If You Strip the Dialogue Tags, Can You Still Tell Who's Speaking?
Here's a test you can run in ten minutes on a scene you already think is fine.
Copy a conversation out of your draft into a blank document. Delete every "he said" and "she asked." Delete the action beats too, the ones where somebody sets down a cup or looks out a window right before they speak. Just the spoken lines, stacked up, nothing attached.
Now read it aloud and see if you can tell who's talking.
Most writers find the first exchange holds, the second one wobbles, and somewhere around line seven they're working it out by logic rather than by ear. Which is the answer. If you have to reason it out, so does your reader, and they're doing it on a train with somebody else's music leaking out of their headphones.
The tag advice can't help you with this
Go and read what ranks for dialogue tags. Reedsy, The Write Practice, Jericho Writers, Story Grid, Louise Harnby. They're careful, useful pages and between them they cover the mechanics properly: how to punctuate a tag, why "said" beats a thesaurus, where action beats go.
The consensus is right, by the way, and I'm not relitigating it. Renni Browne and Dave King make the case in Self-Editing for Fiction Writers about as well as it can be made: plain attribution is nearly invisible to a reader, while an ornate verb pulls the eye off the dialogue and onto the author. They also warn against smuggling emotion into the tag, propping up a flat line with an adverb instead of fixing the line. Both points hold.
But notice what every one of those pages assumes. They assume your problem is WHICH WORD to attach. And the question that actually matters is whether you should need to attach one.
One of the ranked pages goes further and recommends the opposite of the repair. The advice is roughly: use expressive verbs like growled or snapped and you can tell your speakers apart without leaning on tags. Read that again. It's proposing you fix indistinguishable voices with more decorated attribution... which is the same crutch in a better outfit. If a reader knows it's Marta because the tag says she growled, they still don't know Marta.
Tag reliance is a symptom
A dialogue tag points at a voice. That's all it does. If the voice it's pointing at is indistinct, the pointer hasn't fixed anything, it's just told you where to look.
So when your tags become load-bearing, when removing them makes the scene genuinely hard to follow, that's not a formatting lapse. It's evidence that two characters are speaking with the same mouth.
This connection isn't something I discovered. Liam Carnahan put it plainly in a 2017 piece for Writers Helping Writers: "Without unique voices, conversations between characters can confuse your readers, and you'll have to be overly reliant on dialogue tags to keep everything straight." That's the whole thesis of this post in one sentence, published nine years ago.
The strip-the-tags test isn't new either. It turns up in craft posts here and there.
What I haven't found anywhere is the two put together into something you can actually RUN, with a decision at the end of it about what to change. That's what the rest of this is.
The audit
Do this on one scene at a time. Two speakers is the cleanest test, and I'll come back to crowded scenes later.
- Pick a scene with at least two named characters and ten or more lines of dialogue. Pick one you're confident about. The confident ones are more informative.
- Strip it. Copy into a new document, delete every tag and every action beat that identifies a speaker.
- Read it aloud, once, at speed. No stopping to reason. The first-pass instinct is what you're measuring, because that's what a reader has.
- Mark every line where you weren't sure, or where you were sure and wrong.
- Look at how the marks fall. This is the whole diagnosis. A single ambiguous line in an otherwise clean scene is a local problem, and you fix that line. Marks scattered all the way down are voice collapse, and no amount of line-fixing will touch it.
- For collapsed scenes, find the shared dimension. Which of these are the two characters doing identically?
- Vocabulary register. Do they reach for the same class of word? Are they both plain, or both ornate, or both technical?
- Sentence rhythm. Do they run the same lengths? Two characters who both speak in eight-to-twelve word sentences will blur no matter what the words are.
- Verbal habits. Hedges, qualifiers, questions answered with questions, sentences trailing off. These are the strongest differentiators available and most drafts give every character the same set.
- Emotional directness. Does this person say the thing, or circle it? Two circlers are indistinguishable. A circler and a blurter never are.
- Change the voice, not the tags. Don't touch attribution while you do this. You'll be tempted, because it's faster and it makes the symptom go away.
- Rerun the strip. Same scene, same process. If the marks are gone, put back only the tags the scene still needs, which will be fewer than you had.
I'm deliberately not giving you a threshold, by the way. I could say something like "more than one line in four is a problem" and it would sound authoritative, but I'd be making the number up, and a made-up number gets quoted back at people as though somebody measured it. What you're looking for is the difference between stumbling once and never being sure, and you'll know which one you've got.
Overlap is fine. Indistinguishability isn't.
Worth guarding against the version of this that makes books worse.
Characters who grew up in the same house should sound related. Two surgeons should share a register. If you respond to a failed audit by giving one of them a catchphrase and the other a regional tic, you've made the scene easier to follow and the book harder to believe.
The test was never whether two voices are DIFFERENT. It's whether a reader can follow the exchange. Two characters can share vocabulary and rhythm completely and still be perfectly trackable if one of them says what he means and the other never does. One dimension, worked properly, will carry a whole scene.
When tags are doing real work
None of this is an argument for stripping tags out of the finished book.
Three or more speakers in a fast exchange need attribution and always will, because voice differentiation doesn't scale past about two before a reader has to hold too much at once. A line where who said it changes what it means needs a tag for the meaning, not the tracking. Coming back to dialogue after a long stretch of narration needs a tag to re-establish the room.
Tags earn their place in all of those. The rule I'd actually give is: a tag should CONFIRM who's speaking, not be the reason you know.
And the action-beat alternative is worth naming precisely because it feels like a solution. Replacing "he said" with him setting down a glass gives you texture and it reads better. It also attributes the line exactly as mechanically as the tag did. If you ran the audit and the beats were what got you through, you haven't fixed anything, you've re-upholstered it.
On a full manuscript
This is a scene-level test and it stays manual, because the read-aloud instinct is the instrument and nothing else can hold it.
What gets hard is knowing WHERE to run it. Nobody strips tags out of ninety scenes.
FirstReader helps with that part of the problem. Dialogue is one of the craft dimensions it reports on, so its scene-level reports will show you where your dialogue-heavy scenes are and how they're behaving structurally, which is enough to pick your candidates instead of guessing at them.
It won't tell you your characters sound alike. That judgement lives in your ear, and it needs you to know who these people are supposed to be. What it'll do is stop you running a ten-minute test ninety times to find the four scenes that needed it.
The short version
Strip the tags. Read it aloud once, fast. Mark what you can't place.
If the marks cluster, don't fix the tags. Find which of the four dimensions your two characters are sharing, change one of them in one character, and run it again.
The tags were never the problem. They were just the thing holding the scene together while you weren't looking.