SEO Recap covering August 15, 2026: Cloudflare blocking Googlebot and Anthropic's SynthID watermark
Daily summary of what matters in SEO, GEO, AEO, and AI search generated with Claude Code (beware of hallucinations)
Generative Engine Optimization (GEO) & Answer Engine Optimization (AEO)
SEO Pulse: Cloudflare AI Blocking, ChatGPT's Index, GA Benchmarks, and Google's SerpApi Refiling
Source: Search Engine Journal
- Starting September 15, 2026, Cloudflare will treat Googlebot and Bingbot as bots that both index for search and gather AI training data, so any setting that blocks AI training (including the old "Block AI bots" switch) will block them too. Review the setting before then or opt out.
- Resoneo found OpenAI's in-house index served sites with no content deal the same way as licensed partners, reading ChatGPT answers captured in July via a source field OpenAI removed in late July.
- OpenAI's index keeps a page's title and ~200 characters from the top, so any template content shown before the first paragraph uses up that space.
- Google's Ask Advisor agent will compare campaign performance in Google Analytics against anonymized averages from similar businesses, though it hasn't said whether it uses the existing customizable peer group or when it rolls out.
- Google amended its DMCA complaint against SerpApi to include content licensing terms after a judge dismissed both claims for failing to show copyright owners authorized its anti-scraping system. The outcome affects whether rank trackers and AI visibility tools can collect Google results at scale.
Anthropic Reveals How Its Text Watermark Works and How It Can Be Defeated
Source: Search Engine Journal
- Anthropic confirmed future Claude models will watermark text outputs to comply with the EU AI Act, using a version of Google DeepMind's 2024 SynthID-Text approach.
- The watermark is a pattern in the word-selection randomness itself, keyed to a watermark key plus preceding words. It adds no Unicode, metadata, hidden characters, or extra tokens, and is undetectable without the key.
- Paraphrasing or a complete rewrite defeats it, but light editing probably will not remove it fully.
- Detection is weaker in several cases:
- Small text samples with fewer words
- Factual content where word choice is constrained
- Proofreading edits that only touch grammar and punctuation
- Anthropic plans a watermark detection API, will use C2PA metadata for image files like JPG, PNG, and SVG, and says watermarking has trivial speed impact and no added token cost.
Published on