Screenplay Dialogue Feedback: Which Notes to Act On

Three people read the same ten pages and you get three different dialogue notes. One calls a scene the best thing in the script. Another calls it flat. A contest reader writes "dialogue needs work" and stops there. Dialogue draws more feedback than any other element of a screenplay, and a smaller fraction of it is usable.

There's a reason for that, and it isn't that your readers are careless. Dialogue is the one part of a screenplay every reader feels qualified to judge, because everyone has been in a conversation. Structure notes tend to arrive with reasons attached. Dialogue notes often arrive with taste attached.

So the problem in front of you usually isn't that you need better dialogue advice. It's that you need a way to tell which of the notes on your desk are worth a rewrite — and a way to check your own pages before you hand them to anyone else.

TL;DR: Act on dialogue notes that name a mechanism, log the ones that report a reaction until three readers flag the same page, and check your pages yourself for the four things that actually break dialogue — on the nose, undifferentiated, overexplained, inert. Each has a pass/fail test you can run alone, and each has a specific repair. Rewriting to please the last person who read it leaves the real defect in place.

Because most dialogue notes report a reaction rather than name a mechanism, and reactions don't transfer between readers.

"This felt clunky" is real information that something is off, and no information about what. Two readers can both be responding to something genuinely wrong on the page and still send you in opposite directions, because each one is describing their experience of the lines rather than the property of the lines that produced it. Neither is lying. Neither is much use.

The structural reason is distance. A reader can articulate a plot problem — "nothing happens between pages 30 and 55" — because plot problems are visible from far away, in the shape of the thing. Dialogue problems are line-level. By the time a reader notices something is wrong, they've already had the reaction, and the reaction is what they can describe. Naming the mechanism underneath it takes a second, more deliberate pass that most readers don't make and aren't being paid to make.

Writers respond to this in two predictable ways, and both are expensive. The first is to rewrite toward whoever read it most recently, which produces a scene assembled from four people's preferences and belonging to none of them. The second is to decide dialogue notes are all subjective and ignore the lot. Both leave the actual defect sitting on the page.

The way out is to sort before you rewrite. Not every note deserves a pass, and the ones that do announce themselves.

Act on any note that names something happening on the page — something a second reader would see the same way, whether or not they share the first reader's taste.

State the test plainly and it does most of the work: could a different reader verify this without sharing the first reader's taste? If yes, it's a mechanism, and it's actionable. If no, it's a reaction. Log it, date it, and wait. When three readers flag the same page, the pattern is the note even if none of them could name what's wrong.

The same problem gets described both ways constantly:

The right column isn't a better class of reader. It's the same note, one question deeper — which means you can often get there yourself. Go back with a single question: which line, and what did you notice? Most readers can point at the line even when they can't name the problem, and the line is the part you need. The diagnosis is yours to make.

Contradictions deserve a different move. Two readers disagreeing about a scene usually means the scene is doing something ambiguous, not that one of them misread it. Look at what they disagree about — that element is carrying either too much weight or not enough, and it's the thing to examine.

One concrete fix: before you open the script, put every note you're holding into two columns. Most writers find the actionable column is shorter than they feared and clearer than they expected.

Nearly every dialogue note that survives scrutiny is one of four problems. Each has a tell you can check on the page without an outside opinion.

On the nose. Characters say exactly what they mean and feel. The tell: strip out the action lines and nothing is lost, because all the emotional content is already in the words. It reads as unreal because people almost never state a position directly when stating it costs them something — so articulate, fully-expressed feeling registers as performance. This is the dialogue-level version of a problem that also shows up at story level, where a script states its argument instead of dramatizing it; if coverage has ever called your script preachy, the same defect at the thematic level is worth reading about.

Undifferentiated. Everyone sounds like the writer. The tell is the name-cover test below. What it costs you is attention: the reader stops tracking who's speaking and starts skimming, and skimming is how a good scene gets a bad note.

Overexplained. Dialogue hauling exposition the scene should be carrying, usually delivered to characters who already know it. The tell: someone tells another person something both of them lived through.

Inert. Nobody wants anything from anyone. The tell: the exchange could stop at any line and lose nothing, because nothing is being pursued. This is the most common cause of "flat" and the least often named, which is why writers so often respond to a flatness note by sharpening individual lines — polishing an exchange that has no engine.

These overlap more than they separate. An inert scene is usually also overexplained, because exposition is what a writer reaches for when the exchange has nothing else to do.

It's worth looking at how produced films handle the same pressures, at the level of what the exchange is doing rather than what is said. The Social Network differentiates characters inside a single fast exchange through rhythm and register — who interrupts, who completes a thought, who changes subject — rather than through distinct vocabularies. No Country for Old Men carries a threat almost entirely by indirection, the pressure sitting in what the speaker declines to state. Jaws has men comparing histories in a scene where the ostensible subject of the conversation isn't the subject of the scene at all. In each case the mechanism is available to you without imitating anyone's voice.

