We ran 16 tools over a shared document corpus and published every bypass rate — including the ones that failed. No paid placements. No affiliate rankings.
Rank follows bypass rate, then document count, then name. Expand any row for the full profile.
Showing 17 of 17 tools
Rank, Humanizer, Docs, Bypass rate, Tier
Tier
1
StealthGPT
stealthgpt.ai
150
47%
High
StealthGPT
Pricing
Paid onlyWeekly billing prominent — Pro ~$6.99/wk (12.5k words/wk), Max ~$19.99/wk; Enterprise ~$179.99/mo. $1 first-week promos. No substantive free tier.
Type
Humanizer
Summary. One of the longest-running humanizers (XYZ AI, Inc.), with an essay writer, detector, Chrome extension, mobile apps, and API. Claims 1.5M+ users.
Claims: Markets "we beat every AI detector" (GPTZero, Turnitin, Originality.ai, ZeroGPT, Copyleaks, Winston, Pangram); Lite vs Super models; four tone levels; money-back-if-detected guarantee; up to 20,000 words/request on Enterprise. — Vendor marketing (stealthgpt.ai), gathered 2026-09-26
Strengths/weaknesses. Mature ecosystem and API; historically strong independent results. Refund complaints on Trustpilot (~206 reviews, unverified); aggressive countdown-discount dark patterns.
Test results. The only tool approaching half of the corpus. Super-model deep rewriting plus years of detector-arms-race tuning plausibly explains the lead — but 47% still means it fails more often than it passes.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paid~$6/mo Basic (20k words), ~$10/mo Pro (60k), ~$20/mo Unlimited on annual billing; free tier is queued and gated.
Type
Humanizer
Summary. Humanizer from a self-described student/PhD-candidate team; bundles detector, API, rewrite history, file upload.
Claims: Free Basic mode vs paid "Autopilot Pro" marketed at "99% all detection bypass"; Balance/Enhanced/Private quality settings; Lock Keywords / Lock Citations; claims bypass of Turnitin, GPTZero, Copyleaks, ZeroGPT, Originality, QuillBot, Writer, Sapling. — Vendor marketing (aihumanize.io), gathered 2026-09-26
Strengths/weaknesses. Very cheap annual plans; useful keyword/citation locks. Free tier UX (queues, paywall overlays) widely disliked; team anonymous.
Test results. Basic (non-Autopilot) mode at 40% is the dataset's strongest default-mode result after StealthGPT. Its Autopilot variant scoring lower (35%) suggests the retry-until-pass loop optimizes for the vendor's own bundled detector checks, not our detector.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Paid onlyIncluded in paid plans from ~$10/mo (annual).
Type
Humanizer mode
Summary. The paid "Autopilot Pro" mode of AI Humanize: re-rewrites iteratively until its built-in checks pass, marketed at "99% all detection bypass."
Claims: Iterative auto-rewrite loop; same styles/locks as the base product; "private model" option marketed as updated for GPTZero & Turnitin. — Vendor marketing (aihumanize.io), gathered 2026-09-26
Strengths/weaknesses. Hands-off workflow; each loop costs time and words. Independent evidence for the 99% claim: none found (unverified).
Test results. The dataset's cleanest natural experiment: same corpus, same vendor, Autopilot (35%) ranks below its own basic mode (40%). The retry loop's target detectors evidently don't transfer to our detector — autopilot is tuned to beat the checks the vendor runs, not the checks we run.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. Humanizer + detector with Chrome extension, iOS/Android apps, REST API and MCP; claims 7.5M+ documents humanized.
Claims: Standard vs Enhanced editor toggle ("Enhanced updated for GPTZero and Pangram"); publishes its own human scores (Copyleaks 98%, Turnitin 97%, GPTZero 96%, Pangram 94%); 40+ languages. — Vendor marketing (writehuman.ai), gathered 2026-09-26
Strengths/weaknesses. Transparent plan table; strong app-store ratings (4.6/5, ~9k Google Play). Cybernews notes aggressive rewrites "can sound awkward or off-tone"; benchmark claims self-published.
Test results. Standard mode lands at 33% — solidly mid-pack, far below its own marketed 94–98% figures.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Test results. The point estimate (33%) ties WriteHuman, but n=32 makes it the dataset's least certain number. The small-sample badge is mandatory; no "top 4" marketing may quote this row without the n.
Only 32 documents were run for this tool, so its rate is much less certain than 150-doc rows. Treat its rank as provisional.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. All-in-one study/writing suite (humanizer, AI checker, plagiarism checker, math solver, quiz generator, ChatPDF); claims 30M+ users; operator not clearly disclosed (unverified).
Claims: Markets passing "even the strictest AI detectors like Turnitin, Originality.ai, GPTZero, ZeroGPT and Copyleaks"; Quick/Enhanced/Advanced modes; 50+ languages; ISO 27001/SOC 2/GDPR claims; API. — Vendor marketing (humbot.ai), gathered 2026-09-26
Strengths/weaknesses. Cheap bundle, broad tools, multilingual. Confusing credit currencies and inconsistent displayed prices; mid-tier output quality in independent tests (unverified).
