Google AI Mode is the most consequential change to search in a decade: a full conversational search experience, powered by Gemini, that answers complex questions with a synthesised response and a set of cited links rather than a page of blue results. Unveiled at Google I/O in 2025 and rolled out from Search Labs into the mainstream, it changes what winning looks like: the prize is no longer a ranking, it is a citation inside the answer. This guide gives you eleven concrete ways to earn those citations, each grounded in how it actually works, including its query fan-out technique, which quietly rewrites the rules of what content gets surfaced. It builds on the same principles as optimizing for Google’s inline summaries, but it is a deeper, more conversational surface with its own demands. No hand-waving, just what to change so your content is the source Gemini reaches for when it builds an answer. Treat it as the leading edge of where search is going, rather than a passing experiment, and the effort you put in now pays off as it captures more of the queries your buyers ask. The tactics that follow are the ones we apply for clients, and every one of them is something a disciplined team can put into practice without waiting for the dust to settle.

Quick Answer

To win citations in AI Mode, create genuinely useful, answer-first content that covers a topic and its related sub-questions deeply, because AI Mode uses a query fan-out technique that breaks each question into many sub-queries and pulls sources for each. Powered by Gemini, AI Mode synthesises an answer from multiple retrieved pages and cites a handful, so the goal is to be the trusted, extractable source across the cluster of questions behind a topic, not just the single head term. The highest-leverage moves are answering the full spread of related questions on a page, leading with clear direct answers, building strong E-E-A-T signals, adding structured data, and establishing unambiguous entity and topical authority. The eleven tactics below turn that into a practical plan you can act on this month.

How AI Mode works, and why query fan-out matters

Every tactic below follows from the mechanism, so it is worth understanding first. AI Mode is not a slightly smarter results page; it is a conversational reasoning layer built on Gemini. When you ask it something, it does not run one search. It uses a technique Google calls query fan-out: it breaks your question into many related sub-queries, runs them simultaneously across its index, retrieves sources for each, and synthesises the results into a single answer with citations. It also holds context across follow-up questions, so a conversation builds on itself.

That fan-out is the single most important thing to internalise. It means the system is not looking for the one page that best matches a keyword; it is assembling an answer from the best sources across a whole cluster of sub-questions. A page that deeply covers a topic and its related questions can be cited many times over in a single answer, while a thin page targeting one exact phrase is easily missed. This is why depth and topical completeness, rather than narrow keyword matching, are the foundation of winning here. The broader framework sits in our answer engine optimization guide, and the page-level habits overlap heavily with optimizing for AI Overviews; what follows is tuned specifically to AI Mode.

11 ways to win citations in AI Mode

These are ordered so the highest-leverage moves come first, but they reinforce each other. The theme running through all of them is that this surface rewards genuine topical authority expressed in a machine-legible way.

1. Answer the whole cluster of sub-questions, not one keyword

Because query fan-out pulls sources for many sub-questions at once, the biggest win is a page that thoroughly answers a topic and every question orbiting it. Map the sub-questions a buyer would ask around your core topic and answer each one clearly on the page or across a tight cluster of linked pages. A resource that covers the full spread can be cited repeatedly in one synthesised answer, which is the closest thing AI Mode has to a jackpot. This single shift, from targeting a keyword to owning a question cluster, is the defining tactic here.

2. Lead every section with a direct, self-contained answer

Gemini lifts passages that state the answer plainly. Open each section with the conclusion in a sentence or two, then elaborate, and make sure each passage makes sense on its own because that is how it will be extracted, stripped of surrounding context. Burying the point under preamble hands the citation to a competitor who got to it faster. This answer-first discipline is the highest-frequency habit that separates cited pages from ignored ones.

3. Make your entities unambiguous

The engine reasons over entities, people, brands, products, concepts, not just strings of words, so clarity about who and what you are is a direct ranking input. Define your brand and its key concepts explicitly, keep your naming consistent across the web, and use structured data to connect your entity to authoritative references. When Gemini can resolve exactly what your brand is and what it is expert in, it surfaces you far more reliably for the questions in your domain. Practically, that means a clear about page, consistent descriptions of your products and category across your own site and third-party profiles, and internal links that reinforce what each concept means. Entity ambiguity is a silent tax: every time the engine has to guess what you are, it is likelier to reach for a competitor it has already resolved cleanly.

