For twenty-five years, digital visibility meant one thing: rank on page one of Google. That era is not over, but it is no longer the whole story. Today, a growing share of buying decisions begin with a question typed into ChatGPT, Perplexity, Gemini, Claude, or Google's AI Overviews and AI Mode — and the answer the customer receives is not a list of ten blue links. It is a synthesized paragraph, and your brand is either in that paragraph or it is invisible.

This guidebook explains, step by step, how to make your brand and products the ones these systems retrieve, cite, and recommend. It is grounded in three kinds of evidence: peer-reviewed academic research (principally the Princeton-led GEO study presented at ACM KDD 2024), official platform documentation (especially Google's first formal guide to its generative AI features), and large-scale citation studies analyzing hundreds of millions of real AI answers. The full reference list appears at the end.

One note on terminology: you will see this discipline called GEO (Generative Engine Optimization), AEO, LLMO, or AI SEO. Google's own position is that optimizing for generative AI search is optimizing for the search experience — and thus still SEO. The labels matter less than the work.

Chapter 01

How AI Search Actually Decides What to Cite

You cannot optimize a system you do not understand. There are two distinct pathways by which your brand can appear in an AI-generated answer, and they require different strategies.

The first pathway is training data. Large language models learn from enormous snapshots of the public web. If your brand has been discussed, reviewed, and described consistently across trusted sources over time, the model internalizes that association and may recommend you from memory alone — no live lookup, no link. Marketers call this "linkless influence": recognition without a click, which is becoming as valuable as traffic itself.

The second pathway is retrieval-augmented generation (RAG), also called grounding. This is how AI Overviews, AI Mode, Perplexity, and the search modes of ChatGPT and Claude work in real time: the system searches a live index, retrieves relevant pages, and synthesizes an answer with citation links back to sources. Google has confirmed its generative features pull from the same Search index that powers traditional results — there is no separate AI index.

RAG adds one wrinkle that changes content strategy fundamentally: query fan-out. Rather than running the user's question as one search, the model decomposes it into several concurrent sub-queries. Google's own example: "how to fix a lawn that's full of weeds" may fan out into "best herbicides for lawns," "remove weeds without chemicals," and "how to prevent weeds in lawn." Your content wins by ranking for these machine-generated sub-questions — one December 2025 analysis found that ranking for fan-out queries boosts citation odds by 161%.

Finally, AI answers are non-deterministic: ask the same question five times and you may get five different answers. There is no fixed "position #1" in ChatGPT. Your goal is not a ranking; it is a mention rate — how often your brand appears across many responses to many related prompts.

Chapter 02

The Foundation — Yes, It's Still SEO

In May 2026, Google published its first comprehensive official document on the subject: Optimizing your website for generative AI features on Google Search. Its central message: generative AI features are rooted in Google's core Search ranking and quality systems, and the best practices for SEO remain the best practices for AI visibility.

The implications are concrete. To be eligible as a supporting link in AI Overviews or AI Mode, a page must simply be indexed and eligible to appear in Google Search with a snippet. No additional technical requirements, no special AI-only schema, no secret markup. Industry data backs this up: the overwhelming majority of links appearing in AI Overviews come from pages already ranking in top organic positions.

Just as importantly, Google's guide names tactics you can safely deprioritize for its AI surfaces: you do not need an llms.txt file (Google assigns it no special meaning), you do not need to artificially "chunk" content, and you should not pursue inauthentic brand mentions. Google also clarified that its full spam policy catalog now applies to AI Overviews and AI Mode citations — sites demoted for spam are excluded from the AI citation pool.

One nuance for multi-platform strategy: Google's dismissals apply to Google's AI features. Other platforms behave differently, and some practitioners still find llms.txt useful for non-Google AI crawlers. Treat Google's guidance as authoritative for Google — and directionally useful, but not gospel, everywhere else.

Chapter 03

Writing Content That AI Systems Choose to Cite

This is where peer-reviewed science comes in. The study GEO: Generative Engine Optimization (Aggarwal et al.) — conducted by researchers from Princeton, Georgia Tech, IIT Delhi, and the Allen Institute for AI, and presented at ACM KDD 2024 — was the first large-scale controlled experiment on what makes content citation-worthy. Across a 10,000-query benchmark, nine content modifications were tested. Five tactics produced gains of roughly 30–40% in citation visibility:

Structure for extraction

Answer first, elaborate second. Lead each section with a direct, complete answer (BLUF — Bottom Line Up Front), then expand. LLMs frequently extract a single passage, so every major section should stand alone — roughly 44% of ChatGPT citations come from the first third of a page.

