How to build content that AI search engines actually cite

The key to AI search optimisation isn’t better writing. It’s a better system. Content that AI search engines cite is built through a structured process: organised around real customer questions, grounded in documented expertise, and designed so every section works as a standalone answer an AI engine can extract and reference. The difference between content that gets cited and content that gets ignored isn’t quality alone. It’s architecture. This piece covers the full picture: how to build an AI content marketing process that earns citations , how to earn AI citations through offsite engagement on Reddit, LinkedIn, and Quora , and how to run a GEO audit so AI search engines can actually find your content .

Generative AI traffic to US websites jumped 1,200% between July 2024 and February 2025 (Adobe Analytics ). That’s not a trend you can wait out. That’s a structural shift in how people find and consume information. The content that earns the citation is whichever source AI selects from its options, and AI is selective. If your content doesn’t meet its criteria for structure, expertise, and extractability, it moves on to someone else’s.

I’m Owen Steer , and at Fifty Five and Five I’ve been building AI content systems for companies like Quisitive and Avalara. The process covers everything from keyword research through to writing and editorial, and I’ve seen firsthand what separates content that gets cited from content that doesn’t. The pattern is consistent: companies that treat content as a system (with documented expertise, structured sections, and both onsite and offsite layers) earn citations. Companies that just write good blog posts and hope for the best get passed over.

GEO optimisation: why ranking isn’t the finish line anymore

GEO optimisation (Generative Engine Optimization) is the practice of structuring your content so AI engines cite, reference, or recommend it when answering questions. It builds on SEO. It doesn’t replace it. Your technical SEO foundation still matters. But ranking on page one of Google is no longer enough if AI engines are answering the question before anyone clicks through to your site.

The researchers at Princeton and Georgia Tech who coined the term GEO found that applying GEO techniques can boost visibility in generative engine responses by up to 40% (Aggarwal et al., 2023 ). That’s a meaningful shift, and it applies directly to how B2B companies should think about their content investment.

How AI overviews decide which sources to cite

Google’s AI Overviews don’t just pull from whatever ranks highest. They apply a separate set of filters on top of traditional ranking signals: content extractability, E-E-A-T strength, freshness, and whether the page answers the question directly in its opening lines. A page can sit at position one in organic results and still get skipped by AI Overviews if its answer is buried three paragraphs deep, hidden behind a gate, or lacks clear author attribution. I’ve seen this with clients who were celebrating page one rankings while their content was completely absent from AI Overviews on the same queries. The selection process is structurally different from traditional organic results, which is why optimising for one doesn’t automatically mean you’re optimised for the other. If you’re ranking but not getting cited, this gap between traditional SEO performance and AI source selection is almost certainly where the problem sits.

What’s changed about how people search is central to why GEO matters. AI queries tend to be longer and more conversational than traditional Google searches. Instead of typing “AI SEO tips,” people ask full questions: “How do I build content that AI search engines actually cite?” The content that answers those full questions, clearly and with structure, is what gets selected. Content that buries the answer or assumes the reader will scroll to find it gets skipped.

So what actually changes when you optimise for GEO versus traditional SEO?

• Answer-first structure: Every section opens with a direct answer. AI engines extract snippets from the beginning of content blocks. If your answer is buried in paragraph four, AI has already moved on to a source that leads with it.
• Standalone sections: Each part of your content must make sense if pulled out in isolation. AI might cite your third H2 without any of the surrounding context.
• Open access: Gated content is essentially invisible to AI crawlers. If the engine can’t read it, it can’t cite it.
• Structured data: Schema markup (article, author, FAQ) gives AI engines additional machine-readable signals about your content and authority. Similarly, an llms.txt file provides AI systems with a structured summary of your site.

GEO has two sides: onsite (how your content is structured and written) and offsite (where else you show up beyond your own website). Most guides about GEO only cover the onsite piece. That’s half the picture. AI engines pull from a wide range of sources when constructing their answers, including Reddit threads, LinkedIn posts, and industry publications. The companies I work with that actually get cited are addressing both sides, because being brilliant on your own domain isn’t enough if you’re invisible everywhere else.