Test results. 20% sits below every marketing superlative on its homepage: four out of five tested documents were still flagged. The suite model (dozens of tools) correlates with a humanizer that is a feature, not the product.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Paid onlyQuick Pass $12.99 one-time (24h); annual-billed Growth ~$8.25/mo (100 humanizations), Business ~$10/mo, Professional ~$18.33/mo; Lifetime $199 (list $300).
Type
Humanizer
Summary. Indie humanizer aimed at students/SEO; bundles detector, plagiarism checker, AI writer, style cloning, extension, API/MCP; company undisclosed (unverified).
Claims: Markets bypassing Turnitin, GPTZero, Copyleaks, Originality.ai, ZeroGPT, Pangram — with a "guarantee 100% pass rates" claim; free re-humanize if unsatisfied; SynthID detector/bypass; 2,000 words per run. — Vendor marketing (rephrasy.ai), gathered 2026-09-26
Strengths/weaknesses. Cheap lifetime deal; generous per-run cap. Sparse independent reviews; "4.8/5 by 6,000+ users" and university-logo social proof unverifiable.
Test results. The widest claim-to-measurement gap in the dataset: it advertises "100% pass rates" and measured 19% — four in five tested documents stayed flagged.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paidFree tier; $2 3-day trial; Unlimited ~$10.99/mo annual (~$19.99 monthly per site schema); business API.
Type
Humanizer mode
Summary. Writing suite (humanizer, detector, plagiarism checker, AI writer, flashcards) by Phrasly LLC, San Diego; claims 5M+ users; markets in-house models "trained on 500k+ real human articles."
Claims: Two humanizer models — Balanced (free) and Phrasly Ultra (paid, 5 modes, released Sept 2026); markets >99% pass rates on GPTZero, ZeroGPT, Turnitin, Originality.ai and 99.8% average Pangram human score (own benchmark); "watermark free"; 800 words/process trial, 5,000 paid. — Vendor marketing (phrasly.ai), gathered 2026-09-26
Strengths/weaknesses. Trustpilot 4.8/5 as displayed on own site (~3,500 reviews). Independent verification of pass-rate claims: none; its vendor benchmark scoring competitors poorly is self-published.
Test results. "Direct" is the dataset label for Phrasly's straight-through (non-Ultra) pipeline. At 15% it trails its own Balanced marketing by an order of magnitude — and the deeper-rewrite "Ultra" tier was not separately run in this cycle, so nothing here supports the Ultra claims either way.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paidFree core; Basic $12/mo (15k words), Pro $19/mo (50k), Advanced $39/mo (200k); 60% off yearly; API offered.
Type
Humanizer
Summary. Free web humanizer; aihumanizer.io now 301-redirects to SuperHumanizer; claims training on 15M human samples filtered to 1M; operator undisclosed (unverified).
Claims: Super Lite (single-step) vs Super Ultra (deeper, "quality-validated") modes; 1,200 words/run, unlimited free uses, no signup; built-in detector; markets bypassing Turnitin, GPTZero, Originality.ai with a self-reported 97% success rate (unverified). — Vendor marketing (aihumanizer.io), gathered 2026-09-26
Strengths/weaknesses. Genuinely free, no-signup access. Little independent coverage; rebrand churn (aihumanizer.io → superhumanizer.ai) makes version history murky — we tested the aihumanizer.io pipeline.
Test results. 14% against a self-reported 97% is the dataset's second-widest claim gap. The free tier's single-pass Lite rewrite is the likely tested path; either way, we found no evidence for the marketed success rate.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
CreditsOwn product line "from $24/mo" (site metadata); $1 1-day trial; credit-based (1 credit per 80 words on metered plans); Grow/Scale/Team tiers (exact per-tier prices client-rendered, unverified).
Type
Humanizer
Summary. Humanizer from AISEO, an established SEO-tools company; positioned around "structural refinement" (sentence rhythm/pacing) rather than synonym spinning.
Strengths/weaknesses. Real company with a real product suite; granular controls unusual in this niche. On-site testimonials only; independent reviews not found.
Test results. 12% — its "structural refinement" thesis produced a bottom-third result while its marketing claims 100% undetectable. A mature SEO vendor's humanizer add-on underperforming dedicated tools matches the Humbot pattern: the humanizer is a line item, not the mission.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. Phrasly's default Balanced model — the free-tier humanizer pipeline (same vendor as Phrasly — Direct).
Claims: Same suite claims; Balanced is the model free accounts run. — Vendor marketing (phrasly.ai), gathered 2026-09-26
Strengths/weaknesses. Polished UI and suite depth; pass-rate claims self-published only.
Test results. Balanced (11%) ran slightly below the Direct pipeline (15%) on the same corpus — both far under the vendor's >99% marketing. Two Phrasly rows in one dataset is deliberate: it shows how thin the differences are between a vendor's own "modes" once an independent corpus is applied.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. Long-standing paraphraser, now part of a writing suite under Learneo, Inc. (Course Hero's parent; exact structure unverified); claims 35M+ writers. Not a dedicated humanizer — though QuillBot now also sells a separate "AI Humanizer" tool and a "Humanize" paraphrase mode; we tested the classic paraphraser.
