Tuesday’s stories share one move: someone asks for evidence and gets less than they came for. Publishers want a regulator to enforce a contract Google already signed, Search Console ships a report that withholds the two fields that would settle anything, a study reverse-engineers an index nobody documents, and Anthropic starts stamping its own output so it can be recognized later.
The Regulator And The Reporter: Two Disputes, Two Referees
APIG is asking France’s competition authority to rule on a contract Google already signed. A rival’s backdoor claim about WPForms Lite got a hands-on install instead of a verdict from either side.
- French Newspapers Say Google Broke AI Overviews Pay Deal APIG, representing close to 300 French newspapers, asked France’s competition authority on August 11 to rule on whether Google breached the 2022 compensation agreement by switching on AI Overviews in the country without publisher sign off. France’s competition authority already fined Google 250 million euros in 2024 over compliance gaps in that same deal. A complaint is an allegation, not a ruling.
- WPForms Lite Backdoor Claim From a Rival Draws a Hands-On Test A competing plugin developer alleged that WPForms Lite’s setup wizard hands its maker a short-lived admin access token on more than five million sites, a framing another commenter disputes since only an admin-triggered wizard is involved. A hands-on install of the plugin found an undisclosed redirect to a different domain during setup, plus bundled plugins installed with no way to opt out.
What The Report Won’t Show: A Rollout Without Queries, An Index Without Documentation
Search Console’s generative AI report just reached every property, minus the two fields that would explain it. A single consultancy’s read of 1,249 ChatGPT answers had to reverse-engineer what OpenAI’s index stores, because OpenAI hasn’t documented it either.
- Search Console’s Generative AI Report Goes Live For Every Property Search Console’s generative AI performance report now shows as live on every property checked, with no Google blog post confirming the wider rollout. It carries impressions, pages, countries and devices, but no clicks and no queries, the same gap our August 7 coverage flagged when AI Mode’s follow-up questions turned out to land in the standard Performance report instead.
- ChatGPT’s Free-Tier Index Doesn’t Reward an OpenAI Licensing Deal Resoneo’s read of 1,249 ChatGPT answers found sites without an OpenAI content licensing deal moved through the same internal search index as licensed partners. The index appears to store only about 200 characters per page, and 83.6 percent of the stored snippets Resoneo checked opened at the H1 instead of the meta description. The finding comes from a single consultancy’s analysis, not an independently replicated study.
The Mark That Outlives The Copy: Anthropic Stamps Its Own Output
Anthropic is now hiding a watermark inside everything Claude writes, worldwide, not just where EU law requires it. The company is explicit about what a detected mark proves, and what it does not.
- Anthropic Starts Watermarking Claude Text Worldwide, Not Just EU Anthropic has started embedding invisible watermarks inside text from its newest Claude models and attaching signed provenance metadata to generated image files, rolling out worldwide rather than only inside the European Union where the transparency rule originates. Anthropic says a detected mark means Claude may have processed the content at some stage, not that Claude authored it.
Today’s Quick Hits
- Google Trials Advertiser Roster Above Mobile Search Ads Google is testing a module that lists advertiser favicons and domains above a cluster of mobile sponsored results, spotted by a marketer. Google has not confirmed the test or its scope.
- New Google Ads Report Flags Missing Import Data A new Campaign Data Import Validation Report lets advertisers check imported non-Google campaign data for missing cost, click and impression figures before that data feeds into reporting.