4. Build genuine topical authority in clusters

Depth signals expertise, and the feature rewards expertise heavily because fan-out repeatedly probes a topic from many angles. Build hub-and-spoke clusters: a comprehensive pillar page supported by focused pieces on each sub-question, all cross-linked with descriptive anchors. This gives the engine a dense, well-connected body of sources to draw from, so wherever the fan-out reaches within your topic, it finds you. Isolated one-off posts cannot compete with a coherent cluster.

5. Earn strong E-E-A-T and trust signals

Gemini is cautious about which sources it puts behind an answer, so credibility is decisive. Show named authors with real credentials, cite primary sources, maintain a credible organisation presence, and earn corroboration from authoritative third parties. These E-E-A-T signals are often the deciding factor between being retrieved and being skipped, especially on topics where accuracy matters, and they compound with every other tactic here.

6. Add structured data and clean semantic HTML

Structured markup removes ambiguity about what your content is and how it is organised, making it easier for the model to parse and trust. Implement relevant schema.org types and pair them with clean semantic HTML: one H1, a logical heading hierarchy, real lists and tables. The specifics of which types earn the most are in our guide to schema markup that wins AI citations, and they apply directly to this surface.

7. Structure content for extraction

Fan-out and synthesis both favour content that breaks cleanly into parts. Use descriptive question-style headings, short paragraphs, numbered steps, and tables for comparable data, so every chunk is a self-contained unit it can lift and attribute. A comparison table or a crisp step list is often surfaced almost verbatim, making structured formats some of the highest-yield content you can add to a page.

8. Keep content fresh and accurate

For anything that changes, currency is an advantage, and accuracy is non-negotiable because a synthesising engine actively avoids sources it cannot trust. Maintain your key pages on a real cadence, update facts, and show honest dates. A current, correct page beats a stale one answering the same question, and it protects you from being replaced by a competitor who keeps theirs alive.

9. Optimise for conversational, follow-up intent

This is a conversation, not a single query, so it rewards content that anticipates the natural next questions. After you answer the main question, address the follow-ups a curious reader would ask, the comparisons, the caveats, the how-do-I-actually-do-this. Content that satisfies the whole line of inquiry is more likely to be cited across a multi-turn conversation, not just the opening question, which is where a lot of conversational-search usage actually happens.

10. Cover multiple formats and modalities

Gemini is multimodal, and the experience can weave text, images and other media into its answers. Support your written content with clear, well-labelled images, diagrams, and where relevant video, each with descriptive alt text and captions so the model understands them. Rich, well-described media gives the model more ways to surface you and can make your source the most complete answer to a question that benefits from a visual. A clearly labelled comparison chart, an annotated screenshot or a short explainer clip can be exactly the element that makes your page the richest source on a question, and rich sources are the ones a synthesising engine prefers.

11. Be present on the third-party sources it trusts

The engine does not only cite first-party pages; it draws on authoritative third-party sources and communities to corroborate and enrich answers. Genuine presence in the reputable publications and discussion spaces of your field increases how often Gemini encounters your brand as a trusted entity, which feeds back into how readily it cites your own pages. This off-site work is some of the highest-leverage optimization for any answer engine, this one included.

Across all eleven, the pattern is the same: win the topic, not the keyword; be clear and trustworthy; and make every passage easy to lift. Do that and query fan-out works in your favour instead of against you.

AI Mode vs AI Overviews: what is actually different

Teams reasonably ask how this differs from the inline summaries they are already optimizing for, because the advice overlaps. The honest answer is that AI Overviews and AI Mode are two points on the same spectrum, and the fundamentals, answer-first content, trust, structure, depth, serve both. But the differences shape priority. An AI Overview is a summary bolted onto a normal results page for a single query; the mode is a standalone conversational experience that reasons across many sub-queries and multiple turns.

That makes topical depth and question-cluster coverage matter even more here than in AI Overviews, because fan-out actively rewards breadth within a topic, and conversational context means a page that answers the follow-ups keeps getting cited as the dialogue continues. If AI Overviews taught you to write a clean, citable answer to a question, it asks you to write the clean, citable answer to a question and every question that naturally follows it. Optimize for the deeper surface and you are, by definition, covered for the shallower one, which is why we treat this as the more demanding standard to build toward. The diagnostic on the traffic drop after AI Overviews explains why sitting still is not an option on either.

Common mistakes that keep you out of the answer

Most failures here are not technical; they are strategic misreadings of how the feature works. These are the ones we see most often when a brand cannot understand why it is never cited.