Claims: Free Standard/Fluency modes (125 words/paraphrase); premium adds Formal, Simple, Creative, Academic, Expand, Shorten, Humanize, Custom; synonym slider; 23 languages. Markets essentially no detection bypass — messaging is "sounds more natural," with an academic-integrity page. — Vendor marketing (quillbot.com), gathered 2026-09-26
Strengths/weaknesses. The most trustworthy brand in the set — it does not claim what it cannot deliver. As a bypass tool it was never the point: synonym-level paraphrase leaves deep statistical structure intact.
Test results. 9% is the score of a tool that never promised bypassing. It sits above six products that did — the dataset's quiet indictment of the "guaranteed undetectable" segment, and the honest control row: light paraphrasing moves detectors ~9% of the time.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Claims: "Bypass all AI Detectors guarantee" — free rewrite if flagged; counters GPTZero, Turnitin, Originality.ai, Copyleaks; Standard/College/PhD tones; Balanced/Enhanced modes; "Shield Guard"; watermark removal; 80 languages; publishes its own "AI Humanizer Benchmark 2026" (self-reported, unverified). — Vendor marketing (gpthuman.ai), gathered 2026-09-26
Strengths/weaknesses. Full toolchain and API; aggressive guarantee marketing. No independent verification of guarantees found; self-published benchmark only.
Test results. 3% — while selling a "bypass all AI detectors guarantee." Its self-published benchmark and our results disagree completely; we tested the product, not the marketing site.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paid300 free words/mo (site text; one reviewer reported 150 — discrepancy, unverified); Essential ~$12.99/mo (10k words), Pro ~$15.99/mo (30k), Unlimited ~$22.99/mo; auto-renewal, non-refundable except per guarantee terms.
Type
Humanizer
Summary. Student-focused paste-in humanizer/"AI bypasser" by Study Solutions A16 LLC; claims 1.6M+ student users; assets served from cdn.bypassai.ai, suggesting shared origin with BypassAI (unverified).
Claims: Markets beating GPTZero, Originality.ai, ZeroGPT, Turnitin, Winston, Content at Scale, Copyleaks, Writefull, NoGPT, AISEO with "99%+ human scores"; a "Turnitin Guarantee" (money-back); summarizer and autotyper (unverified); English only. — Vendor marketing (grubby.ai), gathered 2026-09-26
Strengths/weaknesses. Cheap, transparent pricing, simple UI. Trustpilot ~3.0/5 (63 reviews, unverified) with unstable-output, meaning-drift, and billing/refund complaints; consistency degrades on long documents; Turnitin added "AI bypasser detection" in Aug 2025, undermining the guarantee's premise.
Test results. 3% — forty-nine out of fifty tested documents stayed flagged, against a marketed "99%+ human scores" and a money-back Turnitin guarantee. The gap between guarantee language and measurement is the largest single-vendor contradiction in the dataset.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
FreeFree tool (1,500 words/use); "premium pricing options" mentioned by Originality.ai, no published plan table (unverified). Monetization funnels to the essay service.
Type
Humanizer
Summary. Free detect-and-humanize side tool of MyPerfectWords.com (Skyscrapers LLC), a paid human essay-writing service (since 2010, from ~$11/page); humanizer launched June 2024 (PR release).
Strengths/weaknesses. Genuinely free, combined detect+humanize. Originality.ai's own Jan 2026 test: output still 100% "Likely AI" on Originality.ai; ZeroGPT only dropped 98.65%→52.28%; reviewers documented badly garbled synonym swaps ("High blood pressure management" → "High Arteries Pressure Governance…").
Test results. 0 of 125. Rule-of-three upper bound ~2.4%. Consistent with Originality.ai's independent finding: naive synonym-level rewriting produces text that is simultaneously worse to read and still machine-flagged. The tool exists to funnel traffic to human writers — we measured the funnel's bait.
* Zero passes observed. With n = 125, the rule of three puts the true rate below 2.4% at 95% confidence — near zero, not proven zero.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Paid onlyConsumer app: free trial 250 words; annual plans from ~$5/mo (10k words) to ~$21.25/mo (50k). API pricing above.
Type
Humanizer API
Summary. Undetectable Inc. (Sheridan, WY): detector + humanizer suite claiming 24M+ users and "#1 Best AI Detector by Forbes" (marketing). We tested its public REST API — the integration that "powers other humanizers" (vendor claim).
Claims: API marketed as passing "every major detector"; models Basic/Stealth/Undetectable; money-back guarantee if output flags non-human; API: Starter $29/mo (72.5k words, then $0.40/1k), Growth $49/mo (245k words, then $0.20/1k), 3 req/sec default, check-for-AI billed at 1/10 humanize cost. — Vendor marketing (undetectable.ai), gathered 2026-09-26
Strengths/weaknesses. The suite's detector heritage is real; the humanizer API's independent reputation is thin (Sitejabber 2.0/5 from 2 reviews — too thin to generalize, unverified). Competitor benchmarks scoring it poorly exist but are vendor-published.
