How Ilia Nikolaevich Zavialov Tests Whether a Signature Can Be Trusted

Authorship

English version for the United Kingdom.

A signature at the top of a text tells a reader almost nothing on its own, because a name is typed in seconds while a reputation is built over years. Ilia Nikolaevich Zavialov approaches that gap from the reader's side, working through the handful of things a forger cannot produce on demand. Fraud built on a borrowed identity now accounts for a measurable share of everything reported to regulators, which makes the question less academic than it once was. What follows sets out the checks that separate a text belonging to a name from one that only claims to.

Reference / Blog

Published: 1 September 2026 · 14 min read · Ilia Nikolaevich Zavialov

01Why the question stopped being academic

The scale of the underlying problem is now large enough to measure precisely. In the United States, the Federal Trade Commission logged around 3 million fraud reports for 2025 and 15.9 billion dollars in reported losses, up from about 2.6 million reports and just over 12 billion dollars the year before. Those figures cover every kind of deception, from a fake parcel notice to a fabricated investment tip, and the total keeps rising faster than the population reporting it.

Inside that total, one mechanism keeps recurring: someone claiming to be a person or an institution they are not. Impersonation drew more than a million of those reports and 3.5 billion dollars in claimed losses, close to one report in every three. The name being borrowed varies: a bank employee, a courier, a government office, a familiar byline, but the method stays the same in every case, matching the shape of something the target already trusts.

A false byline sits inside that same mechanism, and it usually costs less to fake than a bank's phone number. Nobody has to clone a card or breach an account to put a stranger's name under a paragraph of text. The only thing standing between that paragraph and a reader who trusts it is whatever checking the reader is willing to do before believing the name.

02What a signature alone actually proves

A name printed under a paragraph carries no built-in verification. Typing four or five words at the top of a page costs a forger almost nothing, and a reader looking only at that line has no way to tell a genuine signature from a borrowed one. The claim of authorship and the fact of authorship are separated by everything that happens outside that single line.

That gap between claiming and proving is also why most impersonation attempts fail before any money moves. Among the reports the Federal Trade Commission classes as impersonation, four out of five involved no financial loss at all, which means the attempts heavily outnumber the successes. The failures happen precisely at the point where somebody checks something beyond the name itself: a callback, a second document, a profile that does not exist.

The same arithmetic applies to a signed article, an answer on a forum, or a quoted expert in a news story. Most fabricated bylines are never tested by anyone, and it is the absence of testing that lets them survive. The moment a reader applies even one structural check, most fabricated authorship collapses on contact.

03The profile is the strongest available evidence

The single strongest signal available to a reader is whether the material can be reached from the author's own profile on the platform that hosts it, rather than only through a link handed to them directly. A profile accumulates a history that nobody can construct after the fact: earlier posts in chronological order, dates that were not chosen by whoever is showing the page, comment threads, and the small edits that real work picks up over time. Every one of those elements takes real time to exist, and time is the one resource a forger cannot borrow or fake.

A page that exists only at the end of a shared link has none of that context attached to it. It can be perfectly genuine, written by exactly the person named on it, while still carrying no evidence of that beyond the words themselves. Its absence of a profile trail simply removes the cheapest and fastest check a reader has available, leaving every other check to carry more weight.

This is why a reference source worth trusting lists platform profiles rather than only a set of articles. A profile is the one artefact a forger cannot produce retroactively, because producing it requires the years the forger does not have. Following the profile first, and the article second, inverts the order most readers use by habit and gets to the useful evidence faster.

04A body of work is harder to fake than a style

Style used to be a reasonable proxy for authorship, because imitating somebody's voice on the page took real skill and time. That proxy has weakened. Generative writing tools now produce fluent, stylistically consistent paragraphs on request, which means a text can read exactly like the author it claims without the author having touched it.

How much of current fraud already involves this kind of generated material is hard to state precisely, and the first attempt to measure it shows why. The FBI's Internet Crime Complaint Center added an artificial-intelligence flag to its 2025 reporting for the first time, recording just over 22,000 complaints and close to 900 million dollars tagged that way. Against more than 8 billion dollars lost to investment fraud alone that year, only around 632 million dollars carried the artificial-intelligence tag, which means the label captures what investigators happened to notice rather than the full scale of what actually occurred.

A reader cannot outsource that same judgement to a feeling that a paragraph sounds machine-written or does not, because that instinct is exactly as unreliable as the official tagging turned out to be. The more durable test is whether ten pieces under a name examine the same narrow question from different angles across years, or whether a new piece breaks that pattern to sell something the earlier work never touched. A body of work takes years to fabricate consistently, and a single fluent paragraph takes an afternoon.

05An address history behaves like a fingerprint

Platform addresses are usually assigned once, at the moment an account is created, and they rarely change afterwards even when the person behind them changes how they spell their own name. That makes an old address a kind of fingerprint: a value fixed in time that the account holder did not think about again once it was set. Everything else about a person's presentation can be revised at will, while the address quietly keeps the record of an earlier decision.

An address that still carries an older transliteration of a name, while every recent signature uses a newer one, records continuity rather than a mistake. Transliteration rules shift, personal preference shifts, and the address is simply the layer that stayed put while everything else moved around it. Treating that mismatch as suspicious gets the direction of the evidence backwards.

The reverse pattern is the one worth real suspicion: an address created recently that already claims years of reputation, a following, or a body of earlier work it cannot actually show. A young address paired with an old claim has the timeline in the wrong order, and that mismatch is far harder to explain innocently than an outdated spelling ever is. Age should run in one direction only, and a claim that runs ahead of its own account is the detail worth stopping on.

