A reasoning model that works through steps before answering — Gemini's shift toward agentic tasks.
Gemini 2.5 Pro arrived as Google’s first “thinking model” — a reasoning model that works through steps before answering, marking Gemini’s shift toward agentic, multi-step tasks. The move from models that answer in one pass to models that reason through a problem is a genuine inflection: it lets complex, multi-step queries be resolved inside the assistant, and it changes what content must offer. For brands, the arrival of thinking models means depth and structure that survive multi-step reasoning become the qualities that keep you present in the answers these models produce.
A thinking model is one that reasons through steps before producing an answer, rather than generating a response in a single pass. Gemini 2.5 Pro, Google’s first such model, works through a problem — breaking it down, reasoning about its parts, and building toward a considered answer — which lets it handle complex, multi-step questions more capably than a model that answers immediately. This deliberate, step-by-step reasoning is what distinguishes a thinking model, and it marks a shift in what AI can do well.
The significance is that thinking models represent a move from answering to reasoning, which is Gemini’s pivot toward agentic, multi-step tasks. A model that reasons through steps can tackle complex questions that a one-pass model would fumble, resolving them inside the assistant rather than requiring manual research. For brands, the arrival of a thinking model means more complex, multi-step questions can be answered in the assistant, drawing on sources the model reasons over — a new kind of visibility opportunity on harder questions.
The introduction of thinking models marks an inflection in AI capability, from models that retrieve and rephrase to models that reason through problems. Gemini 2.5 Pro being Google’s first thinking model reflects this shift, and it signals a direction toward agentic systems that plan and execute multi-step tasks rather than responding in one shot. This is the beginning of a broader move toward AI that reasons and acts, of which thinking models are an early, foundational step.
This matters because reasoning changes which questions resolve in the assistant. A model that reasons through steps can handle the complex, multi-step research that a one-pass model could not, which means more of that research happens in the assistant, drawing on the sources the model reasons over. For brands, the shift to reasoning reframes the visibility opportunity toward the complex questions thinking models can now handle — and toward the depth and structure that serve a reasoning process.
The mechanic that matters is how a thinking model handles a complex, multi-step question. It works through the problem — decomposing it into parts, reasoning about their relationships, retrieving relevant information for each, and building toward a considered answer — a more deliberate process than a one-pass response. This step-by-step reasoning lets it tackle complex questions capably, reaching into sources across the facets of a problem and reasoning over what it finds to construct a coherent answer.
For brands, the key implication is that a thinking model reaches into content across the steps of its reasoning and rewards depth and structure that serve that process. Content that answers only the surface of a topic is quickly exhausted as the model reasons; content with genuine depth and clear structure keeps providing relevant information across the reasoning steps. Depth and structure that survive multi-step reasoning become the qualities that keep content useful to a thinking model tackling complex questions.
The defining implication of thinking models is that depth and structure survive multi-step reasoning. When a model reasons through a complex question, working across its facets, shallow content that covers only the obvious is exhausted quickly, while deep, well-structured content keeps providing relevant, reliable information across the reasoning steps. A thinking model rewards content with genuine depth and clear structure, because that is what remains useful as it reasons through a problem toward a considered answer.
For brands, this argues for comprehensive, well-structured content on complex topics — genuine depth organized so a thinking model can navigate and draw on it across its reasoning, rather than thin pages built to intercept a simple query. Content that covers a topic thoroughly, clearly structured, is what survives the multi-step reasoning a thinking model performs. Depth and structure are the qualities that keep you present in the answers thinking models produce on the complex questions they now handle.
A thinking model works through a problem step by step, reaching across sources. Shallow content is exhausted fast; genuine depth and clear structure are what keep you present in the answers a reasoning model builds.
The core implication is that thinking models let complex, multi-step research resolve inside the assistant, and the sources a reasoning model draws on for those questions shape the answers. Visibility on complex questions — being a source the model reasons over — is a specific opportunity, and it rewards depth and structure over thin content. A complete visibility strategy accounts for the complex questions thinking models can now handle, ensuring your content survives the multi-step reasoning they perform.