Test results. 0 of 80 documents passed. Rule-of-three puts the true rate below ~3.8% with 95% confidence. A zero at n=80 is a measurement, not an absence of measurement — the API path was fully exercised. Whatever the consumer app does, its programmatic output carried no detectable evasion signal in our testing.
Only 80 documents were run for this tool, so its rate is much less certain than 150-doc rows. Treat its rank as provisional.
* Zero passes observed. With n = 80, the rule of three puts the true rate below 3.8% at 95% confidence — near zero, not proven zero.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paidFree tier, no signup (~200-word visible cap; reviewers call free use "unlimited" — unverified); Day Pass $4.99/24h (25k words); Pro $6/mo annual (50k words/mo + Autopilot); Unlimited $15/mo annual (fair use 30k words/day).
Type
Humanizer
Summary. Free/freemium humanizer founded by Abdulla Abdurazzoqov (previously built and sold AIHumanize.com); small team, some ex-Hyundai AI research (self-reported); engine "Version 6," marketed as trained on 30M+ human samples; iOS/Android app and API.
Claims: Markets bypassing GPTZero, Originality.ai, Turnitin "and more"; Stealth vs Premium models; Standard or "Autopilot" (re-rewrites until it passes); built-in detector scoring; per-sentence alternatives; 25+ languages; claims no text storage. — Vendor marketing (texttohuman.com), gathered 2026-09-26
Strengths/weaknesses. Third-party reviews (345tool, marketspur, contentbuffer) report it effectively bypassed GPTZero while preserving meaning; praised as best free option; criticized for lacking advanced controls. Little independent Turnitin testing.
Test results. No run exists for this tool in the current cycle. It is listed for completeness; we will score it when it enters the corpus.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Paid onlyWeekly billing prominent — Pro ~$6.99/wk (12.5k words/wk), Max ~$19.99/wk; Enterprise ~$179.99/mo. $1 first-week promos. No substantive free tier.
Type
Humanizer
Summary. One of the longest-running humanizers (XYZ AI, Inc.), with an essay writer, detector, Chrome extension, mobile apps, and API. Claims 1.5M+ users.
Claims: Markets "we beat every AI detector" (GPTZero, Turnitin, Originality.ai, ZeroGPT, Copyleaks, Winston, Pangram); Lite vs Super models; four tone levels; money-back-if-detected guarantee; up to 20,000 words/request on Enterprise. — Vendor marketing (stealthgpt.ai), gathered 2026-09-26
Strengths/weaknesses. Mature ecosystem and API; historically strong independent results. Refund complaints on Trustpilot (~206 reviews, unverified); aggressive countdown-discount dark patterns.
Test results. The only tool approaching half of the corpus. Super-model deep rewriting plus years of detector-arms-race tuning plausibly explains the lead — but 47% still means it fails more often than it passes.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paid~$6/mo Basic (20k words), ~$10/mo Pro (60k), ~$20/mo Unlimited on annual billing; free tier is queued and gated.
Type
Humanizer
Summary. Humanizer from a self-described student/PhD-candidate team; bundles detector, API, rewrite history, file upload.
Claims: Free Basic mode vs paid "Autopilot Pro" marketed at "99% all detection bypass"; Balance/Enhanced/Private quality settings; Lock Keywords / Lock Citations; claims bypass of Turnitin, GPTZero, Copyleaks, ZeroGPT, Originality, QuillBot, Writer, Sapling. — Vendor marketing (aihumanize.io), gathered 2026-09-26
Strengths/weaknesses. Very cheap annual plans; useful keyword/citation locks. Free tier UX (queues, paywall overlays) widely disliked; team anonymous.
Test results. Basic (non-Autopilot) mode at 40% is the dataset's strongest default-mode result after StealthGPT. Its Autopilot variant scoring lower (35%) suggests the retry-until-pass loop optimizes for the vendor's own bundled detector checks, not our detector.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Paid onlyIncluded in paid plans from ~$10/mo (annual).
Type
Humanizer mode
Summary. The paid "Autopilot Pro" mode of AI Humanize: re-rewrites iteratively until its built-in checks pass, marketed at "99% all detection bypass."
Claims: Iterative auto-rewrite loop; same styles/locks as the base product; "private model" option marketed as updated for GPTZero & Turnitin. — Vendor marketing (aihumanize.io), gathered 2026-09-26
Strengths/weaknesses. Hands-off workflow; each loop costs time and words. Independent evidence for the 99% claim: none found (unverified).
Test results. The dataset's cleanest natural experiment: same corpus, same vendor, Autopilot (35%) ranks below its own basic mode (40%). The retry loop's target detectors evidently don't transfer to our detector — autopilot is tuned to beat the checks the vendor runs, not the checks we run.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. Humanizer + detector with Chrome extension, iOS/Android apps, REST API and MCP; claims 7.5M+ documents humanized.
