Gemini 3.5 lands: what changes for brand citations

What happened

Per the Google AI blog, Google has shipped Gemini 3.5, positioning the release as "frontier intelligence with action." The framing matters more than the version bump. Google is no longer selling a model. It is selling a system that decides, retrieves, and executes on behalf of the user.

The headline shift is agentic. Gemini 3.5 is being marketed as a model that takes action, not one that simply answers. That changes what gets cited, when, and why. When a model executes a multi-step task, the surfaces it touches (search, tools, structured data, partner APIs) become the new battleground for brand visibility. Static "best of" listicles lose ground. Sources the agent trusts mid-task gain it.

Google's positioning is also a direct response to OpenAI's agent push and Anthropic's Claude tool-use traction. The competitive frame is no longer who writes the best paragraph. It is who completes the task.

Why it matters for your brand

For B2B brands targeting financial services, multilaterals, industrial groups, and policy institutions, Gemini 3.5 collapses two distinct visibility problems into one.

The first problem: getting cited in an answer. The second: getting selected as a trusted action surface. A bank's research page might be cited in a Gemini answer about inflation forecasts. The same bank's API or rate calculator might be invoked by an agentic Gemini flow when a user asks it to model a mortgage scenario. Those are two different content products, ranked by two different signals, but increasingly served by the same model. If your team is only optimising for the first, you are losing the second by default.

For financial services specifically, this means structured data is now a brand asset. Gemini 3.5's agentic mode will preferentially call sources that return clean, machine-parseable outputs. A wealth manager whose insights pages render as PDF blobs will be invisible to an agent that needs JSON. A reinsurer whose catastrophe data is locked behind a login will be skipped for one whose public dataset is indexable. The communications team and the data team now share a KPI.

For multilaterals and the UN system, the implication is sharper. Gemini 3.5 will be the default model for millions of users asking policy questions: climate finance flows, refugee statistics, SDG progress. If UNDRR, the World Bank, or WHO are not the agent's first call for their own domains, a secondary aggregator will be. Multilaterals have historically published authoritative data and let intermediaries package it. That model breaks when the intermediary is an LLM that prefers the cleaner feed. The institutional brand value of being the source erodes if the agent does not recognise you as the source.

For major industrial groups, the agentic shift changes procurement-adjacent content. When a buyer asks Gemini to "compare low-carbon cement suppliers in EMEA with shipping availability to Lagos," the model is no longer surfacing a Forbes article. It is querying supplier specifications, sustainability disclosures, and logistics data. Holcim's competitive position in that answer depends on whether its product data is exposed in a format Gemini's agent can use. Marketing budget that goes to brand films will not move that needle. Budget that goes to structured product content, sustainability disclosures in machine-readable formats, and partner integrations will.

For philanthropic and policy institutions, the risk is dilution. Gemini 3.5's action layer will increasingly summarise positions rather than quote them. A foundation that publishes a 60-page report on financial inclusion will see its position absorbed, paraphrased, and acted upon without the foundation being named in the user-facing surface. The defence is not louder publishing. It is making the foundation's framework so embedded in the underlying data that any agentic answer in the category routes through it.

The signal in context

Every major model release in 2024 and 2025 has pushed in the same direction: from answer to action. OpenAI shipped Operator. Anthropic shipped computer use. Perplexity shipped Comet. Gemini 3.5 is Google's most aggressive entry in that race, and because Gemini is wired into Search, Workspace, and Android, it has the largest default distribution of any agentic model on the market. The citation patterns that mattered in the GPT-4 era (being quoted in a paragraph-length answer) are now one of several visibility surfaces. The others (being called as a tool, being read as structured data, being trusted as an action target) are growing faster.

The brands that will hold visibility through the next 18 months are the ones that treat their content estate as a machine-readable product, not a publishing programme. That reframe is uncomfortable for communications teams who built their authority on long-form thought leadership. It is also unavoidable. Gemini 3.5 has made the cost of ignoring it explicit.