Match the shape of AI queries. The average ChatGPT prompt runs about 23 words, versus 3–4 for a classic Google search. Structure content around real question clusters — how, why, what's the difference, which is better — with genuine FAQ sections.

Prioritize information gain. AI can generate generic summaries instantly; it cannot generate your proprietary data, client case results, testing screenshots, or your named expert's contrarian take. Original research earns citations precisely because no one else has it.

Keep it genuinely fresh. LLM-powered search favors recent content, but changing a date stamp without changing substance fools no one. Update when facts materially change.

Chapter 04

Entity Engineering — Teaching Machines Who You Are

LLMs do not think in keywords; they think in entities — distinct, identifiable things (your brand, products, founders) and the relationships between them. If a model cannot cleanly resolve who you are, it cannot confidently recommend you, no matter how good your content is. Entity recognition builds the trust required to be a candidate for an answer; content optimization increases the odds of being the chosen source.

Start on your own site. The first paragraph of your homepage, about page, and core service pages should contain a clear, machine-parseable entity definition — one sentence stating what your company is, what it does, for whom, and since when. That sentence is not marketing flourish; it is data infrastructure.

Support the entity with structured data: JSON-LD schema — Organization, Product, Person, FAQ, Article, Review — deployed accurately and site-wide. Be clear-eyed about what this does: Google states no special schema is required for AI inclusion. Its value is upstream — it disambiguates your entities and keeps machine understanding precise. Helpful, not required; do it anyway.

Finally, work toward the reference layer: Wikipedia, Wikidata, and Google's Knowledge Graph. Wikipedia remains among the most-cited individual sources across AI platforms, and communications professionals now describe it as infrastructure rather than PR. A Wikipedia page is not achievable for every business — notability standards are real and paid manipulation backfires — but accurate Wikidata entries, complete Business Profiles, and consistent industry-directory presence are achievable for everyone.

Chapter 05

The Third-Party Battlefield — Where Citations Actually Come From

Here is the uncomfortable truth that reshapes budgets: much of what determines whether AI recommends your brand happens on websites you do not own. Being praised on trusted third-party platforms increases your odds of being cited far more than self-promotion ever will.

The research is remarkably consistent about where the gravity is. A 2026 index synthesizing more than 680 million citations across ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude found Reddit the most-cited source across every major engine. Peec AI's 30-million-source analysis reached the same conclusion, with YouTube, LinkedIn, Wikipedia, and Forbes rounding out the top five — and review platforms like Yelp and G2 dominating recommendation-style queries.

The third-party playbook

Community presence (Reddit & niche forums). AI engines treat community threads as authentic, experience-based signal. The rule is unforgiving: participate genuinely or not at all. Astroturfed praise gets detected — by communities and by spam enforcement.

Video (YouTube). Among the most-cited domains in AI search overall. Transcribed, well-titled explainer videos create citable assets in a channel most professional-services competitors still neglect.

Review platforms. For any "best X for Y" query — the commercial gold of AI search — models lean on G2, Capterra, Trustpilot, Yelp, and vertical equivalents. Reviews shape entity sentiment: the qualitative attributes AI associates with your brand.

Editorial & digital PR. Forbes and Business Insider rank among ChatGPT's most-cited sources; Claude prefers legacy journalism. The most reliable way in: become the statistic. Publish original research, then pitch the data. Journalists cite data; AI cites the journalists; AI cites you.

Professional platforms (LinkedIn). Top-five cited across several engines, with particular weight in B2B queries. Consistent expert publishing under named authors ties personal authority to your brand entity.

Chapter 06

Know Your Platforms — One Strategy, Five Dialects

The foundational work applies everywhere, but each engine has a measurable accent, and a mature program allocates effort accordingly.

Vertical patterns cut across platforms too: SaaS citations skew toward G2 and Reddit; health defers to government and hospital domains; finance rewards Bloomberg and SEC filings. Run your top 20 customer questions through each engine, log the sources, and target those specific watering holes.