Zero-click search is killing your traffic and AI is accelerating it

58.5% of US Google searches now end without a single click to any website (SparkToro, 2024 ). Searches that trigger AI Overviews push that number even higher. For B2B marketers who built their entire content strategy around organic traffic, this isn’t a minor inconvenience. It’s a fundamental change in the business model that underpins most content marketing.

Informational queries were the backbone of B2B content strategy for a decade. How-to guides, explainers, comparison posts, buying guides. Write something genuinely useful, rank for it, capture the click, nurture the lead. That model is breaking. Not because the content stopped being useful, but because the distribution mechanism changed. AI answers the question inline, and only a handful of sources earn the citation that sits alongside the answer. The rest get summarised, paraphrased, or ignored entirely.

The B2B companies I work with are all seeing this. Traffic to their resource pages is flat or declining, even when their Google rankings haven’t moved. The rankings still show page one positions, but the clicks aren’t coming through the way they used to. It’s a disorienting experience: you’re “winning” at SEO by traditional metrics but losing the thing SEO was supposed to deliver (actual visitors who might become customers).

Most of these companies are still optimising for clicks when the real competition has shifted to citations. They’re measuring the wrong outcome. A page that generates zero clicks but gets cited as the authoritative source in an AI Overview seen by 10,000 people might be more valuable than a page that captures 200 clicks from a traditional search result. But if you’re only tracking clicks, you’d never know.

The question has changed. It’s no longer “will this rank?” It’s “will AI choose to cite this when it answers the question?” That reframe changes everything about how you build content. You’re not writing for a search results page anymore. You’re writing to be selected as the authoritative source by an AI engine that has access to thousands of alternatives. And AI engines are picky. They choose sources that demonstrate expertise, that structure information clearly, and that make it easy to extract a definitive answer.

E-E-A-T SEO matters even more when AI picks your sources

E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) has become a gatekeeping filter for AI citations. Not a quality signal that nudges you up a few positions. A binary filter that determines whether AI even considers your content as a potential source. Analysis of 2,400 AI Overview citations found that 96% come from sources with strong E-E-A-T signals (Wellows ). If your content doesn’t demonstrate documented expertise, AI engines skip it entirely.

The data gets more interesting when you look at how E-E-A-T interacts with ranking position. Pages ranking #6 to #10 with strong E-E-A-T signals are cited 2.3x more frequently than #1-ranked pages with weak E-E-A-T (Wellows ). That flips the traditional SEO hierarchy on its head. Documented expertise beats domain authority. Content with proper bylines, publication dates, and structured author information gets cited up to 40% more often (ZipTie.dev, citing Wellows data ).

What does this mean in practice? You need a system that captures and surfaces expertise before you start writing. Not after. Not as an afterthought where you stick an author bio at the bottom and call it done.

When I was building Quisitive’s content process, this was the core problem to solve. Quisitive is a premier Microsoft Partner that needed blog content performing in both traditional search and AI-powered search. They had three subject matter experts, each with deep knowledge in their own domain: agentic AI, cybersecurity, and business applications. The challenge wasn’t writing good content. It was making AI engines recognise each of those people as genuine experts worth citing.

The approach I built uses separate author profiles for each SME. Each profile captures their real career background, specific voice patterns, characteristic phrases, and actual project stories. When the content goes through the writing process, it doesn’t sound like generic AI content with a name stuck on top. It sounds like that specific person wrote it, because the system is drawing on their real experience and their actual way of explaining things.

This is where most GEO content falls short: telling you E-E-A-T matters without showing you how to build a system that delivers it consistently. Author profiles that capture voice, background, and real stories. Company context files that capture services, credentials, and case studies. A writing process that enforces E-E-A-T signals in every section, not just the “about the author” box at the bottom of the page.