Claims: Standard vs Enhanced editor toggle ("Enhanced updated for GPTZero and Pangram"); publishes its own human scores (Copyleaks 98%, Turnitin 97%, GPTZero 96%, Pangram 94%); 40+ languages. — Vendor marketing (writehuman.ai), gathered 2026-09-26
Strengths/weaknesses. Transparent plan table; strong app-store ratings (4.6/5, ~9k Google Play). Cybernews notes aggressive rewrites "can sound awkward or off-tone"; benchmark claims self-published.
Test results. Standard mode lands at 33% — solidly mid-pack, far below its own marketed 94–98% figures.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Test results. The point estimate (33%) ties WriteHuman, but n=32 makes it the dataset's least certain number. The small-sample badge is mandatory; no "top 4" marketing may quote this row without the n.
Only 32 documents were run for this tool, so its rate is much less certain than 150-doc rows. Treat its rank as provisional.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. All-in-one study/writing suite (humanizer, AI checker, plagiarism checker, math solver, quiz generator, ChatPDF); claims 30M+ users; operator not clearly disclosed (unverified).
Claims: Markets passing "even the strictest AI detectors like Turnitin, Originality.ai, GPTZero, ZeroGPT and Copyleaks"; Quick/Enhanced/Advanced modes; 50+ languages; ISO 27001/SOC 2/GDPR claims; API. — Vendor marketing (humbot.ai), gathered 2026-09-26
Strengths/weaknesses. Cheap bundle, broad tools, multilingual. Confusing credit currencies and inconsistent displayed prices; mid-tier output quality in independent tests (unverified).
Test results. 20% sits below every marketing superlative on its homepage: four out of five tested documents were still flagged. The suite model (dozens of tools) correlates with a humanizer that is a feature, not the product.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Paid onlyQuick Pass $12.99 one-time (24h); annual-billed Growth ~$8.25/mo (100 humanizations), Business ~$10/mo, Professional ~$18.33/mo; Lifetime $199 (list $300).
Type
Humanizer
Summary. Indie humanizer aimed at students/SEO; bundles detector, plagiarism checker, AI writer, style cloning, extension, API/MCP; company undisclosed (unverified).
Claims: Markets bypassing Turnitin, GPTZero, Copyleaks, Originality.ai, ZeroGPT, Pangram — with a "guarantee 100% pass rates" claim; free re-humanize if unsatisfied; SynthID detector/bypass; 2,000 words per run. — Vendor marketing (rephrasy.ai), gathered 2026-09-26
Strengths/weaknesses. Cheap lifetime deal; generous per-run cap. Sparse independent reviews; "4.8/5 by 6,000+ users" and university-logo social proof unverifiable.
Test results. The widest claim-to-measurement gap in the dataset: it advertises "100% pass rates" and measured 19% — four in five tested documents stayed flagged.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paidFree tier; $2 3-day trial; Unlimited ~$10.99/mo annual (~$19.99 monthly per site schema); business API.
Type
Humanizer mode
Summary. Writing suite (humanizer, detector, plagiarism checker, AI writer, flashcards) by Phrasly LLC, San Diego; claims 5M+ users; markets in-house models "trained on 500k+ real human articles."
Claims: Two humanizer models — Balanced (free) and Phrasly Ultra (paid, 5 modes, released Sept 2026); markets >99% pass rates on GPTZero, ZeroGPT, Turnitin, Originality.ai and 99.8% average Pangram human score (own benchmark); "watermark free"; 800 words/process trial, 5,000 paid. — Vendor marketing (phrasly.ai), gathered 2026-09-26
Strengths/weaknesses. Trustpilot 4.8/5 as displayed on own site (~3,500 reviews). Independent verification of pass-rate claims: none; its vendor benchmark scoring competitors poorly is self-published.
Test results. "Direct" is the dataset label for Phrasly's straight-through (non-Ultra) pipeline. At 15% it trails its own Balanced marketing by an order of magnitude — and the deeper-rewrite "Ultra" tier was not separately run in this cycle, so nothing here supports the Ultra claims either way.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paidFree core; Basic $12/mo (15k words), Pro $19/mo (50k), Advanced $39/mo (200k); 60% off yearly; API offered.
Type
Humanizer
Summary. Free web humanizer; aihumanizer.io now 301-redirects to SuperHumanizer; claims training on 15M human samples filtered to 1M; operator undisclosed (unverified).
Claims: Super Lite (single-step) vs Super Ultra (deeper, "quality-validated") modes; 1,200 words/run, unlimited free uses, no signup; built-in detector; markets bypassing Turnitin, GPTZero, Originality.ai with a self-reported 97% success rate (unverified). — Vendor marketing (aihumanizer.io), gathered 2026-09-26
Strengths/weaknesses. Genuinely free, no-signup access. Little independent coverage; rebrand churn (aihumanizer.io → superhumanizer.ai) makes version history murky — we tested the aihumanizer.io pipeline.
Test results. 14% against a self-reported 97% is the dataset's second-widest claim gap. The free tier's single-pass Lite rewrite is the likely tested path; either way, we found no evidence for the marketed success rate.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
CreditsOwn product line "from $24/mo" (site metadata); $1 1-day trial; credit-based (1 credit per 80 words on metered plans); Grow/Scale/Team tiers (exact per-tier prices client-rendered, unverified).