This makes depth on complex topics a visibility imperative. Auditing whether Google’s AI answers cite you on the complex, multi-step questions in your category, understanding who is cited instead, and building the depth and structure that survive multi-step reasoning is the work. Gemini 2.5 Pro’s arrival as a thinking model is a prompt to ensure your content holds up under the reasoning these models perform, because the complex research they now handle draws on the sources with genuine depth.
For brands, thinking models reward genuine depth and clear structure on complex topics, an invitation to compete on substance. The fundamentals apply — answer-first, evidenced, structured, entity-clear content — but the emphasis is on depth and navigable structure: covering a topic thoroughly, organized so a thinking model can draw on it across its reasoning. Thin content that answers only a simple query is less suited to the complex questions thinking models handle, while deep content survives their reasoning.
This rewards brands that build genuinely comprehensive, well-structured resources on complex topics, because those resources are what a thinking model reaches into as it reasons. A deep resource covering a topic’s facets, clearly organized, is what survives multi-step reasoning and keeps you present in the answers thinking models produce. Brands that invest in depth and structure are positioned for an assistant powered by thinking models that handle the complex, high-value research shaping decisions.
Understanding how a thinking model selects sources sharpens the approach. It reasons through complex questions across their facets, favoring content that is relevant, authoritative, and — crucially — deep and well-structured enough to serve a multi-step reasoning process. Shallow content is exhausted quickly; deep, clearly-structured content keeps providing relevant information as the model reasons. Evidence and clarity help the model trust and navigate what it draws on across its reasoning steps.
Depth and structure are especially decisive because a thinking model reasons more deeply than a one-pass model. For brands, this argues for comprehensive, well-structured, evidenced content on complex topics — the combination that survives multi-step reasoning. Being the deep, navigable, credible source is what keeps you cited in the answers thinking models produce, which rewards genuine substance over the thin content that might suffice for a simple, one-pass query.
Thinking models redistribute visibility on complex questions toward brands with genuine depth and clear structure. The winners are those whose content covers a topic thoroughly, organized so a thinking model can navigate it across its reasoning, because that depth survives the multi-step reasoning the model performs. The losers are brands with thin content that is quickly exhausted on complex questions, and those relying on surface-level pages that a thinking model reasons past in favor of deeper sources.
The determining factor is depth and structure — whether your content holds up under the multi-step reasoning a thinking model performs. Brands that invest in comprehensive, well-structured resources are positioned to win on complex questions; those with shallow content cede that ground. As thinking models handle more complex research in the assistant, the value of genuine depth rises, rewarding brands that cover their topics thoroughly and penalizing those with thin content that cannot survive multi-step reasoning.
Thinking models resolve hard, multi-step questions in the assistant — drawing on the sources with real depth. DUNkē tracks whether your brand is one of the sources Google’s AI answers cite, across its surfaces and seven other engines, per prompt and against competitors, so you know if you hold up on complex queries.
The response to thinking models is to build the depth and structure that survive multi-step reasoning on your complex topics. Start by auditing whether Google’s AI answers cite you on the complex, multi-step questions in your category and who is cited instead — a baseline revealing whether your content holds up under the reasoning these models perform. That audit often shows gaps on exactly the complex, high-value questions where depth matters most.
From there, the work is building genuine depth: create comprehensive resources covering a topic’s facets thoroughly, organized with clear structure a thinking model can navigate; lead with clean, quotable, evidenced answers; ensure your entity is clearly defined; and keep content fresh. Then measure whether your presence on complex queries improves and double down on what works. Building depth and structure that survive multi-step reasoning is how brands stay visible on the complex research thinking models now handle.
Situating Gemini 2.5 Pro clarifies its place. As Google’s first thinking model, it marked the start of Gemini’s reasoning capability, reaching billions through Google’s surfaces and grounded in its index. Other engines have advanced their own reasoning models; the broader direction across AI is toward models that reason through steps. Gemini 2.5 Pro was an early, foundational step in this shift for Google, introducing the thinking-model paradigm that subsequent models would build on.