Chapter 07

Technical Access — Letting the Machines In

A brief but non-negotiable chapter. Every retrieval-based system depends on a crawler reaching your content — and an astonishing number of sites block AI crawlers by accident, through stale robots.txt rules, aggressive CDN bot protection, or inherited firewall settings.

Beyond access, the checklist is the familiar one: server-side rendering or render-safe JavaScript (several AI crawlers execute little or no JavaScript, so critical content must exist in the initial HTML), fast load times, clean semantic HTML with logical heading hierarchy, XML sitemaps, and accurate canonicals. If content is locked inside images or scripts without HTML equivalents, assume the machines cannot read it.

Chapter 08

Measuring What Was Never Measurable Before

Traditional rank tracking cannot see inside a synthesized paragraph, so GEO requires its own scoreboard, built in four layers.

Mention rate. Assemble a prompt panel — 50 to 200 realistic questions your customers actually ask, spanning discovery ("how do I…"), comparison ("X vs Y"), and recommendation ("best [category] for [audience]") intents. Run it across the major engines on a recurring schedule, logging separately whether your brand is mentioned and whether your site is cited. Because outputs are non-deterministic, run multiple trials and track the frequency, not any single result. Dedicated platforms (Profound, Evertune, Peec AI, Semrush's AI toolkit) automate this — but a disciplined manual panel in a spreadsheet is a perfectly credible start.

Sentiment and accuracy. It is not enough to be mentioned; audit how you're described. Models sometimes carry outdated pricing or a competitor's attribute mistakenly attached to your brand. Every inaccuracy is a content assignment: publish the correction prominently and get it corroborated on third-party sources.

Referral traffic and conversion. Segment AI-platform referrals in GA4 (chatgpt.com, perplexity.ai, and peers). Expect small volumes and high intent — measure this channel on conversion rate and revenue per visit, not raw sessions. Google's own data points the same way: clicks from pages with AI Overviews are higher quality, with users spending more time on site.

Citation velocity. Track how quickly new content enters AI answers after publication — a leading indicator of your domain's standing with retrieval systems. Review monthly to see whether authority is compounding.

Chapter 09

What Not to Do

Every visibility gold rush attracts snake oil. Four warnings, each grounded in the sources referenced in this guide.

Chapter 10

The 90-Day Action Plan

The whole guidebook, compressed into a quarter's work.

Then repeat. GEO is not a project with an end date; it is the new shape of the discipline. The brands showing up in AI answers a year from now are the ones doing this ordinary, verifiable, well-sourced work today — while their competitors are still asking whether AI search is real.

It's real. Get cited.

For twenty-five years, digital visibility meant one thing: rank on page one of Google. That era is not over, but it is no longer the whole story. Today, a growing share of buying decisions begin with a question typed into ChatGPT, Perplexity, Gemini, Claude, or Google's AI Overviews and AI Mode — and the answer the customer receives is not a list of ten blue links. It is a synthesized paragraph, and your brand is either in that paragraph or it is invisible.

This guidebook explains, step by step, how to make your brand and products the ones these systems retrieve, cite, and recommend. It is grounded in three kinds of evidence: peer-reviewed academic research (principally the Princeton-led GEO study presented at ACM KDD 2024), official platform documentation (especially Google's first formal guide to its generative AI features), and large-scale citation studies analyzing hundreds of millions of real AI answers. The full reference list appears at the end.

One note on terminology: you will see this discipline called GEO (Generative Engine Optimization), AEO, LLMO, or AI SEO. Google's own position is that optimizing for generative AI search is optimizing for the search experience — and thus still SEO. The labels matter less than the work.

Chapter 01

How AI Search Actually Decides What to Cite

You cannot optimize a system you do not understand. There are two distinct pathways by which your brand can appear in an AI-generated answer, and they require different strategies.

The first pathway is training data. Large language models learn from enormous snapshots of the public web. If your brand has been discussed, reviewed, and described consistently across trusted sources over time, the model internalizes that association and may recommend you from memory alone — no live lookup, no link. Marketers call this "linkless influence": recognition without a click, which is becoming as valuable as traffic itself.

