1. Classify the document context
Identify whether the text is an essay, research paper, article, business report, application, or internal document before interpreting AI-writing risk.
Methodology
WriteGo is built around writing-integrity workflows: score the document, show the evidence, reduce false positives, and keep humans in control of high-stakes decisions.
Benchmark
WriteGo has published the evaluation protocol described below, but has not yet published an auditable performance report that supports a precise headline accuracy rate. Until the dataset definition, versioned test procedure, subgroup results, and reproducible materials are public, detector scores should be treated as probabilistic review signals.
Methodology content version 2026-07-15 · reviewed by WriteGo product and engineering · public protocol draft with auditable performance results pending.
No benchmark average can prove how an individual document was authored. Text length, language, editing, translation, genre, model changes, and mixed authorship can all change the signal, so high-stakes action requires independent context and human review.
Update note: This revision separates the public evaluation protocol from unpublished performance claims. Numeric results will be added only with dataset definitions, a versioned test procedure, subgroup results, and reproducible supporting materials.
Detector output is review evidence, not a final judgment. Reports show sentence-level signals, confidence ranges, and reviewer notes so teams can make defensible decisions.
The public protocol calls for versioned model families, mixed-authorship samples, edited drafts, human writing, multilingual text, and domain-specific prose. Performance results remain pending until the supporting materials are auditable.
The product favors transparent risk bands over absolute accusations. Strong workflows combine AI-likelihood signals with source context, writing history, drafts, and policy.
Submission handling, access, and retention must follow the published privacy policy, account settings, and configured workflow. Reviewers should avoid submitting sensitive text without checking the applicable controls.
Review workflow
The methodology separates triage from judgment. Reviewers should understand what was scanned, why a passage was flagged, which false-positive patterns apply, and what policy-based action is appropriate.
Identify whether the text is an essay, research paper, article, business report, application, or internal document before interpreting AI-writing risk.
Use the document-level score for triage, then inspect the sentence or paragraph evidence that caused the risk band.
Check whether the text is short, translated, templated, ESL, heavily edited, or citation-heavy before escalating a result.
Record the prompt, assignment, source material, drafts, reviewer notes, and policy threshold that shaped the final decision.
Accept the text, request revision, ask for disclosure, escalate for review, or dismiss the signal when supporting evidence is weak.
Limitations
Trustworthy AI detection is transparent about uncertainty. WriteGo avoids framing a single score as a final verdict.
Not yet. WriteGo has published its evaluation approach, but a versioned benchmark report with dataset definitions, subgroup results, and reproducible supporting materials is still pending. We therefore do not present a precise accuracy rate as a verified public claim. Detector output remains probabilistic review evidence, not proof of individual authorship.
No. WriteGo treats AI-detection output as review evidence, not proof. High-stakes decisions should include drafts, sources, policy, and human judgment.
Methodology explains how scores are calibrated, what evidence reviewers see, how false positives are handled, and when a result should be escalated.
Teams should inspect flagged passages, compare supporting context, check false-positive patterns, document reviewer notes, and choose a policy-based next action.