For brands, the takeaway is that the shift to reasoning rewards depth and structure across engines, so investing in genuine depth serves visibility broadly. The deep, well-structured content a thinking model rewards is what other reasoning models reward too, because they all reason more deeply than one-pass models. Gemini 2.5 Pro introduced the thinking-model paradigm for Google, and the depth-oriented approach it rewards applies across the reasoning models the industry is building — one more reason to invest in depth now.
Gemini 2.5 Pro marking Gemini’s shift toward agentic tasks is worth dwelling on, because reasoning is a foundation for agentic capability. A model that can reason through steps is on the path toward planning and executing multi-step tasks — the essence of agentic AI. Gemini 2.5 Pro as a thinking model is thus an early step toward agents that reason and act, which points to a future where the assistant does not just answer but accomplishes multi-step tasks on a user’s behalf.
For brands, the agentic-shift dimension reinforces that depth, structure, and machine-legibility are becoming more important, because agentic systems that reason and act need content they can navigate and use across multi-step processes. As thinking models evolve toward agents, being the deep, well-structured, legible source that serves multi-step reasoning and action becomes valuable. Gemini 2.5 Pro’s introduction of reasoning is an early marker of this agentic direction, which argues for investing in depth and structure now.
There are genuine uncertainties around thinking models. How they select and reason over sources on complex questions, how their citation behavior evolves, how much complex research shifts into the assistant, and how the agentic capabilities they foreshadow develop are all open questions. There are also broader considerations about how reasoning models affect engagement with sources and how being cited on complex queries translates to demand.
For brands, though, these uncertainties do not change the fundamental calculus. Thinking models like Gemini 2.5 Pro power the complex answers a share of customers get in the assistant, and being citable — through depth and structure that survive multi-step reasoning — pays off regardless of the details, because it rests on qualities every reasoning model rewards. The concrete risk is not that the models change; it is having thin content that cannot survive multi-step reasoning on complex questions. That risk is addressed by building genuine depth now.
The developments to track are the ones that signal how reasoning reshapes search: how thinking models advance, how much complex research shifts into the assistant, how their citation behavior on hard questions evolves, and how the agentic capabilities they foreshadow develop. Each will tell you how much of your customers’ complex research is mediated by thinking models and how the rules of being cited on hard questions are taking shape. The through-line is that thinking models reward depth and structure that survive multi-step reasoning.
For your own program, watch your presence on complex, multi-step queries in Google’s AI answers over time, per prompt and against competitors, as the measure of whether your content holds up under a thinking model’s reasoning. Gemini 2.5 Pro’s arrival is a reason to establish that baseline and invest in depth, and any gaps are a signal to act — because on complex questions thinking models now handle, thin content that cannot survive multi-step reasoning is a visibility gap on the queries that matter most.
The deepest way to read Gemini 2.5 Pro is as the point where AI begins to think — to reason through problems step by step rather than answer in one pass, opening the door to handling complex questions and, eventually, agentic tasks. This is a genuine inflection, because reasoning is what lets the assistant tackle the hard, multi-step research that shapes important decisions. Gemini 2.5 Pro as Google’s first thinking model is an early marker of AI moving from answering to reasoning.
That reframe is the strategic takeaway. As AI begins to think, being the deep, well-structured source that survives multi-step reasoning becomes the visibility discipline on complex questions, rewarding brands that cover their topics thoroughly. Gemini 2.5 Pro introducing reasoning is a signal to invest in genuine depth, because the complex research thinking models handle draws on the sources with real substance — and the brands that build that depth are positioned for an assistant that reasons, and eventually acts.
Winning complex queries under a thinking model comes down to building genuine depth in practice: comprehensive resources covering a topic and its facets thoroughly, so that as the model reasons through a multi-step question, your content keeps providing relevant information rather than being exhausted after the obvious point. Depth is what remains useful across the reasoning steps, while shallow content runs out quickly. The practical work is identifying the complex questions in your category and building content deep enough to serve them.
