Why did Google break AI Overviews for 'disregard'?

Google’s AI Overviews have been failing to return summaries for certain command-style words, including “disregard.” Multiple reports describe a pattern where the search experience interprets these terms less like normal query words and more like instructions—so the AI overview may refuse to answer or behave unpredictably instead of producing a helpful summary.

The problem matters because it points to a broader UX and safety tension in AI-first search: when interfaces mix conversational generation with keyword-style searching, the same token can be interpreted very differently depending on context. In traditional search, typing a word is straightforward; with AI Overviews, the system is effectively performing an additional “reasoning layer” before presenting results. If that layer treats a query like an instruction prompt, users can see failures even for simple informational lookups.

For users, the practical takeaway is that some single-word searches—especially those that resemble imperative commands—may not behave reliably in AI Overviews. That can increase time-to-answer and push people back to older search modes or alternative engines.

For the web ecosystem, the issue also underlines the fragility of a model-based summary layer: when it doesn’t understand or handle specific phrasing, it doesn’t just return poorer answers; it can stop working in a way that’s harder to diagnose than classic ranking glitches.

Overall, the “disregard” failure is a concrete example of why AI search reliability is still uneven, even as Google rolls out major Search changes aimed at making answers feel more conversational.