Use AI to Find Where Your Verbs Are Doing Nothing

Why AI Drafts Default to Empty Verbs

Open a fresh AI draft and count the verbs in the first paragraph. I'll bet you lunch you find at least three was, two had, and a seemed doing all the work while the actual event hides in a noun. It's not a bug in the model — it's a feature of how language models predict. Statistically, "she was nervous" is a safe, high-probability continuation. "She pinched the skin between her knuckles until the white showed" is a riskier bet, so the model hedges toward the safe sentence unless you push it.

Three patterns show up again and again:

• Linking verbs — was, were, seemed, felt, became. These connect a subject to a description. They don't move anything. "The room was cold" tells you a temperature but nothing anyone did about it.
• Nominalizations — actions that got stuffed into nouns. "She made a decision" instead of "she chose." "He gave her a look" instead of "he stared." The verb is technically there (made, gave) but it's a hollow shell around a buried action.
• The "had" habit — the model loves past perfect for backstory and then forgets to come back. Three consecutive had clauses flatten a scene into summary.

None of these are grammatically wrong. That's exactly why they slip through. Your eye glides over them because there's no friction. But a page of coasting verbs reads like a synopsis of a scene rather than the scene itself. If you're just getting started with AI drafting and want the bigger picture of where this fits, the write a book with AI overview covers the whole pipeline — this guide is the verb-level pass inside it.

A scene with hollow verbs isn't badly written. It's unwritten — the events are there but nobody's hands are on them.

The fix isn't "replace every was with a stronger verb," which produces thesaurus-mangled prose. The fix is a diagnostic pass: find the verbs that are doing nothing, identify the physical action hiding inside each one, and rebuild the sentence around that action. That's what the rest of this guide does, prompt by prompt.

Building a Verb Inventory: Prompting AI to Tag Every Finite Verb by Category

Before you rewrite anything, you need an honest map of what's on the page. AI is genuinely good at this mechanical tagging work — it's pattern-matching at scale, which is precisely its strength. The trick is asking for a categorical inventory rather than vague feedback like "the verbs are weak."

Why it works: the category set forces the model to commit. Ask for "weak verbs" and you get a shrug. Ask for finite verbs tagged by function and you get a table you can actually scan. The ratio at the end is your headline number — a healthy action scene runs maybe 40% concrete or better; a draft coasting on was will sit under 15%.

Run this on three passages that feel different to you — a fight scene, a quiet conversation, a transition chapter. The contrast tells you where the problem concentrates. In my experience, dialogue-heavy scenes come back with surprisingly healthy verb counts and interiority or exposition passages are where the corpses pile up.

This pairs well with the Filter Word Finder, which catches the just, really, quite, felt layer of vagueness that sits on top of the verb problem. Do the filter pass first, then the verb pass; otherwise you're cataloguing verbs in sentences that are about to get shortened anyway.

The Nominalization Hunter: Finding Actions Trapped Inside Nouns

Here's the one that surprises people. "She made a decision" looks active — there's a subject, a verb, an object. But the actual happening, the deciding, has been compressed into a noun. The verb made is just a delivery mechanism. Grammarians call this nominalization, and it's the single most common way AI prose goes flabby without a single was in sight.

Other offenders: gave a look, took a breath, had a thought, offered a smile, let out a laugh, reached a conclusion, came to a stop, put up a fight. Each one buries a real, specific, physical act under a generic verb-plus-noun wrapper.

That third column is the whole ballgame. It's not enough to know "gave a look" should be "stared" — you need the physical specifics that make the replacement land in this character's body. The model will propose eyes and jaw; you pick the one that fits the person. A nervous character pinches the bridge of their nose. A contemptuous one doesn't blink.

Two cautions. First, don't hunt nominalizations in dialogue — people talk in these phrases, and "I made a decision" is exactly what a defensive character says out loud. Second, some nominalizations are load-bearing. "The investigation stalled" is fine if the investigation is the subject of a whole subplot and you don't want to name the detective. Use judgment; the table is a menu, not a mandate.

