About us

Honest detection. Rewriting that reads well.

We make Blurwrite to help you examine writing and revise it with care. We show uncertainty in detector readings, check rewritten facts against the source and judge revisions by how they read.

Blurwrite is for people who need to judge a piece of writing fairly and improve it without changing what it says. Our suite brings AI detection and AI humanizing together with paraphrasing, grammar, summarizing and counting tools. A detector should report its reading with its uncertainty. It should also publish its own error rate, so you can see where a result deserves caution rather than confidence.

We believe a rewriter should be judged by whether its output reads well and preserves what you meant. A lower detector reading is not a measure of good editing. We show the draft and rewrite readings side by side, but our aim is readable text in your voice. We keep detection and editing accountable to their different purposes.

Tools for careful readers and writers

We build for students revising drafts, teachers reviewing writing, editors checking copy and people writing in a second language. These tasks need more than a label or a rewritten paragraph. You need context for a reading and a way to examine changes. We want our tools to support that review, without turning a statistical signal into a judgment about a person.

We also want you to know what happens to your text. Text pasted into a tool is processed to produce the result. It is not stored, except in the history of a signed-in account. We do not train models on it, sell it or share it. The counters work in your browser, where your text stays. Free use needs no card.

How we read for AI writing

Our detector uses trained models to read how predictable each sentence is in its surrounding text. We interpret that score through a calibration curve fitted separately for each language, using human and machine writing, including writing by non-native speakers. We measure false flags alongside AI recognition and set the decision line with human writers in mind. We measure again when the model or decision threshold changes.

You get a figure, its range, a High, Medium or Low confidence rating and sentence marks pointing toward AI or human writing. We do not search for banned words, watermarks or database matches. Very short text gets no reading; short text gets Low confidence. Longer text gives more context. Human writing can be flagged. Rates vary by language and writing type; our English tests flag argumentative essays more often than encyclopedia prose.

How we rewrite for readability

Our humanizer rewrites AI-assisted drafts for readability and your own voice. It varies sentence length, removes filler and keeps meaning. Nothing invisible is added; the result is plain text. We check numbers, names and quotations against the original. We repair changes and show anything we could not repair beside the result. Standard keeps the register; the paid Academic, Professional and Casual modes change it. We judge the result by readability and faithfulness to the draft.

Our detector reads the draft and rewrite, with Before and After readings shown together. It is trained to recognize rewritten and paraphrased text, so the After reading often stays high. That is deliberate: rewriting should not make the detector less useful to readers who rely on it. We make no promise about other detectors' scores. Paid plans also create several rewrites and keep the most faithful one. The Pro plan adds a Custom mode that learns from your writing sample.

What we hold ourselves to

Publish the error rate
We publish our own measurements by language, including false flags on human writing. Results that fall outside our publication rule remain visible.
Evidence, not a verdict
We show the range and confidence with each reading. A signal pointing toward AI writing calls for review, not an automatic conclusion.
Include non-native writers
We include writing by non-native speakers in language calibration. A language goes live only when its human false-flag rate meets our publication rule.
Keep the facts intact
We check rewritten numbers, names and quotations against the draft. We repair changes and show unresolved differences beside the result.
Your text stays yours
We process your text for the result, without training on it, selling it or sharing it. Storage is limited to the history of a signed-in account.

Our measurements, with their limits

We publish measurements by language with dates, separating calibration results from held-out tests. We show false flags alongside AI recognition, including test results that do not meet our publication rule.

Read our accuracy results

The people checking our work

Our writers and editors test the tools and write guides on detection, academic writing and editing. They check article claims against sources, remove untraceable facts and link to our accuracy page rather than repeat figures.