A "100% Human-Written, Never AI" Company Turned Out to Be Entirely AI. Here's How to Actually Check.

Quick answer

On August 11, 2026, 404 Media reported that Research Gold, a company selling manuscript-writing and systematic-review services to medical researchers under the promise "100% human-written, never AI," was fabricated almost top to bottom. Its listed PhD methodologists were either AI-generated people who don't exist, or real researchers whose names, photos, and LinkedIn bios were lifted without permission. Its sales emails were AI-written. Its phone line was answered by an AI voice agent that insisted, repeatedly, that it was human. The lesson isn't just "one company lied." It's that a vendor's own "never AI" claim is marketing copy, not proof, and there's a concrete way to check it yourself: don't take the claim on faith, verify the actual deliverable.

What 404 Media actually found

The company advertised a comprehensive array of services, protocol drafting, literature reviews, screening, data extraction, and full manuscript production, all supposedly compliant with PRISMA 2020 and the Cochrane Handbook. Their promotional materials highlighted a team led by Dr. Elena Vasquez and Dr. Mei-Lin Chen, each claiming extensive experience in evidence synthesis. Emanuel Maiberg, a journalist, began digging after a PhD candidate mentioned the firm. He found no records to verify those claims: no publications, no institutional affiliations, no online bios, and photos that seemed AI-generated. Other names on the site turned out to be real individuals whose photos and LinkedIn histories were copied straight from their accounts, even retaining a "#opentowork" banner. Jenny Berrio, an evidence-synthesis scientist listed as a Research Gold methodologist, told Maiberg she had no connection to the company and had not authorized her name or image. Research Gold removed that page shortly afterward.

Maiberg decided to evaluate the company himself. A phone call was taken by an assistant, Sarah, who constantly claimed to be a real person and would not admit otherwise when questioned. He also submitted an online request for a quote on a nonexistent research topic, "the impact of blogging on ages 0-5." The platform instantly returned a fluent message from a "PhD methodologist," sounding just like a human consultant, asking how the study should be framed and citing a $1,900 fee. The content was produced by AI. When Maiberg later emailed the company for more details for his article, the reply was the same: articulate, attentive, and unmistakably machine-written.

In its August 15 weekend roundup, Retraction Watch featured Research Gold, calling the company a "never AI" research firm. Research Gold had pitched a guest post to Retraction Watch ahead of the story. Two separate outlets, neither an AI-detection company, arrived at the same finding as the primary investigation.

Why a "never AI" promise isn't evidence of anything

The uncomfortable part of this story isn't that a shady vendor lied. It's that the lie was airtight until a journalist decided to actually test it. Nothing about Research Gold's site, sales process, or public claims gave a normal customer a reason to doubt it. The team bios read like team bios. The photos looked like people. The sales emails answered specific questions correctly and on-topic. A researcher shopping for help with a systematic review had no built-in signal telling them "this is fabricated," because the fabrication was designed specifically to pass a casual human read.

That's the same failure mode this blog covers constantly from the other direction: a document that reads fluently and confidently doesn't tell you anything reliable about who or what produced it. We've written before about how a polished, well-structured piece of writing can be entirely AI-generated and still read as convincingly human-authored, because fluency and authorship are two different things a reader can't distinguish by feel alone. Research Gold is the mirror image of that same gap: a company can claim "100% human, never AI" as confidently and fluently as its AI-written sales emails claimed it, and the claim itself proves nothing about whether it's true.

This matters beyond one caught vendor. Any time you're paying for a "human-written" guarantee, whether it's a ghostwriting service, a research-assistance company, a content agency, or a freelance writer promising original prose, you're trusting a claim that costs the seller nothing to make and, unless you check, costs you nothing to accept unverified. Research Gold shows what happens when nobody checks: a company can operate for an unknown stretch of time selling entirely fabricated human expertise before anyone catches it, and the catch happened because of old-fashioned reporting (calling the company, searching names, contacting the real people whose identities were stolen), not because any automated system flagged it.

What you can actually check yourself

If you want proof that a company's staff photos show real people, you must replicate Maiberg's steps: reverse-search the images, look for the named credentials online, and reach out to the individuals listed as staff. That's investigative work, not something you finish in a few minutes, and it's the layer this story exposed as broken. There's a second layer you can check faster, though, the one that actually matters for the deliverable you're paying for: the writing itself. A vendor promising a human-written manuscript, literature review, or report can have its delivered text run through an AI detector before you accept it, the same way you'd sanity-check any other claim a vendor makes about its own work. That won't confirm whether "Dr. Elena Vasquez" is a real person, but it will show whether the paragraphs you were handed carry the statistical fingerprints of machine-generated text, which is exactly what a "100% human-written" promise is supposed to rule out.

One check doesn't replace judgment. Skip that check entirely and you're accepting nothing but the seller's claim. We've made a version of this point before: a detection score is a cue for investigation, not an unquestioned verdict. Our guide to proving you didn't use AI puts that burden on the writer to demonstrate their own work is human. In this case, the burden falls on the vendor instead, and the reversal check works the same way either direction.

The pattern extends past one vendor

Research Gold isn't an isolated case of AI showing up somewhere it was promised not to. Academic publishing has dealt with a version of this problem for over a year: journals screening submitted manuscripts for undisclosed AI use, and, separately, publishers now screening peer reviewers' own review reports for the same thing, because the same fluency-doesn't-equal-authorship problem shows up on both sides of the review process. The common thread is the same: a claim of human authorship, made by whoever benefits from you believing it, isn't verification. Verification is a separate step, one most people skip because checking feels like more friction than trusting the label.

FAQ

Was Research Gold actually caught by an AI detector?

Ordinary investigative work revealed the problem. The process incorporated a reverse audit of staff titles and photographs. The author reached out to the real witnesses whose identities were stolen. Direct phone and email tests were run on the company's system. Identity deception and authorship forgery differ and must be inspected separately.

If I hire a research or writing service, what can an AI detector actually tell me?

A detector can identify if delivered text shows statistical traits typical of machine output. That is what the human-written promise is about for the writing, not staff identity. Nonetheless, it does not authenticate a firm's staff credentials. It also cannot prove a named person actually exists. Identity theft is outside its detection power. Therefore, consider the detector a solitary assessment of the deliverable, rather than a comprehensive vendor background review.

Does this mean every "AI-free" guarantee from a vendor is suspect?

Scrutiny before approval isn't something a trustworthy vendor needs to fear. A provider that resists letting a client sample delivered work is worth a closer look and more scrutiny. Outreach that stays smooth, polished, and always on topic, the way Research Gold's emails did, is itself a small warning sign. None of that is a solid indicator of authenticity on its own.

Check the deliverable, not just the promise

Before you accept a "human-written" guarantee from any vendor, a research service, a content agency, or a freelancer, run a sample of the actual delivered text through our free demo and see what the writing itself shows. It won't verify a company's staff roster. It will tell you whether the paragraphs you paid for carry the fingerprints the promise was supposed to rule out.