AI Passive Voice Fixer
Revise passive sentences where active voice makes the meaning clearer. Keep useful passive constructions and return the complete edited text. Try the same tool available in your dashboard, then carry your input, options and result into your account.
Correct the document while preserving its voice, paragraph order and meaning.
Add at least 3 words to document.
The complete document comes back with grammar, spelling and punctuation fixed in place.
What passive voice actually is, and what keeps getting mistaken for it.
Most people who have been told to avoid the passive cannot reliably identify it, which is a reasonable outcome given how it is usually taught. It is worth ten seconds to fix.
The passive puts the receiver of an action in the subject position. A form of “to be” plus a past participle. “Sam wrote the report” is active. “The report was written by Sam” is passive, and so is “the report was written”, where the actor has disappeared entirely.
“Was” is not the passive. This is the error behind most false flags, by hand and by software. “The meeting was long” contains no action at all, so it cannot be passive. “She was running” is active — she is doing the running. Any checker that highlights those is pattern-matching on a word rather than parsing a sentence.
Some sentences are genuinely ambiguous. “The door was closed” might describe a state or report an action by an unnamed person. No tool resolves that without knowing what you meant, which is why what gets flagged is worth reading rather than accepting.
And the passive is not the same as vagueness. They get conflated constantly. “It should be noted that consideration was given to the matter” is bad writing because it says nothing, and fixing only the passive gives you “we considered the matter”, which still says nothing. The passive was the symptom.
Five cases where the passive is the right choice.
A tool that flags every instance is teaching a rule that good writers break constantly and deliberately.
- When nobody knows who did it“The server was compromised on Tuesday.” If you knew who compromised it you would say so, and until you do the passive is the only honest construction available. Rewriting this to invent an actor is worse than leaving it passive.
- When the actor does not matter“The samples were refrigerated for 48 hours.” Nobody cares which technician opened the fridge door, and naming one would imply it mattered. This is why scientific method sections are written in the passive and why style guides for those disciplines require it.
- When it keeps the subject where the reader expects itEnglish sentences flow best when known information comes first. If a paragraph is about your application, “the application was rejected” keeps the reader oriented, while “the committee rejected the application” swings the focus to a committee nobody was thinking about. The passive here is doing real work for the reader.
- When naming the actor would be an accusation“A mistake was made in the invoice.” Sometimes evasive, and sometimes exactly right — the error needs correcting and assigning blame in a customer email would help nobody. The passive lets you state a fact without making it about a person. Judgement, not a rule.
- When you want to end on the actorThe end of a sentence is its most emphatic position. “The proposal was approved unanimously by the board” lands differently from “the board unanimously approved the proposal”, because it puts the board last. If that is the emphasis you want, the passive is the tool that produces it.
Under 15–20%, and why the target is not zero.
The conventional guidance across the major style guides puts passive constructions below fifteen to twenty per cent of sentences. The part worth dwelling on is that the figure is deliberately not none.
A document with zero passive constructions has usually been over-edited. Not by a writer making judgements sentence by sentence, but by somebody applying a rule to a highlighted list. The result is prose where the actor is named in places nobody wanted one, and where the emphasis has quietly moved in sentences that were fine.
The rate matters less than which instances. Five passive constructions doing real work — reporting a fact whose actor is unknown, keeping a paragraph's subject in place — are worth more than a document at three per cent where the surviving instances are all evasions. Read the flagged sentences.
And the genre changes the target entirely. A scientific methods section is conventionally passive throughout, and many journals expect it. Legal drafting uses it for the same reason. Applying general-purpose writing advice to either produces prose that is wrong for its context, which is a worse outcome than a high percentage.
The one construction worth removing nearly every time is the actorless evasion: “mistakes were made”, “it has been decided”, “it should be noted”. There, the passive is hiding a person who exists and could be named, and that is the thing the rule was originally about.
Passive voice in ChatGPT, Claude, Gemini, DeepSeek, Grok, LLaMA and Perplexity text.
Models use less passive voice than people do, having absorbed the same advice. Where they use it, the reasons differ.
ChatGPT passive voice fixer
Uses less passive voice than most human writing, and it is not always an improvement.
ChatGPT has absorbed the advice to prefer the active voice and applies it fairly consistently, so its drafts often score better on passive rate than a person's would. What it produces instead is a different uniformity: subject-verb-object, sentence after sentence, which is one of the reasons its prose reads as flat.
Where it does use the passive, it tends to be the evasive kind — “it should be noted”, “it is widely considered” — which is the one construction genuinely worth removing every time, because there is no actor and no source behind it.
Claude passive voice fixer
The highest passive rate of the major assistants, and often correctly so.
Claude hedges and qualifies, which pulls it toward constructions that avoid naming an actor. A lot of that is appropriate — it is writing carefully about things where the actor is genuinely unknown — and a checker set to flag every instance will mark up prose that is doing its job.
