twaingpt.com does not resolve to a product any more. It answers with one 301 to https://verva.com/?from=twaingpt, and that is the whole redirect chain — one hop, then a 200.
I went down this road because I wanted to know which company I would be handing a document to. That turned out to be three questions, not one, and only two of them have a clean answer in public.
Here is the short version. TwainGPT is an old name for Verva. undetectable.com is a second brand run by the same named company. undetectable.ai is a different company with a similar name. Those are three different relationships and it is very easy to flatten them into one.
The evidence for each link, and what it does not cover
The rename is the easy one. The redirect above, plus a banner across the top of every Verva page reading TwainGPT is now Verva, plus a footer that says Copyright © 2026 Verva, LLC. Redirect plus continuity notice is enough to call it a rename.
undetectable.com is where it gets interesting. Its footer also says Copyright © 2026 Verva, LLC — same legal name. But footers are cheap. The thing that convinced me was the link targets: I collected every <a href> on that homepage and filtered for auth routes. Six links, six identical destinations, all of them https://verva.com/login?ref=undetectable. The anchor text on those six is Try for free, Get started for free, Get Started, Get Started, Get Started, Get started for free. There is no separate account system behind that brand; every door opens onto the same login.
And then this, which is the reason I am writing any of it down:
<link rel="canonical" href="http://localhost:3000"/>
That is the raw HTML of undetectable.com, served today, and the rendered DOM agrees. Not a staging subdomain. Not a protocol mismatch. A dev server address, shipped to production, with http:// on it. Meanwhile verva.com and verva.com/pricing both emit correct self-referencing canonicals. Same operator, two brands, one of them built by someone who did not have the env var wired up.
I want to be careful about what that proves, because it is tempting to overread it. It proves the tag is wrong. It does not prove the two brands share a balance, a queue, or a model. Matching numbers and a shared login are strong evidence of a shared operator, not of one backend.
What it does mean practically: do not use a canonical tag as an identity test. The visible URL, the legal terms, the footer and the login targets all still correctly identify that site. The one machine-readable field meant to say "this is who I am" is the only one that is wrong.
undetectable.com and undetectable.ai are not the same company
This is the one that actually costs people money, because the names differ by a TLD.
undetectable.ai's terms open with: We are Undetectable Inc., doing business as Undetectable AI ("Company," "we," "us," "our"). That page names Undetectable Inc. five times and Verva, LLC zero times. Its pricing-page footer reads ©2026 Undetectable Inc. above 1309 Coffeen Avenue, Sheridan, WY. Its canonicals are correct. Its pricing page and humanizer page are entirely its own.
So in the public paperwork these are two separate legal identities. I did not investigate ultimate ownership, investors, or any partnership that would not show up on a terms page, and neither should you read this as ruling one out.
The units do not line up, which is the part that will bite you
I put all three into one table because every number I found was measured against a different denominator.
| Surface | Free / entry figure stated | Per-input ceiling shown | Paid allowance shown |
|---|---|---|---|
| verva.com | Homepage FAQ has a "Can I use Verva for free?" entry; I could not get the accordion to expand, so I have no figure |
0 / 2,000 words counter in the editor |
Basic 8,000 / Premium 30,000 / Ultimate Unlimited "Humanizer Words" |
| undetectable.com | no word figure stated; 250 occurs 0 times in 3,879 visible characters with the FAQ expanded |
0 / 2,000 words counter in the editor |
Basic card shows $8 and 8,000
|
| undetectable.ai | Free Trial is limited to 250 words. |
it can handle up to 10,000 characters in one go |
10K / 20K / 35K / 50K words per month |
Words per input, words per month, "Humanizer Words" per plan, and characters per run are four different units. A 2,000-word input counter and an 8,000-word plan allowance are not in conflict and neither one tells you that 8,000 words fit in a single operation.
One more thing worth knowing before you compare prices: Verva's pricing page defaults to the Annually view, and the annual view is the cheaper one. Flipping it to Monthly today changed the prices and left the allowances alone.
| Plan | Annual view | Monthly view | Humanizer Words (both views) |
|---|---|---|---|
| Basic | $8 / month, billed annually | $10 / month, billed monthly | 8,000 |
| Premium | $20 / month, billed annually | $25 / month, billed monthly | 30,000 |
| Ultimate | $40 / month, billed annually | $50 / month, billed monthly | Unlimited |
If you quote one of those numbers to a colleague, say which toggle you were looking at.
What I could not test, and would not guess at
The Verva humanizer page has one <input type="file">, and its accept attribute reads:
.txt,.md,.rtf,.csv,.json,.docx,text/plain,text/markdown,text/csv,
application/json,application/vnd.openxmlformats-officedocument.wordprocessingml.document
That attribute tells you what the file picker will let you select. It says nothing about whether the server accepts the upload, finishes the job, or gives you a .docx back. I did not put a file in the picker, did not press Humanize, did not create an account and did not pay anyone, so I have no claim to make about any of that.
I did search the rendered page for the words that would tell a user what comes out the other end. Visible text only, which is why the .docx in that accept attribute does not count — nobody reads attributes. Across 7,289 visible characters on the Verva humanizer page: download 0, DOCX 0, same format 0, editable 0, formatting 0, footnote 0, citation 0. The same search on the same page found Upload Doc once, Paste Text once, 2,000 once, one file input and two textareas — so the search could see things that were there. On undetectable.com the same seven terms also came back at zero, with the same controls found as positive checks.
Those zeroes are scoped to those two pages on the day I read them. They are not a claim that the product cannot return a file. They are a statement that the page does not say it does, which is a different and much smaller claim.
If you are evaluating one of these
Keep five things on the same line whenever you write a limit down: domain, access tier, amount, unit, scope. Drop any one of them and a correct number ends up attached to the wrong product, which is exactly how "Verva gives you 2,000 free words" gets into circulation.
Then test the output path before the input path matters. Make a throwaway file with one paragraph that has a citation, a number and a term that must not change, run only that, and open what comes back.
Disclosure, since it is relevant: I work on HumanPen, which sells a competing document humanizer, so treat everything above as a competitor reading public pages and nothing more. If you want the longer version with every page and snapshot listed, it is at humanpen.net/blog/verva-ai-humanizer-review.
All readings above are from rendered first-party pages on 15 September 2026. Pricing pages, plan labels and metadata change; recheck the exact domain before you pay anyone.
Top comments (0)