Just because something is coherent doesn’t mean it’s good. Building a consistent AI writing style that carries real voice and identity is part of solving this. Generic AI content (even high-quality generic AI content) fails E-E-A-T because it has no real experience behind it. The content that gets cited comes from identifiable experts with documented track records, writing about topics where they have genuine, verifiable experience.

The four layers of an AI content strategy that actually works

An AI content strategy that earns citations needs four layers running in parallel. Not just better blog posts. The companies I work with that get this right share a pattern: they don’t treat content as a standalone initiative. They treat it as a system with multiple layers reinforcing each other.

1. Onsite content: LLM-first content built around priority buyer questions. Structure every page for extractability: clear sections with direct answers upfront, proof blocks with sourced statistics, and summaries that work in isolation. Answer the question in the first two sentences of every section, because that’s what AI engines extract first. This is the foundation layer, the content AI actually crawls and decides whether to cite. I break down exactly how to build an AI content marketing process that delivers this consistently.

It helps to understand how these models actually find their sources, because it changes how you prioritise content. ChatGPT with browsing, Perplexity, and Google AI Mode all retrieve information differently: some rely heavily on real-time web search, others lean on training data supplemented by retrieval-augmented generation (RAG), which is where the model pulls from live sources to supplement what it already knows. The practical implication is that content appearing consistently across multiple platforms and source types gets compounding visibility. If your content only exists in one place, you’re visible to one retrieval method. If it appears across your site, Reddit, LinkedIn, and industry publications, multiple AI systems find it through multiple paths. No single platform covers all retrieval methods. That’s why the next three layers matter as much as the first.

2. Offsite engagement: Show up where buyers ask questions and evaluate options: Reddit, LinkedIn, Quora, and Medium. The responses need to be platform-native and genuinely useful, not thinly veiled marketing copy. Link back to onsite proof when it adds real value to the conversation. The research from XFunnel makes the case: across 40,000 AI responses and 250,000 citations, earned media dominates over owned media (XFunnel ). If you only exist on your own website, you’re invisible to a significant portion of the sources AI engines draw from when building their answers. I cover the full approach to earning AI citations through offsite engagement in a separate piece.

3. SEO and GEO hygiene: Information architecture, internal linking between pillar and cluster pages, indexation, and crawl health. Extractable answer blocks and structured sections throughout. A refresh cadence that keeps content current, because AI engines show a documented preference for recent sources over older ones. This is the technical layer that nobody finds exciting, but neglect it and your best content never gets crawled in the first place. My GEO audit guide walks through the exact checklist I use.

4. Authority building: Elevate credible experts and named authors. Maintain consistent points of view supported by proof. PR, podcasts, events, partner ecosystems. This is the layer that compounds over time and creates the authority signals AI engines weigh most heavily when deciding which sources to trust.

How to improve your brand’s visibility in AI models

Brand visibility in AI isn’t about gaming an algorithm. It’s about showing up consistently where AI models look when they’re building answers. ChatGPT’s recommendations are shaped by a combination of its training data and real-time web search (when browsing is enabled). Perplexity pulls from live sources every time. Google AI Mode uses its own search index. What they all have in common: they favour brands that appear across multiple trusted sources, not just on their own domain. If the only place your brand is mentioned is your own website, AI models have one data point. If your experts are contributing to Reddit threads, publishing on LinkedIn, getting quoted in industry publications, and maintaining a well-structured site, that’s a pattern AI models can recognise and reference. This is exactly what we built for Avalara (more on that below): the offsite layer isn’t optional for AI visibility. It’s the difference between being a single source and being a recognised name.

All four layers feed each other. Onsite content gives offsite engagement something to reference. Offsite visibility builds the authority signals that make onsite content more citable. Technical hygiene ensures AI can actually find and process everything. Authority building amplifies the reach of all three other layers.

I’ve seen this integrated approach work in practice with Avalara. They had a rich existing content library on tax compliance, but the content wasn’t showing up in AI citations. The problem wasn’t content quality. It was that Avalara was only running one layer (onsite) and ignoring the rest.