• Targeting keywords instead of questions. Because fan-out probes a whole cluster of sub-questions, a page built around one exact phrase is easy to miss. Build around the question and everything that orbits it, not a single string.
• Thin, shallow coverage. A synthesising engine has no reason to lean on a page that only half-answers the topic. Depth is not optional; it is the mechanism by which you get cited repeatedly in one answer.
• Ambiguous entities. If Gemini cannot resolve exactly who you are and what you are expert in, it will surface a competitor it understands better. Inconsistent naming and missing entity signals quietly cap your ceiling.
• Burying the answer. A passage that hides its conclusion under preamble loses the citation to whoever stated it plainly. Lead with the answer, every time.
• Ignoring the conversation. Optimising only for the opening question and none of the follow-ups leaves most of a multi-turn exchange to competitors who anticipated the next step.
• Chasing it as a trick. There is no shortcut around genuine authority and clarity; the engine is explicitly built to reward substance, so gaming tactics fail fast here.

A workflow for building citable pages

Tactics stick when they become a routine, so here is the sequence we run on any page we want surfaced in conversational search. It turns the eleven ideas above into something a writer or editor can follow without having to remember the theory each time.

Start with the question map. Before writing a word, list the core question the page answers and the cluster of sub-questions a curious buyer would ask around it, including the natural follow-ups. That map is your outline, because it mirrors how fan-out will interrogate the topic. Draft answer-first. Write each section to open with a clear, self-contained answer to one of those questions, then elaborate, so every passage can stand alone when it is lifted. Add the trust and structure layer. Attach a credentialed author, cite primary sources, mark the content up with schema, and use clean semantic headings, lists and tables so the page is both trustworthy and easy to parse.

Make the entities explicit. Define your brand and key concepts clearly, keep naming consistent, and connect them to authoritative references so the engine can resolve exactly what you are. Enrich with media. Where a visual helps, add a well-labelled image, diagram or short video with descriptive alt text, giving the model more ways to surface you. Then measure and iterate. A few weeks after publishing, run the question cluster through the feature, note where you are cited and where a competitor won, and feed the gaps back into the next revision. Treated this way, optimization stops being a one-off effort and becomes a compounding routine: every new page ships in citable shape, and every existing one is pulled up to the same bar on a rolling basis. That discipline, more than any single clever tactic, is what steadily grows your presence in conversational search while competitors keep reacting after the fact. If you want a team to run it at scale, that is the day-to-day work of our answer engine optimization practice, backed by the tooling in our roundup of the best AEO and GEO tools.

Who should prioritise this now

Urgency varies by business, and being honest about that saves wasted effort. The brands that should move first are those whose buyers ask complex, research-heavy questions before they purchase: software and technology companies, professional services, healthcare, finance, and any considered purchase where people genuinely investigate before deciding. Those are exactly the questions a conversational engine is built to answer, so the citation is both winnable and valuable. If your audience already researches you and your category through AI assistants, the work pays back quickly.

For a simple local business or a transactional store, the calculus is gentler. The fundamentals, clear answers, trust signals and depth, still help, but the informational, research-led queries that this surface excels at are less central to how your buyers act, so it belongs on the roadmap rather than at the top of it. The sequence that works for almost everyone is the same one we recommend across every engine: get the content genuinely worth citing first, make it machine-legible, build the off-site authority, and measure. This surface simply raises the reward for doing that well, because a conversational engine that reasons across a topic rewards genuine depth more visibly than a ten-link results page ever did, and punishes thinness more obviously.

Turning this into a content roadmap

Reading a list of tactics is easy; turning it into a plan your team executes is where the value is created, so here is how to sequence the work over a quarter without boiling the ocean. The mistake most teams make is trying to optimise everything at once, which spreads effort so thin that nothing reaches the bar the engine actually rewards. A focused roadmap beats a broad one every time, because a synthesising engine cites depth, and depth is only possible when you concentrate.

In the first month, pick the three topics that matter most commercially and map the full question cluster behind each: the core question, the sub-questions, and the natural follow-ups a buyer asks on the way to a decision. That map becomes your content brief. In the second month, build or rework the pages to cover those clusters properly, answer-first, well-structured, credibly authored and marked up, so each topic is served by a coherent hub-and-spoke set rather than a scattering of thin posts. Resist the urge to add more topics until the first three are genuinely comprehensive, because half-covering six topics wins nothing while fully owning three can win a great deal.

