Does Humbot bypass Pangram? No one has measured it — because Pangram’s own researchers rated Humbot’s output unreadable, graded it worst-tier, and then excluded worst-tier tools from every detection test they ran.
Key Takeaways
- Pangram’s own paper rates Humbot L3, the worst tier, with a verbatim note: “Sentences are uninterpretable and random additions to the text make it unreadable.”
- L3 tools were excluded from every detection experiment in that paper. Humbot is mentioned exactly twice in the whole document, never with a detection number.
- Humbot’s built-in checker advertises results from “All Popular AI Detectors” — Originality.ai, GPTZero, ZeroGPT and Copyleaks. Pangram is not among them.
- So a Humbot user verifying their own output structurally cannot see a Pangram result inside the product.
- A rival detector’s October 2025 test found Humbot moved Originality.ai from 100% AI to 100% AI and GPTZero from 100% to 98%.
- The pricing page advertises a “30-Day money back guarantee”; the actual refund policy is 7 days and void past 1,000 words — a third of the entry plan’s monthly allowance.
- The honest caveat: a badly-degrading tool can sometimes evade more. Humbot is unmeasured against Pangram, not proven to fail.
Want to bypass Turnitin in 2026? Grab the free prompt pack.
Get the exact text-humanization prompts I use to drop an AI score by hand — copy, paste, submit. Free, straight to your inbox.
Send me the free prompts →Does Humbot bypass Pangram?
Unmeasured, and that is the precise word. There is no Humbot-versus-Pangram result anywhere — not from Pangram, not from Humbot, not from an academic, not from a competing vendor, and not on Reddit, where the two names have never appeared in the same thread.
What makes this case sharper than most is that Pangram had Humbot in front of it and made a documented decision not to test it. And Humbot, for its part, has built a product that cannot show you the answer either.
This page reviews the published record and reports no first-party test. Every claim is dated and attributed, and where a source has a commercial stake, we say so.
What did Pangram’s own paper say about Humbot?
The DAMAGE paper (Masrour, Emi and Spero of Pangram Labs, January 2025) audits humanizing tools for output quality. Its verbatim assessment of Humbot AI:
“Sentences are uninterpretable and random additions to the text make it unreadable.”DAMAGE, Table 9 — arXiv:2501.03437, January 2025
That is an L3 grade, the bottom of three tiers, defined in the paper as tools that “often added nonsensical phrases, words, and characters, constructed incorrect and uninterpretable sentences, and often distorted the meaning of the text.” On the paper’s fluency measure, L3 tools won 2.67% of head-to-head comparisons, against 26.0% for the top tier.
Then the disqualification. Introducing its detection experiments, the paper states it will “narrow our scope to L1 humanizers… because their subtle changes are the hardest to detect by eye… making them most relevant in real-world adversarial attacks.” Humbot is L3. It appears exactly twice in the entire paper — once in the roster of tools studied, once in the quality audit — and never alongside a detection figure.
Why isn’t Humbot in Pangram’s benchmark?
Pangram’s August 2025 benchmark scores 20 named tools — Undetectable AI at 90.3%, StealthGPT at 95.6%, QuillBot at 100.0%. Humbot is not among them, and no reason is given. A later Pangram post on how students avoid detection names only Grammarly, QuillBot and Undetectable AI.
By July 2026 the question became unanswerable in principle. The Pangram 4 technical report tests 13 commercial humanizers, selected by “top web traffic in early 2026”, and labels every one of them “Commercial A” through “Commercial M”. Whether Humbot is in that set cannot be established from outside.
| Pangram publication | Date | Humbot present? | Detection figure? |
|---|---|---|---|
| DAMAGE paper | Jan 2025 | Yes — audited, rated L3 | None — L3 excluded from all experiments |
| Humanizer benchmark, 20 tools | Aug 2025 | No | None |
| How students avoid AI detection | Oct 2025 | No | None |
| Pangram 4 technical report | Jul 2026 | Unknowable — 13 tools anonymised | None attributable |
Why can’t Humbot’s own checker answer this?
Because of what is in the panel. Humbot sells a built-in AI checker under the heading “Our AI Checker = All Popular AI Detectors”, promising “simultaneous detection results from top scanners like Originality AI, GPTZero, ZeroGPT, and Copyleaks all with a single click”, and adding that “the results our AI checker gets are the exact results you can get yourself on those popular AI detectors.”
Pangram is not in that list. Neither is it named anywhere else on the site — the homepage promises output that “passes even the strictest AI detectors like Turnitin, Originality.ai, GPTZero, ZeroGPT and Copyleaks”, and Pangram appears in none of those enumerations.
The practical consequence matters more than the marketing point. A student who humanizes text in Humbot, checks it in Humbot, and sees green has verified nothing about the detector their institution may actually be running. We cover that gap generally in how much AI detectors disagree.
What do independent tests show?
