"ray peat aspirin thyroid" and "What did Ray Peat say about" are two very different things when it comes token prediction. Apart from the prompt, the Gemini vs Grok result looks more like a labelling/curation or class/reward failure than a specific attempt at malice.
Google is making a mistake by attempting to integrate this with traditional search in my opinion but they really have no other way to push it and they need interaction to mine for improvement. xAI has Twitter, closedAI has the public memory for popularising it. People are probably more familiar with DeepSeek than Gemini for the competitive coverage.
Google and Apple remain concerning for attestation in their mobile operating systems and the push to have us all scanning digital papers.