Filtering Verbs

The most common filtering verbs: saw, heard, felt, noticed, realized, thought, knew, wondered, watched, smelled, sensed, found, decided, understood, observed.

Not all of these work identically. Saw and heard are the most purely mechanical — in most cases, they can be removed without loss. Realized and understood are more interesting: they signal the moment a character processes new information, which is sometimes exactly the dramatic event, and sometimes just a filter in front of it. Felt is the most overused and the most broadly damaging, because it appears in front of both emotional states ("she felt angry") and physical sensations ("she felt cold"), and both uses distance the reader from what’s being described.

Secondary filter constructions that function the same way but escape notice: she could see, he was aware that, it seemed to him, she had the impression that, he found himself noticing. These are filtering verbs in disguise — they do the same work but require more words and feel more literary, which makes them harder to catch in revision.

Why does filtering create distance? The cognitive mechanism is indirect but consistent.

A direct sentence — "The door was open" — puts a proposition in the reader’s mind without attribution. The reader processes it the same way the character does: as a perception. A filtered sentence — "She noticed the door was open" — puts two propositions in the reader’s mind: that someone noticed, and that the door was open. The first proposition is about a person’s cognitive state. The second is about the world. To get to the second, the reader has to pass through the first.

This creates a subtle separation. The reader is always one step outside the character’s experience — observing someone who is having a perception rather than sharing it. At the sentence level, this is barely perceptible. Over a chapter, the reader is perpetually at a slight remove, processing events through a layer of reported consciousness. Over a novel, this sustained distance is the difference between an immersive read and a technically competent one that nonetheless keeps the reader from losing themselves in it.

John Gardner’s psychic distance scale (in The Art of Fiction, 1983) provides the theoretical framework. Distance level 1 is pure authorial perspective: "It was winter of the year 1853." Distance level 5 is the character’s unmediated consciousness: "the goddamn cold." Filtering verbs reliably push prose toward levels 2 and 3 — interior report — and away from levels 4 and 5 where immersion lives. See Narrative Distance for the full scale and its implications.

Filtering verbs are not categorically wrong. They’re wrong when they’re habitual — when they appear not because the writer made a choice but because the writer instinctively reports rather than renders.

The legitimate uses are when the act of noticing or perceiving is itself the dramatic point. "She realized, too late, that she’d been wrong about all of it" — here the realization is the event, not the thing realized. The filter makes the lag legible: she understood after the moment when understanding would have mattered. Remove the filter and you have a bare fact; keep it and you have a tragedy of timing.

"He noticed she was crying" — in the right context, this is different from "She was crying." The first gives us his observation, his act of seeing; it might matter enormously that he noticed, this character, in this moment, with these implications for what he does next. The second gives us the bare fact. Which version serves the scene depends on whose consciousness the scene is tracking and what that tracking means.

Similarly, when unreliable perception is the point, the filter marks unreliability. "She thought she heard footsteps" puts the doubt on the table. "She heard footsteps" asserts. The filter isn’t weakness here; it’s the precise instrument.

The revision test: is this filter here because the act of perceiving is dramatic, or because I wrote it automatically? If automatic, cut it. If meaningful, keep it and make sure it’s earning its place.

This distinction matters, and conflating them causes overcorrection.

An internalization tag — "he thought," "she decided," "he told himself" — signals that what follows is interior monologue. It marks a transition from narration to the character’s direct voice. He thought: she was lying to him. The tag is doing structural work; it tells the reader they’ve entered the character’s head.

A filtering verb on perception — "he noticed she was lying" — is different. It’s not marking a transition; it’s adding a layer of report to what could be a direct perception. The filtering verb doesn’t enter the character’s head; it observes the character having a perception from the outside.

Both can be removed in certain contexts, but they’re not the same kind of word. Blanket anti-filter-word advice often fails to distinguish them, which leads writers to delete legitimate internalization tags in the name of "removing filters" and end up with interior monologue that lacks any signal to the reader that they’ve crossed into the character’s head.

The reliable distinction: does the filter precede a factual observation (filtering) or the character’s interpretation, decision, or direct inner speech (internalization)? Factual observation — cut the filter. Interpretation or direct inner speech — the tag may be serving a real function.

The instinct to report rather than render is almost universal in early drafts, and it isn’t random. It forms for two reasons.

The first is psychological safety. Writing "she noticed the door was open" is lower-risk than writing "the door was open." The filtering construction creates a subtle buffer: the sentence is not asserting a fact about the world but reporting someone’s perception of it. This feels more epistemically honest to new writers, as if claiming direct access to the reader’s experience requires a confidence they don’t yet have.

The second is drafting speed. In a first draft, filtering verbs are quick and reliable. The writer knows what the character perceives; the filter verb records it efficiently. Rendering perception directly — choosing the precise sensory details that convey the fact without naming it — takes more time, more attention to the specific texture of experience. The first draft values capture over craft. Filtering lets the writer get the content down.

This is fine for drafting. The problem is when filtering habits survive into final prose unchallenged. Revision is where filtering verbs should be evaluated systematically: not hunted down and eliminated, but treated as questions. Is this one doing meaningful work? Or is it automatic?

Filtering verbs are the primary diagnostic in Deep POV. The deep POV commitment is to maintain the closest levels of Narrative Distance — to write at Gardner’s levels 4 and 5 consistently — and filtering verbs are the primary mechanism that breaks that commitment. A manuscript that claims to be in deep POV but is saturated with filtering verbs is not in deep POV; it’s in reported interiority, which is a different register. The distinction between deep POV and close third person often reduces, in practice, to the density of filtering.

Free Indirect Discourse — the grammatical form that delivers the character’s voice without explicit attribution — is the alternative that deep POV typically reaches for. "The door was open. Who had left it like that?" reads as FID: the second sentence is the character’s direct thought, delivered without "she wondered" or "she thought." The technique is as old as Jane Austen; the term "filtering verbs" is modern workshop language for the habitual opposite.