From Reliable Dictionary Cards to Unpredictable AI Overviews
For years, typing a single word into Google Search reliably pulled up a structured dictionary card with clear definitions. That basic reference feature has quietly been handed over to Google AI Overviews, and for some queries, the swap is going badly. Instead of serving a licensed dictionary definition, Search now routes certain word lookups through an AI model that generates a conversational answer in the same top slot. The result: users seeking quick vocabulary checks are greeted with freeform sentences as though they were chatting with a bot. Crucially, the dictionary card often doesn’t appear alongside the AI response, so the deterministic definition is effectively displaced. What used to be a straightforward database lookup has become an open-ended generation step, undermining one of Google Search’s simplest and most frequently used utilities.
When ‘Ignore’ Means “Understood” Instead of “To Refuse to Pay Attention”
The most visible failures involve action verbs that look like commands. Users who type “ignore,” “disregard,” or “dismiss” into Search expect to see definitions first. Today, AI Overviews sometimes treats those dictionary queries as instructions. Instead of defining “ignore,” the system responds with lines such as “Understood. I have disregarded your previous message,” mimicking a chat assistant resetting context. Variations like “ignore synonyms” are also misread: rather than listing alternatives for “ignore,” the AI cheerfully claims it will avoid using synonyms and invites the user to explain their work. Similar misfires appear for words including “remember,” “start,” “finished,” and “forget.” Only a subset of action terms are affected, and other vocabulary still returns normal definitions, but these glitches are enough to disrupt everyday workflows built on quick word lookups.
A Symptom of AI Misinterpretation: Command vs. Search Intent
These dictionary lookup issues expose a deeper design flaw: Google AI Overviews struggles to separate search intent from natural language commands. The trigger words share a grammatical shape—short imperatives that read naturally as instructions for a chatbot. Even without prior conversation, the model defaults to its instruction-following posture, obeying the word as a directive rather than treating it as a term to define. That behavior might be acceptable inside a dedicated chat interface, but in Search it breaks expectations. A slot that once behaved predictably now varies by model output and context, and two people entering the same word may not see the same response. By replacing a deterministic dictionary module with a generative AI layer, Google has introduced ambiguity where users historically expected rigid, reference-grade consistency.
Why This Feels Like a Regression in Core Search Functionality
On the surface, this might look like a quirky edge case. In practice, it represents a regression in Google Search’s core search functionality. Dictionary lookups are a foundational habit—quick, low-friction checks for spelling, usage, or meaning. Now, for certain high-frequency verbs, that routine is interrupted by chat-style replies that neither define the word nor acknowledge the user’s intent. Because the dictionary card is replaced rather than supplemented, the AI misinterpretation leaves no obvious fallback on the same screen. It also undermines trust: people may hesitate to rely on AI Overviews for basic facts if it cannot reliably handle single-word queries. A feature used billions of times has become less dependable precisely because a generative system was bolted onto what used to be a simple, structured reference experience.
Google’s Response and What Needs to Change Next
After users on X highlighted the strange responses to words like “disregard” and “ignore,” others quickly reproduced the behavior across devices and accounts, confirming it wasn’t an isolated glitch. Google has acknowledged that AI Overviews are misinterpreting some “action-related queries” and says a fix will roll out soon, though it has not detailed which verbs are affected, how the patch will work, or when definitions will reliably return to the top of results. The key challenge is not just patching a list of problematic words, but improving the routing logic and model behavior so Search can robustly distinguish dictionary lookups from chat commands. Until Google restores a dependable dictionary path—ideally with deterministic cards prioritized over generative replies—its flagship AI overlay will continue to feel like it’s breaking something Search already solved.