- On 1 October, Google added a rule: check all AI content by hand before you publish it. This includes the title, the meta description, the structured data and the alt text.
- On 5 October, Google said that invented authors are deception. This includes AI photos, made-up names and false credentials.
- OpenAI and Anthropic now put an invisible mark on AI text. The mark shows that a model wrote the text. It does not show that a person checked it.
On 1 October, Google updated its guidance on generative AI content. The new sentence is: “It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.” The review also applies to the title, the meta description, the structured data and the image alt text.
On 5 October, Google updated its page on helpful content. Fabricated creator profiles are now “a form of deception”. Google uses deception as a signal of a low-quality page.
Google did not announce a penalty or a ranking change. But its September spam update ran from 24 September to 8 October. Use the new guidance as your checklist now.
Why an AI mark does not replace a person
On 5 October, OpenAI announced a watermark for ChatGPT and Codex text in the EU. Anthropic marks text from supported Claude models and adds Content Credentials to files. Google marks its images with SynthID. The EU AI Act requires these marks from 2 August 2026, and from 2 December 2026 for systems that were already on the market.
A mark is a weak signal. Anthropic says that a detected mark “is not fully conclusive”. In OpenAI’s tests, a change of 25% of the words cut detection to 17%. A mark shows that a model touched the text. Only a person can show that the text is correct.
The check, in six steps
- Check each number, name and date against a source. Add the link.
- Read the title, the meta description, the structured data and the alt text.
- Put the name of the person who checked the page on the page.
- Add one line that says how you made the page.
- Label each AI image in its caption. Keep its metadata.
- Add one thing that only you have: data, a case or a photo.
Download the full check (nine points, plain text, free).
Why this helps you get cited
Last week, ChatGPT changed to GPT-6. When a model changes, AI answers cite different sources. In May, SISTRIX measured a 47% change in 48 hours. A model can verify a page with dated facts, linked sources and a named author. To see your position, use our free 20-question sheet (CSV).