Type
Humanizer
Summary. Humanizer from AISEO, an established SEO-tools company; positioned around "structural refinement" (sentence rhythm/pacing) rather than synonym spinning.
Strengths/weaknesses. Real company with a real product suite; granular controls unusual in this niche. On-site testimonials only; independent reviews not found.
Test results. 12% — its "structural refinement" thesis produced a bottom-third result while its marketing claims 100% undetectable. A mature SEO vendor's humanizer add-on underperforming dedicated tools matches the Humbot pattern: the humanizer is a line item, not the mission.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. Phrasly's default Balanced model — the free-tier humanizer pipeline (same vendor as Phrasly — Direct).
Claims: Same suite claims; Balanced is the model free accounts run. — Vendor marketing (phrasly.ai), gathered 2026-09-26
Strengths/weaknesses. Polished UI and suite depth; pass-rate claims self-published only.
Test results. Balanced (11%) ran slightly below the Direct pipeline (15%) on the same corpus — both far under the vendor's >99% marketing. Two Phrasly rows in one dataset is deliberate: it shows how thin the differences are between a vendor's own "modes" once an independent corpus is applied.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Summary. Long-standing paraphraser, now part of a writing suite under Learneo, Inc. (Course Hero's parent; exact structure unverified); claims 35M+ writers. Not a dedicated humanizer — though QuillBot now also sells a separate "AI Humanizer" tool and a "Humanize" paraphrase mode; we tested the classic paraphraser.
Claims: Free Standard/Fluency modes (125 words/paraphrase); premium adds Formal, Simple, Creative, Academic, Expand, Shorten, Humanize, Custom; synonym slider; 23 languages. Markets essentially no detection bypass — messaging is "sounds more natural," with an academic-integrity page. — Vendor marketing (quillbot.com), gathered 2026-09-26
Strengths/weaknesses. The most trustworthy brand in the set — it does not claim what it cannot deliver. As a bypass tool it was never the point: synonym-level paraphrase leaves deep statistical structure intact.
Test results. 9% is the score of a tool that never promised bypassing. It sits above six products that did — the dataset's quiet indictment of the "guaranteed undetectable" segment, and the honest control row: light paraphrasing moves detectors ~9% of the time.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Claims: "Bypass all AI Detectors guarantee" — free rewrite if flagged; counters GPTZero, Turnitin, Originality.ai, Copyleaks; Standard/College/PhD tones; Balanced/Enhanced modes; "Shield Guard"; watermark removal; 80 languages; publishes its own "AI Humanizer Benchmark 2026" (self-reported, unverified). — Vendor marketing (gpthuman.ai), gathered 2026-09-26
Strengths/weaknesses. Full toolchain and API; aggressive guarantee marketing. No independent verification of guarantees found; self-published benchmark only.
Test results. 3% — while selling a "bypass all AI detectors guarantee." Its self-published benchmark and our results disagree completely; we tested the product, not the marketing site.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paid300 free words/mo (site text; one reviewer reported 150 — discrepancy, unverified); Essential ~$12.99/mo (10k words), Pro ~$15.99/mo (30k), Unlimited ~$22.99/mo; auto-renewal, non-refundable except per guarantee terms.
Type
Humanizer
Summary. Student-focused paste-in humanizer/"AI bypasser" by Study Solutions A16 LLC; claims 1.6M+ student users; assets served from cdn.bypassai.ai, suggesting shared origin with BypassAI (unverified).
Claims: Markets beating GPTZero, Originality.ai, ZeroGPT, Turnitin, Winston, Content at Scale, Copyleaks, Writefull, NoGPT, AISEO with "99%+ human scores"; a "Turnitin Guarantee" (money-back); summarizer and autotyper (unverified); English only. — Vendor marketing (grubby.ai), gathered 2026-09-26
Strengths/weaknesses. Cheap, transparent pricing, simple UI. Trustpilot ~3.0/5 (63 reviews, unverified) with unstable-output, meaning-drift, and billing/refund complaints; consistency degrades on long documents; Turnitin added "AI bypasser detection" in Aug 2025, undermining the guarantee's premise.
Test results. 3% — forty-nine out of fifty tested documents stayed flagged, against a marketed "99%+ human scores" and a money-back Turnitin guarantee. The gap between guarantee language and measurement is the largest single-vendor contradiction in the dataset.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
FreeFree tool (1,500 words/use); "premium pricing options" mentioned by Originality.ai, no published plan table (unverified). Monetization funnels to the essay service.
Type
Humanizer
Summary. Free detect-and-humanize side tool of MyPerfectWords.com (Skyscrapers LLC), a paid human essay-writing service (since 2010, from ~$11/page); humanizer launched June 2024 (PR release).
Strengths/weaknesses. Genuinely free, combined detect+humanize. Originality.ai's own Jan 2026 test: output still 100% "Likely AI" on Originality.ai; ZeroGPT only dropped 98.65%→52.28%; reviewers documented badly garbled synonym swaps ("High blood pressure management" → "High Arteries Pressure Governance…").
