Habr AI explains why the "write like a human" prompt doesn't save texts from formulaic prose

Habr AI has published an analysis of why the request "write like a human" rarely makes an AI text feel alive. The main idea is simple: the problem is not individual markers such as lists, gerund phrases or bureaucratese, but the fact that models often do not understand what type of speech and what authorial stance the task calls for in the first place.

Not about the words

The author takes issue with the popular practice of "cleaning" a text of everything that gives away a neural network at first glance. If an editor mechanically bans lists, turns of phrase, parenthetical constructions and an academic tone, they treat the symptoms rather than the cause. This approach sometimes makes paragraphs shorter and punchier, but it adds no thought, internal logic, or sense that someone is really standing behind the text. The form changes, but the feeling of emptiness remains.

According to Habr AI, the command "write like a human" sets the model a false goal from the start. It does not explain what exactly needs to be written: an instruction, a news item, a column, an analysis or a controversial comment. Instead, the algorithm receives an abstract demand to imitate subjectivity. Hence the standard result: an averaged, safe and overly smooth text that resembles everything and nothing in particular. It replaces a concrete editorial brief with a vague mask of humanity.

Styles and tasks

At the center of the argument is an old philological idea: different tasks require different functional styles and types of text. There is no single universal "human writing" that works equally well in news, instructions and opinion journalism. When an author or editor first defines the task, the genre and the expected role of the text, the requirements for the language become much more precise, and working with the model becomes noticeably more practical. This is where philology proves more useful than formulaic prompt engineering.

• An instruction relies on the imperative, a sequence of steps and clear wording.
• A short news item is built around facts, context and minimal interpretation.
• An analytical text requires conflict, paradox and the author's own line of thought.
• Opinion journalism works when a position, not an impersonal summary, can be heard in the text.

A simple conclusion follows: you cannot edit all AI material with the same set of bans. Where a dry structure is appropriate, lists only help. Where argument and intonation are needed, the trouble is not bureaucratese as such but the absence of a point of view. The author compares the model's standard answer to a meaningless, universal "42" given to any question: the form may look decent, but in substance it misses the task.

Why the text is dead

Special emphasis is placed on the difference between a person who simply types and a professional editor. A copywriter or philologist usually has years of training behind them: reading, analyzing literary and journalistic texts, a sense of rhythm, an understanding of subtext, attention to syntax. That is why such a person notices not only clichés but a subtler problem: the lack of an inner movement of thought, which makes a text feel lifeless even without obvious "AI markers."

"Write like a human" is a command to the algorithm to imitate subjectivity.

This is exactly why fighting the "smell of AI" through cosmetic editing looks like a dead end. A model can be taught to avoid certain words and constructions, but that does not amount to the emergence of intention, a position or intellectual risk. If the task requires real reasoning, a single prompt is not enough: you need either a strong human editor or a much more precise task definition in which the genre, conflict, audience and quality criteria are set in advance. Without this, the model almost inevitably slides into the safe middle.

What this means

For editors and content teams, the takeaway is unpleasant but useful: a good AI text starts not with banning "neural network" turns of phrase but with choosing the right genre and authorial role. The more precisely the task is described, the less magic there is in the prompts and the better the chance of getting a text that works as intended rather than merely disguising itself as human. This changes both the approach to editing and the process of defining the task itself.