The paradox at the heart of the current crisis in digital publishing is now visible in granular form. While Google AI Overviews have devastated publisher traffic overall — reducing organic search referrals by 42% compared to pre-AIO baseline — one content category has not merely survived but accelerated: breaking news.

This finding, documented in a large-scale analysis by Define Media Group across the top 15 national and local news brands in their portfolio, inverts the intuitive expectation. One would assume that breaking news, the most easily summarizable content, would suffer most in an environment where AI systems increasingly synthesize and answer queries without human click-throughs. Instead, breaking news traffic is up 103% across all Google Surfaces since November 2024, while evergreen content — how-tos, explainers, reference material — is down roughly 40%.

The explanation lies in the technical constraints of large language models when applied to real-time events. Google AI Overviews have a visibility rate of only 15% for news content, nearly three times lower than for health or science information. This differential is not accidental; it reflects deliberate decisions by Google about where AI-generated summaries can operate safely and where they cannot.

The reason is latency. Large language models trained on historical data cannot reliably retrieve and synthesize information at the pace required by a breaking news event. A model trained weeks or months ago cannot accurately summarize a story that began 45 minutes ago. Beyond latency is the hallucination problem: all major LLMs remain capable of generating confident-sounding but factually incorrect information. In breaking news, where information drives financial, medical, and safety-related decisions, the consequences of an AI hallucination can be genuinely harmful.

This constraint has protected the Top Stories carousel — the Google Search feature where breaking news lives. A query like 'iran war' or any live developing event does not trigger an AI Overview. Instead, the Top Stories carousel appears, presenting headlines and images that actively encourage users to click through to the publisher's story. AI Overviews are completely absent from this result type.

The implications are profound. For news publishers facing a 42% decline in organic search traffic overall, and a forecast 43% drop by 2029, breaking news has become the sole reliable traffic engine. Publishers that can staff and resource real-time reporting — that can move fast enough to capture the first hours of a developing story — have a structural advantage. Publishers that have invested in evergreen content, the supposedly 'safe' long-tail strategy, have watched that investment evaporate.

The strategic question for newsrooms is no longer whether to do breaking news. It is whether they can afford not to. And whether the economics of real-time reporting, which requires expensive staffing and infrastructure, can be sustained by the traffic it generates.