Test results. 0 of 125. Rule-of-three upper bound ~2.4%. Consistent with Originality.ai's independent finding: naive synonym-level rewriting produces text that is simultaneously worse to read and still machine-flagged. The tool exists to funnel traffic to human writers — we measured the funnel's bait.
* Zero passes observed. With n = 125, the rule of three puts the true rate below 2.4% at 95% confidence — near zero, not proven zero.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Paid onlyConsumer app: free trial 250 words; annual plans from ~$5/mo (10k words) to ~$21.25/mo (50k). API pricing above.
Type
Humanizer API
Summary. Undetectable Inc. (Sheridan, WY): detector + humanizer suite claiming 24M+ users and "#1 Best AI Detector by Forbes" (marketing). We tested its public REST API — the integration that "powers other humanizers" (vendor claim).
Claims: API marketed as passing "every major detector"; models Basic/Stealth/Undetectable; money-back guarantee if output flags non-human; API: Starter $29/mo (72.5k words, then $0.40/1k), Growth $49/mo (245k words, then $0.20/1k), 3 req/sec default, check-for-AI billed at 1/10 humanize cost. — Vendor marketing (undetectable.ai), gathered 2026-09-26
Strengths/weaknesses. The suite's detector heritage is real; the humanizer API's independent reputation is thin (Sitejabber 2.0/5 from 2 reviews — too thin to generalize, unverified). Competitor benchmarks scoring it poorly exist but are vendor-published.
Test results. 0 of 80 documents passed. Rule-of-three puts the true rate below ~3.8% with 95% confidence. A zero at n=80 is a measurement, not an absence of measurement — the API path was fully exercised. Whatever the consumer app does, its programmatic output carried no detectable evasion signal in our testing.
Only 80 documents were run for this tool, so its rate is much less certain than 150-doc rows. Treat its rank as provisional.
* Zero passes observed. With n = 80, the rule of three puts the true rate below 3.8% at 95% confidence — near zero, not proven zero.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Free tier + paidFree tier, no signup (~200-word visible cap; reviewers call free use "unlimited" — unverified); Day Pass $4.99/24h (25k words); Pro $6/mo annual (50k words/mo + Autopilot); Unlimited $15/mo annual (fair use 30k words/day).
Type
Humanizer
Summary. Free/freemium humanizer founded by Abdulla Abdurazzoqov (previously built and sold AIHumanize.com); small team, some ex-Hyundai AI research (self-reported); engine "Version 6," marketed as trained on 30M+ human samples; iOS/Android app and API.
Claims: Markets bypassing GPTZero, Originality.ai, Turnitin "and more"; Stealth vs Premium models; Standard or "Autopilot" (re-rewrites until it passes); built-in detector scoring; per-sentence alternatives; 25+ languages; claims no text storage. — Vendor marketing (texttohuman.com), gathered 2026-09-26
Strengths/weaknesses. Third-party reviews (345tool, marketspur, contentbuffer) report it effectively bypassed GPTZero while preserving meaning; praised as best free option; criticized for lacking advanced controls. Little independent Turnitin testing.
Test results. No run exists for this tool in the current cycle. It is listed for completeness; we will score it when it enters the corpus.
Profile facts from vendor sites and public reviews, gathered 2026-09-26. Marketing claims are the vendor's, not our findings.
Bypass rates are measured by our detector. Passing a detector does not make text human, accurate, or safe to submit. We have no commercial relationship with any tool listed.
Current cycle · 150 documents per tool unless flagged · last run date not supplied
Why this board exists
Humanizer marketing is noise. Measurements are not.
Every humanizer on this page advertises a 95–100% bypass rate. Measured on the same 150-document corpus, the best of them passed our detector 47% of the time. Two passed zero.
BypassRank is not affiliated with, sponsored by, or selling for any tool listed. We run each product's real output through our detector, publish the failures alongside the successes, and show the sample size behind every number. When our data embarrasses a vendor's marketing, the data ships.
How we test
Our methodology
One corpus, every tool, identical conditions. Here is what our detector is, what a bypass rate means, what the sample sizes do — and what we refuse to claim.
A tool's bypass rate is the share of its tested documents whose humanized output our detector classified as human-written. It measures detection outcomes only — not readability, factual accuracy, or meaning preservation.
Thirteen tested rows ran 150 documents each. AI Humanize Pro ran 32, Undetectable's API 80, and MyPerfectWords 125 — 2,187 documents in total. Smaller samples carry greater uncertainty; rows below 100 documents carry a Small sample badge. Rank describes this run and is not proof that one tool will outperform another on new documents.
Undetectable's API passed zero of 80 documents; MyPerfectWords passed zero of 125. The approximate rule-of-three upper bounds are 3.8% and 2.4% respectively at 95% confidence; zero observed passes does not prove a zero underlying pass rate. The source table's two additional columns were never defined — they are preserved as unlabeled source data in expanded panels and never affect ranking. TextToHuman has no measurement in this cycle and remains unranked.
