Conversational AI systems often exhibit safety hazards such as hallucination, sycophancy, overconfidence, and anthropomorphism, which are hard for users to spot during everyday interactions. We introduce Safety Nudges, a browser‑based lightweight tool that flags concerning behavior in‑situ when it is detected in chatbot conversations.\ \ We conducted a two‑week field study with 45 frequent chatbot users, gathering interaction logs, surveys, and feedback on individual nudges. Participants found the tool useful, clear, and minimally disruptive; almost all reported heightened awareness of potential AI harms. However, increased awareness alone did not consistently translate into observable behavioral changes.\ \ Our findings suggest that user‑facing safety nudges can complement model‑level safeguards by prompting users to critically evaluate AI responses in context. Effective nudge design must prioritize relevance, proper calibration, and user control to enhance conversational AI safety.\ \ Review: Safety Nudges highlights the promise of user‑side interventions, yet achieving concrete behavior change will require deeper integration of nudges with user decision‑making processes.