Boilerplate, a fact sheet you can print without checking, logo files, and five story angles with sources attached. If you need something that is not here, email support@humanflow.pro and we will answer the same week.
Background a journalist usually wants first: who runs this, how we source and test including why we publish no bypass rates, and the corrections log, which is the fastest way to judge whether the rest of it holds up.
Two lines, ready to paste.
HumanFlow is an AI writing tool that rewrites machine-drafted text to read naturally and scores it sentence by sentence for AI detection. It is built and operated by Lacewing Technologies LLC, an AI engineering studio registered in the State of Wyoming, United States.
Unusually for the category, HumanFlow publishes no bypass rates and makes no claim to beat AI detectors, on the grounds that no vendor can measure that honestly.
Every figure below is read from the product's own configuration, so it cannot drift from what the site actually offers.
| Product | HumanFlow — AI Humanizer & Detector |
| Company | Lacewing Technologies LLC |
| Entity | Single-member limited liability company, State of Wyoming, United States |
| Formed | 7 August 2026 (filing 2026-002051166) |
| Registered office | 30 N Gould St, Ste N, Sheridan, WY 82801, United States |
| Parent company site | lacewingtechllc.com |
| Website | humanflow.pro |
| Free plan — humanizer | 2,000 words / month |
| Free plan — detection | 10,000 words / month |
| Free plan — single scan | 1,500 words |
| Paid plans | Writer $18/mo, Studio $29/mo, Team $59/mo |
| Annual discount | 40% |
| Tones | 3 — natural, casual and professional |
| Languages | English only |
| Platforms | Web. iOS in development; no browser extension |
| API | Private beta — no public endpoint |
| Card payments | Not live yet |
| Competitors documented | 17, each with dated figures |
Not published yet: founding date, founder name and bio, headshot, and employee count. We have left these blank rather than approximate them, because a media kit is the one place a guess gets printed as fact. Email us and we will send them.
Awaiting details
Name, role, two-line bio, headshot and LinkedIn will appear here. Deliberately not filled with a placeholder person — see the note above.
Vector, scales to any size. PNG variants and a dark-background lockup are not packaged yet — ask and we will send them.
Please keep the wordmark as one word, lowercase, with "flow" in the accent colour. Do not recolour the mark, stretch it, or set it on a background that drops the contrast below legible.
Five things we can talk about with evidence attached. The first three are about the category rather than about us, and we will say the same things on the record.
Liang et al., published in Patterns (2023), ran 91 TOEFL essays written by real people through seven detectors. They flagged an average of 61.22% as AI-generated; 89 of the 91 were flagged by at least one tool. The same detectors read US 8th-grade essays almost perfectly.
Turnitin's stated false positive rate applies only to documents where at least 20% of the qualifying text is flagged. Below that it shows an asterisk instead of a score, because its own testing found more false positives there. Most coverage drops the condition.
Turnitin published research reporting near-identical false positive rates for English language learners and native speakers above 300 words. It is a real result and it was paid for by Turnitin — a disclosure that rarely survives into secondary coverage.
Competitors advertise figures from 99.5% to 99.9% success against named detectors, and one sells a guarantee against all of them. None ships a corpus, a date, a detector version or a repeatable method. We publish no such number and explain why.
Our benchmark protocol is public with no data attached: corpus design, sample sizes, detector selection, scoring rules and a commitment to release raw per-sample CSVs. First results are scheduled for Q4 2026 and publish unchanged if our own tool performs badly.
support@humanflow.pro — press enquiries get answered within a week, usually sooner. We will go on the record about detector accuracy, false positives and our own limitations, including the ones that are commercially inconvenient.
What we will not do is give you a bypass rate, because we do not have one worth printing — here is why.