A testing tool arrives, a citation share collapses, a data promise holds
The day's 3 verified stories: Google added multi campaign experiments and new planning tools to AI Max; Reddit's share of ChatGPT citations fell 86 percent in three weeks; OpenAI previewed a safety layer that works without breaking zero data retention on frontier models. The full read of each story, its sources and what it means for your business live on its own page.
Google adds multi campaign experiments and new planning tools to AI Max
Per a Google announcement dated August 20, 2026, AI Max for Search campaigns is gaining three new capabilities. Advertisers will be able to test different budgets and return on investment targets across multiple Search campaigns in a single A/B test, a capability arriving in September. AI Max experiments can now run with brand and location controls enabled, so testing no longer requires removing those guardrails. Performance Planner shows how bidding and budget target changes may affect existing campaign performance and can apply suggestions in one click. Search Engine Land confirmed the news the same day; the announcement shares no performance statistics and no customer examples.
Reddit's share of ChatGPT citations fell 86 percent in three weeks
Per data from the visibility tracker Promptwatch covering July 7 to August 17, 2026, reddit.com's share of ChatGPT Search citations fell from a 3.83 percent average between July 18 and August 7 to 0.52 percent across August 14 to 17. The mechanism pointed to is an August 8 jump in the use of site: operators from 0.37 percent to 16.8 percent, though as Search Engine Journal stresses the timing gap weakens that explanation. Over the same window the decline was more gradual on Google surfaces, 11.3 percent in AI Overviews and 30.5 percent in AI Mode. Per Forbes, OpenAI did not respond to a request for comment.
OpenAI previews a safety layer that keeps zero data retention on frontier models
Per an OpenAI announcement dated August 19, 2026, the company continues to promise eligible API customers that prompts and responses are not retained, and is previewing a layer called Private Safety Processing. The system is designed to identify risky patterns across related interactions without giving personnel access to the underlying content. In ZDR deployments content stays on infrastructure the customer controls, and when a risk is identified OpenAI receives only a narrowly defined signal indicating the type of activity. Rollout and a technical white paper are planned for September. Per The Next Web, consumer tiers are outside the scope.
The agenda moves fast, the system stays calm.
Let us talk through what these changes mean for your business and which step fits you.