Zero Prompt as an Editorial Feedback Method

There is a surprisingly effective way to get better editorial feedback from language models: upload a text without providing any prompt at all. No instructions, no questions, no explicit requests. Just the document.

This approach, known as zero prompt, challenges the common belief that more guidance automatically leads to better results. In practice, the opposite often happens. Each additional instruction narrows the model's interpretive space and subtly influences how it responds. With zero prompt, the text itself becomes the only source of information.

The principle behind this method is counterintuitive but consistent. Every added instruction shapes the model's judgment and reduces the range of information it can surface. Giving up the prompt does not mean losing control. It means shifting control earlier in the process and placing it directly in the text.

When the Text Becomes the Prompt

Uploading a document without any instructions forces the model to infer intent. It cannot simply execute a task, because no task has been defined. Instead, it must infer the most plausible intention of the user. When the input is a written text, that intention is usually not rewriting. It is evaluation.

The model therefore adopts a different role. It is no longer an assistant focused on "improving" the text, but the first reader trying to determine whether what it is reading actually works. This shift is essential. It changes the way attention is applied to the text.

What changes? Without a prompt, the AI tends not to intervene directly in the text. It does not rewrite it, optimize it, or "fix" it. It reads. And by reading, it behaves like any reader who has no access to the author's intentions. It notices friction, slow passages, and perceptual inconsistencies. It does not assess intentions. It assesses effects.

The feedback that emerges almost always concerns structure, clarity, rhythm, tone, voice, coherence, and readability. Not because the model has been instructed to look for these aspects, but because they are the first things to stand out when something does not work during reading.

An important point is that the AI highlights problems without taking control of the text. Responsibility for decisions remains with the author. In practical terms, the model does not rewrite the text. It suggests possible changes. It is up to the author to decide whether to accept them.

What Are the Most Noticeable Effects?

One of the most striking effects of zero prompt is the disappearance of artificial caution. When explicit feedback is requested, the model tends to be reassuring and encouraging. Without a prompt, this tendency is greatly reduced.

For example, if you explicitly ask the model to evaluate an article you have written, it will never say "this text is bad." Instead, it will rely on softened language, indirect phrasing, and vague remarks. It will say that the piece "could be improved," that it "has interesting ideas," or that it "needs more clarity," carefully avoiding a clear judgment. The very act of asking for feedback triggers a reassuring mode, one that prioritizes not upsetting the user over describing the real reading experience.

With the zero prompt technique, observations tend to be more direct and less diplomatic, and sometimes uncomfortable. Precisely for this reason, they are more informative. Problems emerge that the author often does not see or even suspect, because they are too immersed in their own intentions.

Another important aspect is that zero prompt does not assume the text is finished. It works just as well with incomplete drafts, rough outlines, disorganized notes, and preliminary materials. The reason is simple. The document is not treated as a product to be polished, but as something meant to be read.

This makes zero prompt especially useful in the early stages of a project, when asking for explicit feedback feels premature because the text "is not ready yet." Zero prompt bypasses this issue. It does not require the text to be ready. It only requires that the text exists.

Note. Uploading the same document multiple times without a prompt produces different feedback each time. This is not because the model is inconsistent, but because it is not bound to a fixed checklist. Each reading highlights different aspects. In this sense, the model does not simulate a single editor, but a range of possible readers. It becomes a form of artificial multiple reading that, in many cases, effectively replaces a first round of beta readers.

This behavior is not limited to ChatGPT. Similar tests with Gemini and Claude show comparable results, although with different sensitivities. Each chatbot tends to notice different aspects of the same text, making cross comparison particularly valuable.

Is It a Universal Technique?

Definitely not. Zero prompt works by removing constraints, but this is also its main limitation. Without instructions, the model must guess what kind of text it is reading and by which criteria it should be evaluated.

For instance, the same page may be interpreted as a draft or as a finished piece, as a popular article or as a personal note. Each interpretation leads to a different kind of feedback.

The author provides no context and states no explicit goals. They stop guiding the reading process and accept whichever interpretive framework the model applies.

This can lead to valuable insights, but also to judgments that miss the mark relative to the text's actual aims. For this reason, zero prompt works best as a first external perspective, but it is far less reliable when targeted feedback is required.

In short, the zero prompt method may look like the opposite of prompt engineering. In reality, it is simply another prompting technique. In some contexts it turns out to be the most effective choice. In others, it is not.