanding to write, because everything they say must be defensible twice.
Craft discussions of the device tend to split it along the axis of intent — whether the distortion is deliberate — and that split is useful, but the more productive question for a writer is narrower: what does this person need to be true, and what would they have to not-notice in order to keep it true?
Answer that and the unreliability writes itself, because it stops being a decision you make sentence by sentence and becomes a consistent pressure on everything the character sees.
The self-deceiving narrator is the richest version
The versions that last are rarely the con artists. Amy Dunne in Gone Girl is a superb piece of construction, but the narrator who stays with you for years tends to be the one who is lying to themselves in public.
Kazuo Ishiguro's The Remains of the Day is the cleanest example I know. Stevens, the butler, tells you about his professional life with enormous dignity and complete conviction. He is not lying. Every fact he reports is probably accurate. What he cannot do is register the meaning of any of it — the political position he served, the woman he loved, the decades he spent making a virtue of not feeling things. The reader assembles the tragedy out of material Stevens hands over without understanding it.
Nothing is concealed. That is the point. The information is all there, the narrator simply cannot see what it adds up to, and the reader's growing ability to see it is the entire emotional engine of the novel.
Humbert Humbert works the same way in reverse: his ornate, ravishing prose is the instrument of his self-justification, and the reader's job is to notice that the beauty of the sentences is doing the work of hiding what is being described. The more gorgeous the language becomes, the more it should alarm you.
Signalling without telegraphing
The reader has to be able to catch it. Not necessarily on the first pass — but the evidence must be sitting there, in plain view, so that a second reading is a different book.
The signals are mostly small.
Insistence. A narrator who keeps returning to how reasonable they are being is telling you something. "I want to be clear that I bore her no ill will" is a sentence that means its opposite, and readers feel it without being able to say why.
Gaps. Time that is passed over quickly, an event described in unusual summary when everything around it is scene, a chapter that ends just before something happens. Vagueness is only noticeable against a baseline of precision, which is why unreliable narrators should be specific about nearly everything else.
Second-hand emotion. A narrator who tells you what other people were thinking, repeatedly and confidently, is characterising themselves. "I could see she was ashamed" is a claim about the narrator, not about her.
Mismatch between report and scene. This is the strongest one, and worth its own section.
Let the dialogue contradict the narration
The most reliable technique is simple: let the narrator interpret a scene for you, and let the scene as rendered say something else.
She asks how his day was, and her voice is light, and she mentions she booked the restaurant he likes. And the narrator tells you that even her kindness is a way of keeping score, that he can see straight through it.
The reader now has two accounts. The dialogue is evidence; the narration is interpretation. Nothing has been withheld and nothing has been announced, and the gap between the two is where the character lives. This works because it puts the reader in the position of a third party in the room — which is exactly the position from which you notice that someone is not seeing their own marriage clearly.
The same trick, inverted: a narrator who reports having told someone the whole truth, and a scene in which the other person says nothing and leaves. The reader does not know what was said. The narrator's certainty about what the silence meant is the tell.
Consistency is what earns the reveal
Internal logic is the thing that separates a satisfying unreliable narrator from a resented one. The narrator can be wrong about the world. They cannot be arbitrary.
If a narrator believes their business partner has been stealing, every observation they make should be consistent with that belief — including their interpretation of evidence that points the other way. Explaining away a contradiction in character is powerful. Ignoring it is not, and readers register the difference immediately even if they could not articulate it.
This is why you have to know what actually happened, in full, before you write a word of the distorted version. Not as a document you will use, but so that every scene can be written twice: once as it occurred and once as this person would report it. The gap between those two drafts is the novel.
The most interesting case: partial self-awareness
The narrators I find most affecting know they are being unfair and cannot stop.
Someone who says, plainly, that they understand their resentment is unreasonable, that the person they resent was young and frightened and has apologised, that they know all of this — and who then spends the next two hundred pages unable to act on any of it. That is not a puzzle to be solved by a reveal. It is a condition, recognisable, and most readers have been in it.
This version has no twist and needs none. The reader is not being misled about facts, only shown the distance between what a person knows and what they can do. It is the hardest to write because there is no mechanism carrying it, only the character.
The test
Read your draft as someone encountering it for the first time, then read it again knowing the truth.
On the first pass, does the narration hold together on its own terms? On the second, is there something on nearly every page that you now read differently? If the second reading only changes three scenes, the unreliability is a plot device. If it changes the whole texture, it is a character.
And the reaction you are aiming for is not "I didn't see that coming." It is the slightly uncomfortable one that follows: that you did not see it because you did not want to, and that the narrator's blindness was a version of your own.
Tags
Taresh Sharan