The second pathway is retrieval-augmented generation (RAG), also called grounding. This is how AI Overviews, AI Mode, Perplexity, and the search modes of ChatGPT and Claude work in real time: the system searches a live index, retrieves relevant pages, and synthesizes an answer with citation links back to sources. Google has confirmed its generative features pull from the same Search index that powers traditional results — there is no separate AI index.

RAG adds one wrinkle that changes content strategy fundamentally: query fan-out. Rather than running the user's question as one search, the model decomposes it into several concurrent sub-queries. Google's own example: "how to fix a lawn that's full of weeds" may fan out into "best herbicides for lawns," "remove weeds without chemicals," and "how to prevent weeds in lawn." Your content wins by ranking for these machine-generated sub-questions — one December 2025 analysis found that ranking for fan-out queries boosts citation odds by 161%.

Finally, AI answers are non-deterministic: ask the same question five times and you may get five different answers. There is no fixed "position #1" in ChatGPT. Your goal is not a ranking; it is a mention rate — how often your brand appears across many responses to many related prompts.

Chapter 02

The Foundation — Yes, It's Still SEO

In May 2026, Google published its first comprehensive official document on the subject: Optimizing your website for generative AI features on Google Search. Its central message: generative AI features are rooted in Google's core Search ranking and quality systems, and the best practices for SEO remain the best practices for AI visibility.

The implications are concrete. To be eligible as a supporting link in AI Overviews or AI Mode, a page must simply be indexed and eligible to appear in Google Search with a snippet. No additional technical requirements, no special AI-only schema, no secret markup. Industry data backs this up: the overwhelming majority of links appearing in AI Overviews come from pages already ranking in top organic positions.

Just as importantly, Google's guide names tactics you can safely deprioritize for its AI surfaces: you do not need an llms.txt file (Google assigns it no special meaning), you do not need to artificially "chunk" content, and you should not pursue inauthentic brand mentions. Google also clarified that its full spam policy catalog now applies to AI Overviews and AI Mode citations — sites demoted for spam are excluded from the AI citation pool.

One nuance for multi-platform strategy: Google's dismissals apply to Google's AI features. Other platforms behave differently, and some practitioners still find llms.txt useful for non-Google AI crawlers. Treat Google's guidance as authoritative for Google — and directionally useful, but not gospel, everywhere else.

Chapter 03

Writing Content That AI Systems Choose to Cite

This is where peer-reviewed science comes in. The study GEO: Generative Engine Optimization (Aggarwal et al.) — conducted by researchers from Princeton, Georgia Tech, IIT Delhi, and the Allen Institute for AI, and presented at ACM KDD 2024 — was the first large-scale controlled experiment on what makes content citation-worthy. Across a 10,000-query benchmark, nine content modifications were tested. Five tactics produced gains of roughly 30–40% in citation visibility:

Structure for extraction

Answer first, elaborate second. Lead each section with a direct, complete answer (BLUF — Bottom Line Up Front), then expand. LLMs frequently extract a single passage, so every major section should stand alone — roughly 44% of ChatGPT citations come from the first third of a page.

Match the shape of AI queries. The average ChatGPT prompt runs about 23 words, versus 3–4 for a classic Google search. Structure content around real question clusters — how, why, what's the difference, which is better — with genuine FAQ sections.

Prioritize information gain. AI can generate generic summaries instantly; it cannot generate your proprietary data, client case results, testing screenshots, or your named expert's contrarian take. Original research earns citations precisely because no one else has it.

Keep it genuinely fresh. LLM-powered search favors recent content, but changing a date stamp without changing substance fools no one. Update when facts materially change.

Chapter 04

Entity Engineering — Teaching Machines Who You Are

LLMs do not think in keywords; they think in entities — distinct, identifiable things (your brand, products, founders) and the relationships between them. If a model cannot cleanly resolve who you are, it cannot confidently recommend you, no matter how good your content is. Entity recognition builds the trust required to be a candidate for an answer; content optimization increases the odds of being the chosen source.

Start on your own site. The first paragraph of your homepage, about page, and core service pages should contain a clear, machine-parseable entity definition — one sentence stating what your company is, what it does, for whom, and since when. That sentence is not marketing flourish; it is data infrastructure.

