Our mission
Fair, clear writing for a multilingual world.
Most people who write English do not write it first. The tools available to them were not designed with them in mind, and the systems evaluating their writing misclassify it at disproportionate rates. Our mission is to address both, without removing the writer from the writing process.
THE PROBLEM AT SCALE
A multilingual world, monolingual tools
These figures are not a niche case. Second-language English is how most of the world writes.
FAIRNESS FIRST
Detection bias is a measurement problem, not a moral one
AI detectors are not malicious — they are classifiers. But the features they reward overlap almost exactly with the markers of careful, learned English, which makes a false flag close to inevitable for the writers working hardest at the language. Writence is designed with this asymmetry in mind.
Read the bias evidence →- AI detectors lean on signals like low perplexity and predictable phrasing — exactly the patterns careful second-language writing produces.
- Independent studies (Stanford 2023, Maryland 2025, and others) have found that non-native English is flagged as “AI” far more often than native English.
- Real-world false-positive rates run well above the figures detector vendors publish.
- The cost of a false flag — a rejected paper, an academic-integrity hearing — falls hardest on the writers with the least benefit of the doubt.
KEEP THE HUMAN
What “keep the human” means in practice
The claim is easy to state. What follows shows how it appears in the product, line by line.
Assist, do not replace
Writence’s tools sharpen sentences and help move a stalled draft forward. They do not write on the writer’s behalf, and they do not hand over text to paste in. The thinking remains the writer’s own.
Make the writing process verifiable
When a writer opts in, the writing process is recorded as a cryptographically signed, append-only chain. Anyone can verify the chain’s integrity — no account and no trust required.
Refuse to train on writer content
Writence does not train models on writer content. Privacy is the default, not a setting that must be located and enabled.
The Authorship Certificate documents how a piece was written. It does not, and cannot, prove legal innocence — it provides the writer with verifiable evidence of the writing process.
A clearer standard
for human writing.
The products built from this mission are the clearest illustration of what it means in practice.
Evaluating Perplexity Bias in Academic Detection
Quantitative assessments demonstrate that standardized models consistently mistake concise, rule-based second-language phrasing for synthetic generation.
The protocol generates a cryptographic signature confirming the record’s integrity, without exposing unpublished findings.