Substack introduces new tool to identify AI-generated content

Yesterday 07:20

By: Azzat Manal

Substack introduces new tool to identify AI-generated content

Substack has introduced a new feature designed to help readers determine whether online articles and posts were written by humans or generated with the help of artificial intelligence.

The publishing platform developed the new system in partnership with Pangram, a company specializing in AI-generated text detection. The tool allows users to analyze articles, posts, comments and replies to assess the likelihood that the content was produced by an AI system.

According to Substack, the initiative is part of broader efforts to increase transparency and address the growing presence of automated content across the internet. The platform's chief executive said the company is particularly concerned about content created with AI but presented to readers as entirely human-produced.

The detection system can currently be used on texts containing more than 100 words that were published from July 21 onward. The results are displayed only to the person requesting the analysis rather than being automatically shown to other readers.

The company has acknowledged that AI detection technology is not perfect and can sometimes produce inaccurate results. However, it said Pangram's system has demonstrated a high level of accuracy in independent testing.

The feature is currently available through Substack's web platform and its iOS application, with support for Android devices expected to be introduced later.

The new tools are also aimed at writers. Authors can use the detection system before publishing their work to see how their content is classified. Substack has also introduced a section called "How I make this," allowing writers to explain how they created their work and whether they used AI tools during the process.

If a writer believes that an AI detection result is incorrect, the platform provides an option to report the result and request that the classification be reviewed.

Substack is also considering additional measures that would give writers more control over how AI is used in comments and replies. The company has indicated that it may also provide readers with greater control over the type of content recommended to them.

The move comes as AI-generated material becomes increasingly common across digital platforms. Industry estimates suggest that a significant share of online posts may now be created entirely or partly with the help of generative AI.

Substack is not alone in attempting to address the issue. Other major technology and social media companies are also exploring ways to identify and limit low-quality automated content, while working on systems that can distinguish between human and machine-generated material.

However, the effectiveness of AI detection remains a subject of debate. As generative AI models become increasingly sophisticated and capable of producing natural-sounding text, detection systems will face the challenge of keeping pace with rapidly evolving technology.