AI Mode dropped citations after the Gemini 3.8 Flash update

Google’s newest AI Mode model started dropping the source citations from some of its answers. On 2 September 2026 Google put its Gemini 3.8 Flash model into AI Mode, and within hours SEOs reported that some broad top-of-the-funnel queries were coming back with far fewer citations than before, in many cases none at all. Google called it a bug and rolled out a fix within days, but the episode is a plain reminder that the visibility multi-location brands earn in AI answers rests on behaviour Google can change without warning.

What happened

Google announced Gemini 3.8 Flash on 2 September 2026 and rolled it into AI Mode the same day. In its announcement Google said 3.8 Flash is available to Google AI Pro and Ultra subscribers across the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.

Soon after the rollout, search analysts testing AI Mode reported that the model was returning answers with far fewer citations than before, and in many cases none, for broad informational queries. Robby Stein, Google Search’s vice president of product, acknowledged the missing citations on X, describing the behaviour as not working as intended and saying a fix would roll out soon.

Google called the behaviour unintended rather than a decision to stop linking to sources, and it surfaced right after the model rollout. Google rolled out a fix in the days after, and source links began reappearing in AI Mode answers.

Why it matters

Citations are the mechanism through which a brand earns visibility and clicks inside an AI answer. When AI Mode names and links its sources, the businesses behind those pages get seen and can get visited. When it answers from the same underlying data but shows no link, that referral path narrows to nothing, and the site owner has no dial to turn.

AI Mode is not a fringe surface where this is a small effect. It is now one of the most-used surfaces in Google Search, so a citation change there moves brand visibility at the scale of the whole audience at once. The unsettling part for anyone planning a strategy around it is how the change arrived: not through a policy update or an algorithm announcement, but in step with a change to the model that writes the answers. What gets cited, and whether anything gets cited, can move with a release most brands never hear about.

What this means for multi-location brands

For a central team responsible for hundreds or thousands of locations, the lesson is not to chase this specific bug, which Google has since fixed, but to treat AI visibility as a monitored outcome rather than a stable given. Track how the brand and its locations surface in AI answers over time, across markets, so a drop like this shows up as a measured signal instead of a mystery slump in traffic. Places AI is built around exactly that question of how a brand’s presence appears inside AI-generated answers, rather than treating the AI surface as a black box.

The durable defence is the same as it always was. A brand that keeps its location data accurate and consistent everywhere, and writes its location and service pages as clear, answer-first paragraphs, stays the most citable source whenever citations return. That accuracy cannot be maintained profile by profile across a large estate. Managing one source of truth centrally, then pushing it out through local business listing management, is what keeps every location eligible to be cited in the first place, and the structural work of shaping content so AI systems can extract it is the generative engine optimization discipline that improves the chance a page gets quoted at all.

3.8 Flash is available to Google AI Pro and Ultra subscribers across the Gemini app, AI Mode in Google Search and Gemini in Google Sheets.

Google, Gemini 3.8 Flash announcement, 2 September 2026

The bottom line

Google moved quickly to restore citations in AI Mode, and this particular gap is closing. What it exposes is structural: brand visibility in AI answers depends on choices Google makes model by model, and no brand controls them. The response is not to react to each swing but to own the fundamentals that survive them, accurate data and citable content across every location, and to watch the AI surface closely enough to notice the next time it moves.