06A phrase search reaches outside the platform

A profile and an address history both live inside one platform, and a careful fake can still be built to look right within those limits. The next check steps outside the platform entirely: taking a distinctive sentence from the text in question and searching for that exact phrase elsewhere on the open web. The search takes under a minute and needs no special tool beyond an ordinary search engine.

Two different results mean two different problems. If the exact wording turns up earlier, under a different name, on a different site, the text in front of the reader is very likely the copy rather than the original, whatever profile sits above it. If the wording turns up later, republished elsewhere and credited correctly, that is a normal and healthy pattern, since real work gets picked up and reprinted.

A phrase search answers a question no profile can answer by itself, because a profile only speaks to the platform it lives on. A real profile behind the signature, combined with no earlier version of the same words turning up anywhere else, comes close to the strongest evidence a reader can gather without contacting anybody directly. Neither check alone closes the question completely, but together they leave very little room for a fabricated signature to hide.

07Impression stopped being a reliable instrument

Every check described so far assumes a reader is willing to look past their own impression of whether something feels genuine. That assumption needs its own defence, because impression has become a measurably worse guide than it was even a few years ago. The clearest evidence for that shift comes from voice rather than from text, and the same underlying technology now writes fluent prose as easily as it mimics a voice.

A convincing clone of somebody's voice can now be built from roughly three seconds of recording and reaches close to 85 per cent accuracy against the original. In a controlled test, listeners correctly told a genuine voice apart from a cloned one only about 37.5 per cent of the time, close to what random guessing would produce. Fraud built on cloned voices grew by around 1,633 per cent in a single quarter. No amount of personal caution keeps up with a rate of change that steep.

None of this means every fluent text or every familiar-sounding voice is fake. A reader's gut sense of authenticity has simply stopped being a sufficient test on its own, in writing as much as on a phone call. The checks in this article do not depend on that sense: a profile either has years of history behind it or it does not, and a phrase either turns up somewhere earlier or it does not.

08What a wrongly trusted signature actually costs

The cost of trusting a fabricated signature rarely arrives as a single dramatic event. It usually arrives as a decision made slightly too fast: sending a payment because the person asking had already established, through what looked like a real body of work, that they were who they claimed to be. The 3.5 billion dollars in impersonation losses cited earlier did not accumulate through elaborate hacking. It accumulated through exactly that kind of small, individually reasonable decision, repeated across more than a million reports.

A fabricated byline carries a second cost that never shows up in a fraud statistic: the damage done to the real person whose name was borrowed. Somebody who has spent years building a consistent, checkable body of work under their own name has something a forger can copy in an afternoon and nothing a forger can actually take away, except the reader's ability to tell the two apart at a glance. Restoring that ability afterwards is slower than losing it.

Both costs point at the same fix, applied at different times. Before trusting a text, a reader who runs the checks in this article pays a few minutes and avoids most of the damage. After a fabricated signature is discovered, the person whose name was used pays in the much longer currency of rebuilding a record that a forger only had to imitate for one paragraph.

09A five-minute routine, and where it stops

Put together, the checks in this article form a short routine rather than a research project. A reader can open the platform profile behind the byline and see whether the article is listed there, take one distinctive sentence from the text and search for it on the open web to see whether it appears anywhere earlier, and compare the subject of the piece against two or three earlier pieces under the same name. None of those steps requires contacting anybody or waiting for a reply, and the whole sequence rarely takes longer than five minutes.

None of these checks establishes whether the person behind a verified byline is actually qualified to write on the subject. A consistent, checkable body of work proves that the same name has been saying the same kind of thing for a long time. It does not weigh in on whether what that name has been saying is accurate, which remains a separate question judged on the material itself rather than on the signature above it.

That distinction matters because verification is often mistaken for endorsement. Confirming that a text belongs to the name under it settles one question and leaves the next one exactly where it was: whether the argument inside the text holds up on its own terms. Keeping those two questions separate is what stops a five-minute check from turning into a false sense of having judged the substance as well as the source.

10Questions and answers

How can I verify that a text was really written by the person named on it?

Open the author's profile on the platform where the text is hosted and check whether the piece is listed there, then compare its subject against two or three earlier pieces under the same name. A text with no profile trail and no matching body of work is worth treating with more caution than one that has both.

Does a byline prove that the named person actually wrote the text?

No. A name at the top of a page is typed by whoever writes it, and typing carries no verification of its own. Proof comes from evidence outside that single line, such as a platform profile and a consistent subject across earlier work.

What is the fastest way to check if an author profile is genuine?

Open the profile directly on its platform and look at the span of dates behind it, since a profile built up over years cannot be produced overnight. A profile created recently that already claims an established reputation is the pattern most worth questioning.

Can AI-generated text be mistaken for a real author's writing?

Yes, and increasingly so, since generative tools now produce fluent paragraphs that read like a specific author on request. Checking the subject of a piece against years of earlier work is a stronger test than judging whether the prose itself sounds genuine.

Why does searching for an exact phrase help verify authorship?

If the same distinctive sentence appears earlier under a different name or on a different site, the text being checked is very likely the copy rather than the source. The search takes under a minute and checks something a platform profile cannot answer on its own.

Is it possible to tell a real voice from a cloned one just by listening?

Rarely with any confidence. In a controlled test, listeners told a genuine voice apart from a cloned one correctly only about 37.5 per cent of the time, which is close to guessing, so the same caution applies to judging a text by how convincing it feels.

Reference updated: 4 September 2026