This week delivered a tidy two-act tragedy for the open web. Act one: Google announced it will save your Lens photos, Search Live recordings, and Translate audio for AI training. Act two: The Atlantic reported that companies are now systematically gaming AI-generated search results, a practice the piece memorably calls sloptimization. Put them together and you have a closed loop: Google trains its models on human queries, those models surface AI-optimized content, and the humans querying the system are increasingly interacting with outputs shaped by the previous generation of the same system.
The Sloptimization Industrial Complex
The Atlantic piece describes a new SEO industrial complex built not for human readers but for chatbot crawlers. Where the original SEO game was about keywords and backlinks, sloptimization is about feeding the machine exactly the tokens it expects. The web stops being a space of human expression and becomes a hall of mirrors. This has a direct material consequence for Google's AI training pipeline: the data it is now explicitly saving from your searches is increasingly data generated by interaction with AI-generated content. The signal is eating the noise that the signal produced.
Privacy Settings as Cultural Symptom
The privacy framing matters here. Google couches the data collection as a feature users can adjust, but as Brewster Kahle has argued in the context of the public internet, the structural choices made by platforms about what gets stored and surfaced determine what culture we can even access. When the archivist is also the advertiser is also the AI trainer, the public interest in open access and the private interest in data extraction are not just in tension. They are in active conflict. The agentic web, as a 2026 arXiv paper on normative infrastructure puts it, requires entirely new governance frameworks that do not yet exist.