If your draft is genre-heavy, adjust the physical detail column to the world. In mystery novel with AI work, "made a decision" might resolve to "slid the case file three inches to the left" — a gesture another character can read. In romance novel with AI drafts, the same phrase often needs a proximity beat: "stepped in until the floorboards complained." The verb inventory doesn't know your genre. You do.

Rewriting Prompt: Swapping Abstract Verbs for Physical, Character-Specific Actions

Now the rewriting. This is where AI earns its keep — but only if you constrain it hard. Left alone, the model will swap "was nervous" for "felt a wave of anxiety," which is the same hollow verb wearing a fancier coat. You have to forbid the abstractions explicitly.

Rule three — observable by a camera — is the one that does the heavy lifting. It's the fastest way to kill "she processed her grief" and force "she folded the dish towel twice, then a third time." Rule four is what stops the rewrite from drifting into generic stage directions; you can pull that physical vocabulary straight from your story bible, where it should already live.

Run this pass scene by scene, never whole chapters at once. Long passages give the model room to get bored and revert to was by paragraph four. Short passages keep the constraint sharp.

One thing I do that the prompt doesn't say: I ask for the table, then I read only the right column as a standalone scene. If the rewritten sentences hold up without the originals beside them, they're real. If they collapse into "she moved, she looked, she turned," the model slid back into generic verbs and I rerun with a sharper physical constraint. You can catch the same drift by running the character consistency check afterward — the pass is built to notice when a character's body language suddenly stops matching who they are.

The good news is this pass gets cheaper and faster the more you do it, because you'll start writing the concrete verbs first. If you're planning the whole draft around this rhythm, How to Budget AI Drafting for a Full Novel is worth reading before you commit to verb-by-verb rewriting at scale.

Guardrails: When "Was" Is the Right Verb and When Flattening Verbs Is Overkill

Here's where I'll push back on the whole enterprise. Was is not a crime. "The knife was on the table" is a perfectly good sentence, and rewriting it to "the knife gleamed on the table" doesn't make it better — it makes it louder. The rule of thumb: was is correct when the state itself is the point, wrong when a held-back action is hiding behind it. You need the inventory to tell the difference, and the inventory will hand you false positives.

Three situations where you should leave hollow-looking verbs alone:

• Genuine states. "The door was locked." Nobody's doing anything. The lockedness is the information. Forcing a concrete verb here ("the deadbolt squatted in its socket") is purple.
• Deliberate rhythm. Short flat sentences can be a stylistic choice — a character in shock, a scene in freefall. Flattened verbs create a hush. Don't replace a hush with a stampede.
• Dialogue and thought. Real people say "I had a thought" and "she gave me a look." Nominalizations are how humans actually talk.

The verb pass is a detector, not a rulebook. Its job is to make you look at every hollow verb once and decide. Most will get rewritten. Some will stay, and now they'll stay on purpose.

One more guardrail worth knowing: the pass can wreck pacing if you run it on the whole manuscript at once. Concrete verbs are heavier per sentence, and a chapter where every clause ends in a physical action reads like a fight scene even when it's a funeral. Weight the pass toward the scenes that need motion and leave the quiet ones alone. If you're working inside the The Five-Pass Revision Order for AI-Assisted Novels, this verb inventory slot fits naturally as pass three, after structure and before line-level polish — it's wasted on a scene that's about to be cut entirely.

When you've done the pass on your draft, run the Manuscript Cleanup Report to confirm the ratio actually moved; it surveys exactly this kind of mechanical drift across the whole file and catches the verbs you missed while you were staring at one paragraph. If you're still deciding whether AI drafting in general fits your workflow, the Entangled Text vs ChatGPT comparison is a useful sanity check before you commit to the toolchain.

The takeaway: Build a two-part ritual. First, run the inventory prompt on any scene that feels flat, and read the CONCRETE ratio before you touch a word. Second, run the nominalization hunter on every passage of interiority or exposition, because that's where the buried actions hide. Rewrite from the table, sentence by sentence, never the whole chapter in one shot — and leave the was clauses that are doing genuine state-work alone. Do this on three scenes and you'll start hearing hollow verbs as you write, which is the real win: the pass stops being a pass and becomes a reflex.