Support the entity with structured data: JSON-LD schema — Organization, Product, Person, FAQ, Article, Review — deployed accurately and site-wide. Be clear-eyed about what this does: Google states no special schema is required for AI inclusion. Its value is upstream — it disambiguates your entities and keeps machine understanding precise. Helpful, not required; do it anyway.

Finally, work toward the reference layer: Wikipedia, Wikidata, and Google's Knowledge Graph. Wikipedia remains among the most-cited individual sources across AI platforms, and communications professionals now describe it as infrastructure rather than PR. A Wikipedia page is not achievable for every business — notability standards are real and paid manipulation backfires — but accurate Wikidata entries, complete Business Profiles, and consistent industry-directory presence are achievable for everyone.

Chapter 05

The Third-Party Battlefield — Where Citations Actually Come From

Here is the uncomfortable truth that reshapes budgets: much of what determines whether AI recommends your brand happens on websites you do not own. Being praised on trusted third-party platforms increases your odds of being cited far more than self-promotion ever will.

The research is remarkably consistent about where the gravity is. A 2026 index synthesizing more than 680 million citations across ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude found Reddit the most-cited source across every major engine. Peec AI's 30-million-source analysis reached the same conclusion, with YouTube, LinkedIn, Wikipedia, and Forbes rounding out the top five — and review platforms like Yelp and G2 dominating recommendation-style queries.

The third-party playbook

Community presence (Reddit & niche forums). AI engines treat community threads as authentic, experience-based signal. The rule is unforgiving: participate genuinely or not at all. Astroturfed praise gets detected — by communities and by spam enforcement.

Video (YouTube). Among the most-cited domains in AI search overall. Transcribed, well-titled explainer videos create citable assets in a channel most professional-services competitors still neglect.

Review platforms. For any "best X for Y" query — the commercial gold of AI search — models lean on G2, Capterra, Trustpilot, Yelp, and vertical equivalents. Reviews shape entity sentiment: the qualitative attributes AI associates with your brand.

Editorial & digital PR. Forbes and Business Insider rank among ChatGPT's most-cited sources; Claude prefers legacy journalism. The most reliable way in: become the statistic. Publish original research, then pitch the data. Journalists cite data; AI cites the journalists; AI cites you.

Professional platforms (LinkedIn). Top-five cited across several engines, with particular weight in B2B queries. Consistent expert publishing under named authors ties personal authority to your brand entity.

Chapter 06

Know Your Platforms — One Strategy, Five Dialects

The foundational work applies everywhere, but each engine has a measurable accent, and a mature program allocates effort accordingly.

Vertical patterns cut across platforms too: SaaS citations skew toward G2 and Reddit; health defers to government and hospital domains; finance rewards Bloomberg and SEC filings. Run your top 20 customer questions through each engine, log the sources, and target those specific watering holes.

Chapter 07

Technical Access — Letting the Machines In

A brief but non-negotiable chapter. Every retrieval-based system depends on a crawler reaching your content — and an astonishing number of sites block AI crawlers by accident, through stale robots.txt rules, aggressive CDN bot protection, or inherited firewall settings.

Beyond access, the checklist is the familiar one: server-side rendering or render-safe JavaScript (several AI crawlers execute little or no JavaScript, so critical content must exist in the initial HTML), fast load times, clean semantic HTML with logical heading hierarchy, XML sitemaps, and accurate canonicals. If content is locked inside images or scripts without HTML equivalents, assume the machines cannot read it.

Chapter 08

Measuring What Was Never Measurable Before

Traditional rank tracking cannot see inside a synthesized paragraph, so GEO requires its own scoreboard, built in four layers.

Mention rate. Assemble a prompt panel — 50 to 200 realistic questions your customers actually ask, spanning discovery ("how do I…"), comparison ("X vs Y"), and recommendation ("best [category] for [audience]") intents. Run it across the major engines on a recurring schedule, logging separately whether your brand is mentioned and whether your site is cited. Because outputs are non-deterministic, run multiple trials and track the frequency, not any single result. Dedicated platforms (Profound, Evertune, Peec AI, Semrush's AI toolkit) automate this — but a disciplined manual panel in a spreadsheet is a perfectly credible start.