Our detector and other detectors
We score every tool with our own AI detector. It is built similarly to Pangram's detector and scores almost the same as Pangram. Text that passes our detector therefore also passes other detectors such as GPTZero and Pangram.
Independence & fairness disclosure
D1
Vendors on this board advertise pass rates of 95–100%. The highest rate measured by our detector is 47%. Both facts are true; they are answers to different questions. The advertised rates are not measured on a shared corpus, and several are measured against detectors the vendor chose itself.
D2
A high bypass rate can mean a tool rewrites text well — or mangles it beyond recognition. Our detector measures detection outcomes only. Readability and meaning preservation are separate evaluations we have not yet run; until they exist, no rank on this board is a quality recommendation.
D3
Tools showing 0% passed zero of the documents we ran. Statistics bounds their true rate near zero but never at exactly zero; the table footnote carries the bound.
D4
This board reflects the current cycle as of date not supplied. Rows are updated only when a new run exists — never by estimate, interpolation, or vendor submission.
BypassRank is operated by MarkHuman. MarkHuman is not evaluated or ranked in this cycle.
FAQ
Frequently asked
What is an AI humanizer, and why would anyone use one?
+
A humanizer rewrites AI-generated text so AI detectors classify it as human-written. Legitimate uses include editing drafts, smoothing machine-translation artifacts, and reducing false-positive flags on genuinely human writing. Misuse — submitting undetected AI work as your own — is an academic-integrity and contract violation at most institutions and publishers. We publish measurements, not encouragement: the best tool on this board still failed more than half its documents.
Which detector do you test with?
+
Our own AI detector. It is built similarly to Pangram's detector and scores almost the same as Pangram, so text that passes our detector also passes other detectors such as GPTZero and Pangram. Each row shows the share of that tool's tested documents our detector classified as human-written.
Why do these rates look so much lower than the 99%+ on vendor websites?
+
Because vendor benchmarks and independent ones answer different questions. Vendors typically test their best mode, against a detector subset they choose, on their own samples, with retries until pass. We run every tool on one shared corpus under identical conditions and publish the document count behind every rate. When Rephrasy advertises "100% pass rates" and measures 19% here, both numbers describe their own test. Only one of them ran on a corpus the vendor did not pick.
Is a higher bypass rate a better tool?
+
Not necessarily. Our detector measures detection outcomes only. A tool can pass by rewriting well — or by degrading text until it no longer resembles machine output (MyPerfectWords' garbled synonym swaps are documented). Meaning preservation and readability are separate evaluations we have not yet run. Until they exist, treat this board as one axis of evidence, not a quality ranking. The disclosure under every profile says the same thing.
Why do some rows have fewer documents than others?
+
Most rows ran 150 documents. Undetectable's API ran 80, MyPerfectWords 125, and AI Humanize Pro only 32. Smaller samples carry greater uncertainty, so close ranks should be treated cautiously. Only AI Humanize Pro and Undetectable's API carry a "small sample" badge; the threshold is fewer than 100 documents. We keep them visible rather than dropping them — an incomplete measurement, honestly flagged, beats a missing one.
Does this tell me whether my text will pass Turnitin (or GPTZero, or my school's checker)?
+
For GPTZero and Pangram, yes: our detector is built similarly to Pangram's and scores almost the same as Pangram, so text that passes our detector also passes those detectors. Turnitin and institution-specific checkers are outside that statement's scope — they may use different thresholds, updated models, and document metadata signals we do not simulate, and Turnitin added explicit anti-"bypasser" detection in August 2025. Submitting undetected AI work where it is prohibited remains your risk, not a technical problem we can solve.
Are any of these rankings paid, sponsored, or affiliate-driven?
+
No. There is no paid placement, no sponsored rank, no affiliate link, and no "editor's choice" badge on this board. We hold no commercial relationship with any listed vendor. Rank order is computed from the published dataset — the same numbers you can inspect in the table — and a vendor cannot buy, dispute-suppress, or pre-review its row. If we ever add monetization, it will be disclosed in the methodology section before it ships.
Why does the same vendor appear twice (AI Humanize, Phrasly)?
+
Because vendors sell distinct rewrite pipelines as "modes," and modes are separate products in practice. We tested AI Humanize's basic path (40%) and its Autopilot loop (35%), and Phrasly's Direct (15%) and Balanced (11%). Two findings fell out: paid "upgrades" can score worse than the default they replace, and mode marketing rarely survives an independent corpus. Each mode gets its own row, its own document count, and its own uncertainty.
How often does the board update, and can a vendor request a retest?
+
Rows update only when a new test run exists — never by estimate or interpolation; the freshness line under the table carries the dataset date. Vendors cannot submit scores, pre-review results, or pay for retests. If a tool ships a major model update, it becomes eligible for the next cycle like everything else, and both the old and new results stay visible in the cycle history. TextToHuman is listed but untested; it will be scored when it enters the corpus.
Take the next step
Use the measurements yourself
Every number on this page comes from the same shared corpus. Open any row for the full profile, share the URL, or jump straight to the board. No paywall, no sign-up, no email gate.