In the third month, shift to authority and measurement. Pursue the off-site presence that corroborates your expertise, earn mentions in the reputable publications and communities of your field, and stand up the tracking that tells you whether your citations are growing. Then review, and let the gaps set the next quarter’s priorities: topics where a competitor is cited and you are not become the next clusters to build. Run this way, the work compounds, because every quarter deepens your authority on more of the questions your buyers ask, and authority is the currency this whole surface trades in. It is slower than a quick-win mentality wants, but it is the only approach that actually moves the needle, and it is exactly how the brands that dominate their category in conversational search got there. The alternative, chasing individual features with shallow content, produces motion without progress and quietly cedes ground to the competitors who chose depth instead.

There is a strategic upside hidden in the difficulty. Because thin, keyword-first content is punished more obviously on a conversational surface, the brands willing to invest in real depth face less competition than they did on a crowded results page. Where a hundred shallow pages once fought over one keyword, a synthesising engine quietly ignores most of them and rewards the few genuinely comprehensive sources. That asymmetry is an opportunity: the barrier to entry is effort and expertise, which most competitors will not sustain, so the teams that do can build a durable, compounding lead while the surface is still young and the field is still thin.

How to measure AI Mode citations

Winning is only meaningful if you can see it, and conversational search is harder to measure than a ranking, so a deliberate approach matters. Start manually by building a list of the conversational questions your buyers ask and running them through the feature, recording whether your brand is cited, for which sub-questions, and alongside which competitors. Because query fan-out means a single question surfaces many sources, note not just whether you appear but how often across the sub-answers, since that breadth is the real signal of topical strength.

Then systematise it with the same discipline we apply to every engine: treat citation share as a first-class metric tracked over time. The full method is in our guide to how to track your brand’s AI citations, and the wider scoreboard, from citation share of voice to AI referral traffic, is laid out in our piece on the AEO metrics that prove it is working. Pair that with an understanding of which sites AI answers cite most and you know both your progress and the benchmark you are chasing. Measurement here is young and imperfect, so track trends rather than obsessing over any single reading, and be transparent about the confidence with stakeholders.

The honest outlook on AI Mode

It is worth being straight about both the stakes and the uncertainty. It represents Google moving its core product toward conversational, answer-led search, and that is a profound shift with real consequences for how much traffic reaches the open web. Some informational clicks that used to be yours will simply not happen, because the conversation resolves the question in place. Pretending otherwise helps no one; the realistic prize on many queries is the citation and the brand authority it builds, plus the qualified click that sometimes follows, not a restoration of old traffic volumes.

At the same time, the details are still evolving, and anyone claiming a precise, guaranteed playbook is overselling. What is durable is the direction: search rewards genuine expertise, made clear and machine-legible, more than ever. The brands that invest in real topical authority and clean structure now will be the ones cited as it expands, while those chasing shortcuts will find there are none. That is why we frame it not as a standalone tactic but as the leading edge of a broader generative engine optimization strategy, run alongside resilient SEO for the high-intent queries that still send clicks. Build for the principle, not the feature, and you are positioned for whatever the feature becomes next.

Frequently asked questions

What is Google AI Mode?

AI Mode is Google’s conversational search experience, powered by Gemini, that answers complex questions with a synthesised response and cited links instead of a traditional list of results. It was unveiled at Google I/O in 2025 and rolled out from Search Labs, and it holds context across follow-up questions like a chat.

What is query fan-out and why does it matter for citations?

Query fan-out is the technique the mode uses to break a question into many related sub-queries, run them at once, retrieve sources for each, and synthesise an answer. It matters because a page that deeply covers a topic and its sub-questions can be cited many times in one answer, so depth beats narrow keyword targeting.

How do I get cited in AI Mode?

Cover a topic and its related sub-questions thoroughly, lead every section with a clear self-contained answer, make your brand and topic entities unambiguous, build strong E-E-A-T and structured data, and keep content fresh. The goal is to be the trusted, extractable source across the cluster of questions behind a topic.

Is AI Mode different from AI Overviews?

They are related but distinct. An AI Overview is a summary on a normal results page for a single query, while AI Mode is a standalone conversational experience that reasons across many sub-queries and multiple turns. The fundamentals overlap, but topical depth and follow-up coverage matter even more in AI Mode.

Does optimizing for AI Mode hurt my traditional SEO?

No. The two reinforce each other. The depth, structure, trust and freshness that win these citations also strengthen traditional rankings, and classic SEO still wins the high-intent, transactional queries that AI Mode rarely resolves in place. Doing both is the durable strategy.

Can I track whether AI Mode cites my brand?

Yes, though it is harder than tracking rankings. Start by running your buyers’ conversational questions through AI Mode and recording where and how often your brand appears across the sub-answers, then use an AI-visibility tool to monitor citation share over time as a first-class metric.