The instances worth changing are the ones where Claude knows the actor and declines to name it. Those read as diffident rather than as careful, and putting the subject back is usually a one-word edit.
Gemini passive voice fixer
Passive rate is misleading on its output for a structural reason.
Bulleted text produces sentence fragments, and fragments frequently have no verb at all, which makes any passive-rate percentage a proportion of a much smaller denominator than you would expect. A Gemini draft can report a low passive rate while its actual prose paragraphs are heavily passive.
Convert the bullets to sentences before measuring, otherwise the number is describing the formatting.
DeepSeek passive voice fixer
Heavily passive, and this is the case where that is most defensible.
Its technical register produces exactly the constructions scientific writing conventionally uses: procedures described without an actor, results reported without an experimenter. Method sections are supposed to read this way, and many journals still expect it.
So a high passive rate on DeepSeek output is worth checking against the document type before acting on it. In a methods section it may be correct; in a customer-facing explanation it is not.
Grok passive voice fixer
The lowest passive rate, as a side effect of register.
Conversational writing is naturally active, because people talking about things name who did them. Grok drafts therefore score well without anyone trying, which is a genuine consequence of the register rather than an artefact.
The editing usually needed on Grok output is elsewhere — raising the register for professional contexts — and doing that tends to introduce passive constructions rather than remove them.
LLaMA passive voice fixer
Inconsistent within a document, which is the pattern to look for.
LLaMA is weights rather than a product, so output depends on the deployment. Passive rate can differ substantially between sections of one document, which is unusual in human writing and visible if you measure by section.
Smaller deployments also produce genuinely mangled constructions — a passive built on the wrong auxiliary, or a sentence that starts passive and finishes active. Those are errors rather than style choices.
Perplexity passive voice fixer
Passive constructions are structural here, because of how it attributes.
Sentences built around citations use the passive to foreground findings over researchers: “it was found that”, “the effect was observed”. With the reference attached that is standard academic convention. With the reference stripped out, as usually happens before pasting, the same sentence reads as a claim with nobody behind it.
So the fix depends on whether the citation is coming back. If it is, keep the construction. If it is not, name who found it or drop the claim.
Passive voice questions.
What is passive voice, exactly?
A construction where the thing receiving the action becomes the subject: a form of “to be” plus a past participle. “The report was written by Sam” is passive; “Sam wrote the report” is active. The most common mistake is assuming any sentence containing “was” is passive — “the meeting was long” is not passive, it has no action in it at all.
Is passive voice wrong?
No, and this is the most useful thing on the page. It is frequently the correct choice: when nobody knows who acted, when the actor is irrelevant, when it keeps the sentence's subject where the reader expects it, and when naming a person would turn a fact into an accusation. A tool that flags every instance is teaching a rule that good writers break constantly.
How much passive voice is acceptable?
The conventional guidance from the major style guides is under 15–20% of sentences, and the important part of that figure is that it is not zero. A document with no passive constructions at all has usually been edited by someone applying a rule rather than reading the sentences.
Why do checkers flag things that are not passive?
Because most of them look for “to be” plus a word ending in -ed, which catches a lot of innocent bystanders. “The door was closed” is ambiguous — passive if someone closed it, a plain description if it was simply shut. Automated detection cannot always tell, so read what is flagged rather than accepting it.
Should scientific writing avoid the passive?
No. Method sections are conventionally passive because the actor genuinely does not matter — nobody cares which technician handled the samples, and naming one implies it was significant. Many journals still expect this. Applying general writing advice to a methods section produces prose that is wrong for its genre.
Does AI text use more or less passive voice?
Less, generally, since the models absorbed the same advice everyone else did. ChatGPT and Grok both score well. Claude is the exception among the major assistants and its passive constructions are often appropriate ones. DeepSeek is heavily passive because its register is technical, which is usually defensible.
What is the difference between passive voice and weak writing?
They get conflated constantly. Weak writing usually means vague nouns, hedging and no clear claim — “it should be noted that consideration was given” is bad because it says nothing, and the passive is incidental. Fixing the passive there leaves you with “we considered it”, which is still saying nothing.
Will fixing passive voice lower an AI detection score?
No, and it may raise it. Converting everything to subject-verb-object makes prose more uniform, and uniformity is what detectors measure. Editing for clarity and editing to move a score frequently pull in opposite directions.
Can it fix passive voice in ChatGPT, Claude and Gemini text?
Yes, and the sections above cover each. Worth knowing that a passive-rate percentage on Gemini output is unreliable, because its bulleted fragments often contain no verb and shrink the denominator the percentage is calculated against.
Is my text stored?
The document is attached to your account so you can return to it, and you can delete it whenever you like. It is not used to train anything.