Four mechanical tests. Each returns an unambiguous answer, none requires another person, and all four take under an hour on a scene. Run them on your worst scene first.

1. The name-cover test. Print the page or paste it with the character names hidden. Try to assign the lines. If you can't, the characters aren't differentiated. This is pass/fail, not a judgment call, which is exactly what makes it useful when you don't trust your own ear yet.

2. The cut-the-first-line test. Delete the opening exchange of the scene. Most scenes start earlier than they need to, and the first lines are throat-clearing — greetings, arrivals, restatements of what the audience just watched. If the scene is better without them, they were the problem, and there's a good chance the last exchange is doing the same thing at the other end.

3. The want test. For each character in the scene, write one sentence: what do they want from the other person, right now, on this page? Not in the story. On this page. If you can't write the sentence, that's the note, and it's the inert diagnosis arriving without anyone having to give it to you.

4. The read-aloud test. Read it out at speed, not carefully. Anywhere you stumble, the line is wrong — usually too long, too written, or missing a beat a person would naturally take.

What these tests can't catch is worth stating, because the tests are not a substitute for a reader. They won't tell you whether your voice is any good, whether the scene is funny, or whether the register suits the genre. What they do is clear the mechanical failures out of the way, so that when someone does read it, their attention goes to the things only a reader can judge. Handing a script to a reader with undifferentiated dialogue still in it means paying for a note you could have written yourself.

One concrete fix: run the name-cover test today, on the scene you're least sure about, before the script goes anywhere.

Five sources, each good at something different and none good at everything. Choose by which of the four failure modes you suspect.

• A writers group. Free, high volume, weekly. Good for hearing your pages read aloud by other people, which surfaces stumbles instantly and is genuinely hard to replicate alone. Bad for dialogue specifically, because dialogue is where taste opinions concentrate and groups drift toward line-editing each other's word choices.
• One trusted reader who reads a lot of scripts. Free, the best signal available, and scarce. Good for everything. Limited by availability and turnaround, and by the fact that you can only spend that goodwill so many times before it thins.
• A contest or festival reader. Roughly $50–$100, bundled with the entry. Good for an outside verdict with no relationship attached to soften it. Bad for dialogue depth — you typically get a paragraph, and how useful that paragraph is varies by reader.
• A script consultant. $200 and up. The most specific, most tailored notes you can buy, including line-level work on individual scenes. The constraint is cost per pass, which makes it hard to use iteratively while a draft is still moving.
• Coverage services and tools. Roughly $10–$50 per script. Good for consistency across a full script, speed, and page-cited specificity you can act on directly. What they don't have is any relationship with your work, which means they won't tell you whether your voice is worth pursuing.

Scanning by price: free (group, trusted reader) · $10–$50 (tools and coverage services) · $50–$100 (contest reads) · $200+ (consultants). Two of those rungs are covered in more detail elsewhere — the free options worth trying and paid tools tested side by side — and the general question of what a machine read and a human read each give you is its own comparison.

StoryNotes published this article, and one of the options listed is ours. It sits in the $10–$50 tier: a graded read that cites the page and the slug line where a scene goes inert and returns concrete rewrites to work from, $20 per script, one-time, no subscription. On dialogue specifically, the value is that it names the mechanism rather than reporting a reaction — the thing most notes leave out. It's detailed enough to rewrite from and useful at whatever stage the draft is in. What it won't do is tell you whether your voice is worth pursuing. To see what page-cited notes on a scene actually look like, there's a sample report on a public-domain script.

Match the repair to the mechanism. These are four fixes, not a style guide.

For on the nose: cut the line that states the feeling and move the information into behavior — what the character does with their hands, what they refuse to answer, what they walk away from. The action line can carry what the dialogue was explaining, and usually carries it harder.

For undifferentiated: give one character a habit of speech the others don't have. Length, formality, whether they ask questions or make statements, whether they finish sentences. Rhythm differentiates more reliably than vocabulary, and it survives being read by someone who has never met your characters.

For overexplained: move the exposition to the moment the audience needs it rather than the moment it's convenient to deliver, and put it in the mouth of the person who has a reason to say it out loud.

For inert: give one character something they need from the other, and give the other a reason to withhold it. An exchange with a want and an obstacle is a scene. Without them it's an update.

The general pass, in order: run the four tests, name the mechanism, apply the matching repair, then delete the first and last line of the scene and read it again. Where this sits in a revision — and what to do before and after a dialogue pass — is covered in the coverage-to-rewrite workflow.

One caveat, and it matters more here than in any other craft area. Dialogue is the most style-dependent element in a screenplay. Plenty of excellent scripts go on the nose deliberately, or let every character sound like the writer as a signature, or run three pages on a subject nobody needs explained. These are diagnostics for lines that aren't working. They are not rules for lines that are. If a scene tests badly and still plays, trust the scene.

One concrete fix: take the note you've been arguing with — the one you keep half-agreeing with and not acting on. Work out whether it names a mechanism or reports a reaction. If it names a mechanism, run the matching repair on that one scene tonight.