Only against detectors other than Pangram, and the results are poor. Originality.ai’s October 2025 review — read it adversarially, since Originality sells a competing detector — reported its own tool at 100% AI both before and after Humbot, Copyleaks unchanged at 100%, and GPTZero moving only from 100% to 98%. Its verdict was that Humbot “does not humanize AI text” but “merely runs every third or fourth word through a thesaurus.”
That description lands close to Pangram’s L3 note, written ten months earlier by an unrelated evaluator. A third assessment, a competing humanizer’s 14-tool benchmark published in August 2026 with its conflict disclosed, counted 321 writing errors across 33 samples for Humbot and two documents broken outright.
Three independent evaluators, none citing the others, reaching the same conclusion about output quality. That is about as close to corroboration as this market offers.
What about the pricing and the guarantee?
The entry plan is $7.99 a month for 3,000 basic and 1,000 advanced words, with a 500-word cap per input. Higher tiers sit around $9.99. A free trial is referenced but no free word allowance is published anywhere we could find.
The guarantee is where the numbers stop agreeing with each other. The pricing page advertises a “30-Day money back guarantee”. The refund policy sets a 7-day window and adds that “if your word usage exceeds 1000 words during this period, you will not be eligible for a refund” — one third of the entry plan’s monthly allowance, which is to say roughly two documents. Refunds are limited to the original purchaser and exclude third-party purchases.
The operator named in the privacy policy and terms is EchoAI Limited, described as a Singapore-based company. No registered address or company number is published. Our full Humbot review covers the product itself.
What should you take from this?
The defensible answer is that the Humbot-versus-Pangram question has never been asked by anyone in a position to answer it, and that Humbot’s own tooling is built so its users cannot ask it either. Everything else is inference — and the inference cuts both ways, because badly degraded text can evade detectors precisely by being bad.
What is not in dispute is the output quality. Three unrelated evaluators, across ten months, independently described Humbot’s text as unreadable, thesaurus-swapped, and error-strewn. Whatever a detector makes of that, a marker reading it will form a view.
If the real worry is being wrongly flagged on your own writing, keep your drafting history — it costs nothing and outperforms any rewriting tool as a defence, and it matters most for the writers detectors treat worst: ESL writers and AI detection and false positives for neurodivergent students. Check before you submit with the best pre-submission check and the Turnitin self-check, and strip the obvious tells by hand first via what to remove before using an AI humanizer.
For the Pangram cells where evidence exists, see Undetectable AI, StealthGPT, GPTHuman, WriteHuman, QuillBot, StealthWriter and Phrasly, plus the overview at bypassing the Pangram AI detector and the Pangram review.
Get the free prompt pack instead
The manual rewriting prompts that lower AI signals without a subscription — and without betting a submission on a number nobody has reproduced. Free, no card.
Frequently asked questions
Does Humbot bypass Pangram?
Nobody has measured it. No Humbot-versus-Pangram result exists from Pangram, from Humbot, from an academic or from a competitor. What does exist is Pangram’s own quality audit, which rated Humbot worst-tier and then excluded worst-tier tools from all of its detection experiments.
What did Pangram’s paper actually say about Humbot?
Its verbatim assessment reads: “Sentences are uninterpretable and random additions to the text make it unreadable.” That earned it an L3 grade — the bottom of three tiers. The paper is explicit that these tiers measure “faithfulness and fluency, not… effectiveness at bypassing AI detectors.”
Doesn’t badly mangled text evade detectors more easily?
Sometimes, and this is the honest complication. Pangram itself observed that “the ‘good’ humanizers in terms of fluency are more detectable, and the ‘bad’ humanizers are less likely to be detectable.” So L3 is not proof Humbot fails. It does mean any evasion comes at the cost of text a human reader can see is broken.
Why can’t I just check Pangram inside Humbot?
Because Humbot’s checker does not include it. The product advertises simultaneous results from “top scanners like Originality AI, GPTZero, ZeroGPT, and Copyleaks” and says those are “the exact results you can get yourself” — but Pangram is not in the panel. You would have to check Pangram separately.
What do independent tests show?
Only against other detectors. Originality.ai’s October 2025 review reported its own detector at 100% AI both before and after Humbot, GPTZero moving 100% to 98%, and Copyleaks unchanged at 100%. Its conclusion was that Humbot “merely runs every third or fourth word through a thesaurus” — though note Originality.ai sells a competing detector.
What does Humbot cost?
The entry plan is $7.99 a month for 3,000 basic and 1,000 advanced words, with a 500-word input limit. Higher tiers sit around $9.99 a month. A free trial is referenced but no free word allowance is published.
Is there really a 30-day money-back guarantee?
The pricing page says so. The refund policy page says something narrower: a 7-day window, and no eligibility at all if your usage exceeds 1,000 words in that period. On the entry plan that is one third of a month’s allowance. Refunds are also limited to the original purchaser.
Who operates Humbot?
Its privacy policy and terms name EchoAI Limited, described as a Singapore-based company. No registered address or company number is published on the site.