Sentiment and accuracy. It is not enough to be mentioned; audit how you're described. Models sometimes carry outdated pricing or a competitor's attribute mistakenly attached to your brand. Every inaccuracy is a content assignment: publish the correction prominently and get it corroborated on third-party sources.

Referral traffic and conversion. Segment AI-platform referrals in GA4 (chatgpt.com, perplexity.ai, and peers). Expect small volumes and high intent — measure this channel on conversion rate and revenue per visit, not raw sessions. Google's own data points the same way: clicks from pages with AI Overviews are higher quality, with users spending more time on site.

Citation velocity. Track how quickly new content enters AI answers after publication — a leading indicator of your domain's standing with retrieval systems. Review monthly to see whether authority is compounding.

Chapter 09

What Not to Do

Every visibility gold rush attracts snake oil. Four warnings, each grounded in the sources referenced in this guide.

Chapter 10

The 90-Day Action Plan

The whole guidebook, compressed into a quarter's work.

Then repeat. GEO is not a project with an end date; it is the new shape of the discipline. The brands showing up in AI answers a year from now are the ones doing this ordinary, verifiable, well-sourced work today — while their competitors are still asking whether AI search is real.

It's real. Get cited.

For twenty-five years, digital visibility meant one thing: rank on page one of Google. That era is not over, but it is no longer the whole story. Today, a growing share of buying decisions begin with a question typed into ChatGPT, Perplexity, Gemini, Claude, or Google's AI Overviews and AI Mode — and the answer the customer receives is not a list of ten blue links. It is a synthesized paragraph, and your brand is either in that paragraph or it is invisible.

This guidebook explains, step by step, how to make your brand and products the ones these systems retrieve, cite, and recommend. It is grounded in three kinds of evidence: peer-reviewed academic research (principally the Princeton-led GEO study presented at ACM KDD 2024), official platform documentation (especially Google's first formal guide to its generative AI features), and large-scale citation studies analyzing hundreds of millions of real AI answers. The full reference list appears at the end.

One note on terminology: you will see this discipline called GEO (Generative Engine Optimization), AEO, LLMO, or AI SEO. Google's own position is that optimizing for generative AI search is optimizing for the search experience — and thus still SEO. The labels matter less than the work.

Chapter 01

How AI Search Actually Decides What to Cite

You cannot optimize a system you do not understand. There are two distinct pathways by which your brand can appear in an AI-generated answer, and they require different strategies.

The first pathway is training data. Large language models learn from enormous snapshots of the public web. If your brand has been discussed, reviewed, and described consistently across trusted sources over time, the model internalizes that association and may recommend you from memory alone — no live lookup, no link. Marketers call this "linkless influence": recognition without a click, which is becoming as valuable as traffic itself.

The second pathway is retrieval-augmented generation (RAG), also called grounding. This is how AI Overviews, AI Mode, Perplexity, and the search modes of ChatGPT and Claude work in real time: the system searches a live index, retrieves relevant pages, and synthesizes an answer with citation links back to sources. Google has confirmed its generative features pull from the same Search index that powers traditional results — there is no separate AI index.

RAG adds one wrinkle that changes content strategy fundamentally: query fan-out. Rather than running the user's question as one search, the model decomposes it into several concurrent sub-queries. Google's own example: "how to fix a lawn that's full of weeds" may fan out into "best herbicides for lawns," "remove weeds without chemicals," and "how to prevent weeds in lawn." Your content wins by ranking for these machine-generated sub-questions — one December 2025 analysis found that ranking for fan-out queries boosts citation odds by 161%.

Finally, AI answers are non-deterministic: ask the same question five times and you may get five different answers. There is no fixed "position #1" in ChatGPT. Your goal is not a ranking; it is a mention rate — how often your brand appears across many responses to many related prompts.

Chapter 02

The Foundation — Yes, It's Still SEO

In May 2026, Google published its first comprehensive official document on the subject: Optimizing your website for generative AI features on Google Search. Its central message: generative AI features are rooted in Google's core Search ranking and quality systems, and the best practices for SEO remain the best practices for AI visibility.

The implications are concrete. To be eligible as a supporting link in AI Overviews or AI Mode, a page must simply be indexed and eligible to appear in Google Search with a snippet. No additional technical requirements, no special AI-only schema, no secret markup. Industry data backs this up: the overwhelming majority of links appearing in AI Overviews come from pages already ranking in top organic positions.

