AI content detection

What does Google say?

Google does not judge the mere use of AI. According to its guidance, using generative tools to produce many pages without adding value for users may violate its spam policy on scaled content abuse. Other points from the guidance:

• Generative output can contain inaccuracies, so content should be fact-checked and reviewed manually.
• Titles, meta descriptions, structured data and alt text must be accurate.
• Sharing how content was created can give readers context.
• AI-generated product images need IPTC DigitalSourceType metadata (TrainedAlgorithmicMedia).

What does the EU AI Act add?

Article 50 of Regulation (EU) 2024/1689 sets transparency obligations. Providers of AI systems that generate synthetic audio, image, video or text must mark outputs in a machine-readable format. Deployers of deepfakes must disclose that content was artificially generated or manipulated. A disclosure duty also applies to AI-generated text published to inform the public on matters of public interest. The article also sets exceptions. The obligations apply from August 2, 2026. For systems placed on the market before that date, the deadline for machine-readable marking runs until December 2, 2026, under the amendments in Regulation (EU) 2026/1744.

This is not legal advice.

How reliable are detectors?

A research group around Debora Weber-Wulff tested twelve freely available detectors and two commercial systems (Turnitin, PlagiarismCheck) in 2023 and concluded that the tools are neither accurate nor reliable. The detectors leaned toward classifying text as human-written, and their performance dropped substantially once the text was obfuscated, for example by paraphrasing. A detector score is therefore at most a hint (assessment) and weak grounds for a verdict about where a text came from.

Last updated: October 7, 2026