Substack показывает читателям, сколько текста рассылки написано с помощью ИИ

What the new tool shows

The email newsletter platform Substack launched a tool on July 22, 2026 that shows readers an estimate of how much of a newsletter's text was written with AI. TechCrunch reports this.

What the new tool shows

Substack's tool estimates the share of AI-generated text in each newsletter and displays this metric to the reader. According to TechCrunch, this is precisely an estimate — an approximate indicator, not a precise measurement: the platform signals that part of the material may have been written by a neural network. The estimate is tied to a specific newsletter issue, not to the author as a whole — that is, the metric is calculated anew for each letter.

For a subscriber, this is a new layer of context. Previously, a reader could not distinguish text written manually by the author from text assembled via ChatGPT or Claude and lightly edited. Now Substack adds a kind of transparency label to the newsletter that is visible right during reading. Substack is not the first to raise the topic of AI in content, but it is one of the first to embed the estimate directly into the reading interface rather than leaving it to the author's discretion.

Why Substack took this step

Substack built its business on the direct connection between authors and their audience and on trust in a specific name in the byline. The launch of the tool on July 22, 2026 reflects a broader industry shift toward transparency around AI-generated content — as TechCrunch notes, the platform's move signals precisely this trend.

The platform's economics reinforce the motivation. Substack retains 10% of the revenue from authors' paid subscriptions, so a reader's willingness to pay is a direct source of income for both the author and the platform itself. The higher the trust in the text, the more stable this revenue, and transparency works for the business model, not against it.

The mass spread of generative models erodes the value of authored text: if a newsletter is written entirely by a bot, the author's personal brand and the reader's willingness to pay for a subscription both fall. The AI-text-share label returns to the subscriber the right to understand what they are paying for, and leaves the choice — whether to trust such a newsletter or not.

Substack gives readers a way to gauge how much of a newsletter is

written by AI — a signal of a broader shift toward transparency around content created with artificial intelligence.

> — from a TechCrunch report, July 22, 2026

How much can this estimate be trusted

The reliability of any estimates of the share of AI text remains disputed. Automated detectors have historically erred in both directions — flagging human-written text as machine-written and vice versa. Back in July 2023, OpenAI shut down its own AI-text classifier, publicly acknowledging its low accuracy.

Substack does not disclose exactly how it calculates its metric, so the number should be treated as an indicator, not a verdict. The risk of false positives falls on authors: a writer who works by hand may receive an inflated AI-involvement estimate and lose some of the audience's trust through no fault of their own. The value of the tool right now lies not in percentage-point accuracy but in the very fact of labeling: the platform is making AI involvement visible by default for the first time.

What this means

Labeling the share of AI text is becoming a new norm for content platforms. For newsletter authors, the signal is clear: audiences increasingly want to know where the human ends and the model begins — and transparency is turning from an option into a competitive advantage. For readers, this is a first step toward a disclosure standard: subscribing to a newsletter is gradually starting to include not just the text itself, but also information about who actually wrote it — a human or a model.