For brands, this means investing in thorough coverage of complex topics — not thin pages targeting a single query, but comprehensive resources a thinking model can draw on repeatedly across its reasoning. Content that covers a topic’s facets in real depth is what survives the multi-step reasoning a thinking model performs. Building depth that survives reasoning is the practical discipline for staying present in the complex answers thinking models produce.
Depth needs navigable structure to be useful to a thinking model reasoning across a problem. Clear headings mapping a topic’s facets, self-contained sections answering specific sub-questions, and logical organization help a thinking model locate and use the right piece of your content as it reasons. Well-structured depth is more useful than depth buried in disorganized prose, because the model can navigate it to find what each reasoning step needs.
For brands, the practical goal is content that is both deep and clearly organized — comprehensive coverage a thinking model can navigate. Disorganized depth, where relevant information is hard to locate, is less useful than clearly-structured depth. Structuring your comprehensive content so a thinking model can navigate it is what makes that depth survive multi-step reasoning and keep you present in the answers thinking models build.
A practical opportunity thinking models create is targeting complex, high-value questions — the multi-step questions that often precede important decisions and carry the highest-value customers. Identifying the complex questions in your category and building the depth to be the source a thinking model cites on them positions you at decisive moments. Being present on the hard questions, where serious research happens, is disproportionately valuable.
For brands, this argues for deliberately targeting the difficult questions, not just the high-volume simple ones. The depth and structure that win complex queries reach customers doing serious research, and being cited there positions you at the point of decision. Targeting complex questions rewards brands that invest in genuine depth on the hard questions a thinking model now handles, capturing visibility with high-value customers.
Investing in depth for thinking models also prepares you for the agentic future they foreshadow, because reasoning is a foundation for agentic capability. A model that reasons through steps is on the path toward planning and executing multi-step tasks, and the deep, well-structured content that serves reasoning is also what agents need to navigate and act across multi-step processes. Building depth now positions you for both the reasoning present and the agentic future.
For brands, this means the investment in depth and structure is doubly worthwhile: it wins the complex queries thinking models handle today and prepares your content for the agents these models evolve toward. As thinking models become agentic, the deep, navigable, machine-legible content that serves reasoning serves agents too. The agentic future depth prepares for is a reason to invest in genuine depth and structure now, ahead of the agentic capabilities thinking models foreshadow.
Thinking models reward substance over tactics, which is a strategic shift worth internalizing. Because a reasoning model works through a problem deeply, it rewards content with genuine depth and rigor, while thin content built to game a simple query is exhausted or reasoned past. This favors a substantive content strategy — genuinely covering complex topics well — over tactical optimization aimed at intercepting individual queries.
For brands, the implication is to invest in genuine substance on complex topics rather than tactical tricks, because reasoning rewards depth that holds up under scrutiny. The brands that build genuinely deep, rigorous content are positioned to win as thinking models handle more complex research, while those relying on thin, tactical content are reasoned past. Reasoning rewarding substance over tactics is a reason to prioritize genuine depth, aligned with building real authority on your topics.
Gemini 2.5 Pro arriving as Google’s first thinking model — a reasoning model that works through steps before answering, marking Gemini’s shift toward agentic tasks — means complex, multi-step questions can resolve inside the assistant, drawing on the sources the model reasons over. For brands, the arrival of thinking models means depth and structure that survive multi-step reasoning become the qualities that keep you present in the answers these models produce, rewarding genuine substance over thin content.
The right response is to build the depth that holds up: audit whether you are cited on the complex questions that matter, create comprehensive, well-structured resources that survive multi-step reasoning, and measure your presence on complex queries as rigorously as on simple ones. The brands that invest in genuine depth and clear structure — the qualities thinking models reward — are the ones that stay visible on the complex research the assistant now handles, while thin content is reasoned past on exactly the queries that decide the most.
“When AI begins to reason through problems step by step, shallow content is exhausted fast. Genuine depth and clear structure are what survive multi-step reasoning — and keep you present in the answers thinking models build.” The Age’X Channel Desk
Get a free GEO audit — the same analysis behind every article here.