The approach combined onsite optimisation (rewriting high-intent existing assets for AI extractability) with offsite amplification (placing expert content from key people at Avalara across Reddit, LinkedIn, Quora, and Medium). The workflow identifies the right topics for the right author, finds the relevant conversations happening with their target audience, and generates responses that actually sound like the person writing them. It started as a managed service process I designed, then became a tool Avalara could run themselves.

The pattern I keep coming back to: the companies that earn AI citations aren’t the ones with the best individual blog posts. They’re the ones running all four layers as an integrated system, where each layer makes the others more effective.

How to optimise your content so AI engines actually cite it

AI content optimisation comes down to seven practical principles you can apply to everything you publish. I built a process that checks all of these systematically, and the gap between knowing the rules and having a system that enforces them consistently is where most content falls short.

1. Answer first: Open every section with a direct answer to the question posed by the heading. Lead with the answer, not the context. If a reader (or an AI engine) can’t find the answer in the first two sentences, they’ll find it somewhere else.

2. Standalone sections: Write every section as if it might be extracted and shown on its own, because it might be. Include enough context for each section to make sense without what comes before or after it.

3. Open access: 99.3% of LLM citations come from open-access sources (SegmentSEO ). If your best content is behind a registration wall, a gate, or a login, AI engines can’t see it. The old lead-gen model of gating your strongest content works against you in AI search.

4. Add statistics with sources: The Princeton GEO study found that adding statistics improves AI visibility by up to 40%, and quotations boost it by 37% (Aggarwal et al., 2023 ). Every claim backed by a sourced statistic is more likely to be cited. Unsourced claims get treated as opinion.

5. Structured data: Schema markup (article schema, author schema, FAQ schema) gives AI engines additional machine-readable signals about your content and your authors. It’s the layer that makes your human-readable content easier for AI to process and attribute correctly.

6. Freshness matters: AI engines show a documented preference for recent content over older sources. Build a refresh cadence into your content calendar: update statistics, add new examples, and make sure publication dates reflect when content was last meaningfully updated.

7. Author attribution: Proper bylines, author bios, and linked author profiles increase citation rates. Anonymous or unattributed content struggles because it lacks the E-E-A-T signals AI engines use as a filter. A named expert with documented credentials always outperforms a faceless brand post.

The process I’ve built handles this end to end: research the question using real search volume data, plan the structure around keyword-backed H2s, write in a specific author’s voice using their real background and stories, then run editorial checks against every one of these principles. The system is the product. Without it, you’re relying on individual writers to remember and apply seven optimisation principles on every paragraph of every post. That doesn’t scale. And the inconsistency shows.

How do you build content that AI search engines cite

The question was how to build content that AI search engines actually cite. The answer is a system, not a single technique.

GEO builds on SEO. Ranking is still important, but it’s no longer sufficient on its own. Your content needs to be extractable and citable, structured so AI can pull standalone answers and attribute them to a credible source.

Zero-click search means your content increasingly needs to work as a citation source rather than a traffic destination. The old model of ranking, capturing the click, and nurturing the lead is being supplemented by a model where AI engines reference your expertise directly in their answers. If your content isn’t built for that model, it gets left out.

E-E-A-T is the gatekeeping filter. 96% of AI Overview citations come from sources with strong E-E-A-T signals. Document your expertise through author profiles, company context, and real case studies before you start writing. Without those signals, AI engines will skip your content regardless of how well it ranks.

Run all four layers of your AI content strategy: onsite content, offsite engagement, technical hygiene, and authority building. The companies that get cited aren’t running one layer well. They’re running all four as an integrated system where each layer reinforces the others.

Structure every section as a standalone answer card. AI may extract any section in isolation, so every part of your content needs to work independently, with enough context to make sense on its own.

The limitations that have always held content marketing back (scaling quality content, making every piece genuinely expert, being visible everywhere your audience looks) are being lifted by AI. I don’t believe that. I can see it. But only if you build the system to match.

If you’re rethinking your AI search optimisation strategy and wondering where to start, get in touch . I’ll walk you through how we approach it.

Frequently asked questions