How to write an unreliable narrator

Every unreliable narrator promises the same deal: trust me, and I'll tell you a story. Then the story quietly breaks that promise. The narrator lies, forgets, distorts, or doesn't understand what's happening around them, and the reader ends up doing detective work on a text that was supposed to be a straight account. Done well, this is one of fiction's most powerful tools. Done badly, it reads like a magic trick where you can see the wire.

What actually makes a narrator unreliable

Unreliability isn't one thing. It's a spectrum of reasons a narrator's account diverges from the truth, and the reason you pick determines everything else about the book.

A naive narrator doesn't lie; they lack the context to interpret what they see, like a child narrating a parent's affair as a run of odd houseguests. The reader assembles the truth from details reported accurately but read wrong.

A self-deceiving narrator is lying, but the lie is aimed inward. They believe it because believing it is the only way to keep functioning: this is the narrator who describes their temper as standing up for himself, their drinking as unwinding.

A biased narrator tells the truth as they see it, filtered through resentment or self-interest. Nothing said is false, but the selection and emphasis tilt the picture.

A deliberately deceptive narrator knows the truth and withholds or falsifies it on purpose, usually while confessing or running a con on the reader.

A narrator with altered perception is unreliable for reasons outside their control: illness, trauma, medication, grief. They report what they genuinely experienced, even when it doesn't match events.

Pick one as your primary mode. You can layer a little bias onto a self-deceiving narrator, but one who is simultaneously naive, biased, and running a deliberate con stops being a character and becomes a plot device. The reader can't track what kind of gap to look for.

The gap between what's said and what's true

The craft problem is simple to state and hard to execute: the reader must sense a gap between what's said and what's true, without the narrator ever seeming aware of creating it.

You can't do this with a single flag or aside; it has to accumulate through details that don't add up alone but do in combination. A useful method: write the true version of events first, off the page, as a private beat sheet. Then write the narration and let the truth leak through details the narrator doesn't think matter, how other characters react to something dismissed, a fact repeated with slightly different details each time, a description that contradicts an emotional claim.

Here's the same beat rendered two ways: first as an overt tell that flattens the effect, then as a plant that respects the reader.

OVERT (breaks the illusion):
"I told her I hadn't touched her diary, even though of course
I had. I always lie about things like that."

PLANTED (preserves it):
"I told her I hadn't touched her diary. She just looked at me,
the way she does when she already knows, and didn't argue. I
never understand why she stops arguing so fast with me."

The second version never admits the lie. It hands the reader a mismatch, her reaction doesn't fit "I'm innocent", and trusts them to do the arithmetic. That trust is what makes unreliable narration feel earned, not announced.

Dialogue is your best tool here because the narrator can't control what other people say back, only report it. If a narrator claims to be calm and everyone else keeps asking if they're okay, the reader believes the room, not the narrator.

When to let the reader in on it

Timing decides whether the technique lands. There are three points at which a reader can pick up on it, each doing a different job.

EARLY (first chapter): reader reads defensively from page one.
    Good for mystery and thriller structures.
FAIR-PLAY (middle third): reader has enough evidence to suspect
    something before confirmation, so the reveal feels earned.
    Good for most literary uses.
LATE TWIST (final act): reader has read in good faith and the
    ground shifts. Only works if a reread makes earlier
    chapters click.

Decide which one before you draft the middle of the book; it changes how much the narrator can get away with along the way.

The two ways this fails

The first failure is revealing the unreliability too soon or too bluntly, so the reader clocks the trick before the story has done anything with it. A line like the overt example above in chapter two leaves the mystery gone: a narrator you already know not to trust for three hundred pages, with nothing left to discover.

The second failure is the opposite: withholding so completely, or piling on so many kinds of unreliability at once, that the reader stops trusting the narrator to mean anything. Once nothing can be relied on, they disengage; there's no foothold left to read from. This usually happens when a late twist retroactively invalidates huge stretches of narration with no prior evidence, feeling less like a revelation and more like the rules changed without telling anyone.

The narrators readers remember, the governess in The Turn of the Screw, or Amy Dunne, work because the unreliability has a shape. The reader can name what kind of liar or self-deceiver they're dealing with by the midpoint, even without seeing everything hidden. That's the target: not total mystery, not total exposure, but a legible untrustworthiness the reader can read around.

One practical side effect of writing this kind of narrator: you end up needing to track what they've actually witnessed versus what they only claim, and it's easy to lose the thread over a long draft. Chaptered's entity mentions track character and place references as you write, so when you need to check whether your narrator could really have known something they just asserted, you can see what they've actually been present for, rather than trusting memory.