Just as importantly, Google's guide names tactics you can safely deprioritize for its AI surfaces: you do not need an llms.txt file (Google assigns it no special meaning), you do not need to artificially "chunk" content, and you should not pursue inauthentic brand mentions. Google also clarified that its full spam policy catalog now applies to AI Overviews and AI Mode citations — sites demoted for spam are excluded from the AI citation pool.

One nuance for multi-platform strategy: Google's dismissals apply to Google's AI features. Other platforms behave differently, and some practitioners still find llms.txt useful for non-Google AI crawlers. Treat Google's guidance as authoritative for Google — and directionally useful, but not gospel, everywhere else.

Chapter 03

Writing Content That AI Systems Choose to Cite

This is where peer-reviewed science comes in. The study GEO: Generative Engine Optimization (Aggarwal et al.) — conducted by researchers from Princeton, Georgia Tech, IIT Delhi, and the Allen Institute for AI, and presented at ACM KDD 2024 — was the first large-scale controlled experiment on what makes content citation-worthy. Across a 10,000-query benchmark, nine content modifications were tested. Five tactics produced gains of roughly 30–40% in citation visibility:

Structure for extraction

Answer first, elaborate second. Lead each section with a direct, complete answer (BLUF — Bottom Line Up Front), then expand. LLMs frequently extract a single passage, so every major section should stand alone — roughly 44% of ChatGPT citations come from the first third of a page.

Match the shape of AI queries. The average ChatGPT prompt runs about 23 words, versus 3–4 for a classic Google search. Structure content around real question clusters — how, why, what's the difference, which is better — with genuine FAQ sections.

Prioritize information gain. AI can generate generic summaries instantly; it cannot generate your proprietary data, client case results, testing screenshots, or your named expert's contrarian take. Original research earns citations precisely because no one else has it.

Keep it genuinely fresh. LLM-powered search favors recent content, but changing a date stamp without changing substance fools no one. Update when facts materially change.

Chapter 04

Entity Engineering — Teaching Machines Who You Are

LLMs do not think in keywords; they think in entities — distinct, identifiable things (your brand, products, founders) and the relationships between them. If a model cannot cleanly resolve who you are, it cannot confidently recommend you, no matter how good your content is. Entity recognition builds the trust required to be a candidate for an answer; content optimization increases the odds of being the chosen source.

Start on your own site. The first paragraph of your homepage, about page, and core service pages should contain a clear, machine-parseable entity definition — one sentence stating what your company is, what it does, for whom, and since when. That sentence is not marketing flourish; it is data infrastructure.

Support the entity with structured data: JSON-LD schema — Organization, Product, Person, FAQ, Article, Review — deployed accurately and site-wide. Be clear-eyed about what this does: Google states no special schema is required for AI inclusion. Its value is upstream — it disambiguates your entities and keeps machine understanding precise. Helpful, not required; do it anyway.

Finally, work toward the reference layer: Wikipedia, Wikidata, and Google's Knowledge Graph. Wikipedia remains among the most-cited individual sources across AI platforms, and communications professionals now describe it as infrastructure rather than PR. A Wikipedia page is not achievable for every business — notability standards are real and paid manipulation backfires — but accurate Wikidata entries, complete Business Profiles, and consistent industry-directory presence are achievable for everyone.

Chapter 05

The Third-Party Battlefield — Where Citations Actually Come From

Here is the uncomfortable truth that reshapes budgets: much of what determines whether AI recommends your brand happens on websites you do not own. Being praised on trusted third-party platforms increases your odds of being cited far more than self-promotion ever will.

The research is remarkably consistent about where the gravity is. A 2026 index synthesizing more than 680 million citations across ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude found Reddit the most-cited source across every major engine. Peec AI's 30-million-source analysis reached the same conclusion, with YouTube, LinkedIn, Wikipedia, and Forbes rounding out the top five — and review platforms like Yelp and G2 dominating recommendation-style queries.

The third-party playbook

Community presence (Reddit & niche forums). AI engines treat community threads as authentic, experience-based signal. The rule is unforgiving: participate genuinely or not at all. Astroturfed praise gets detected — by communities and by spam enforcement.

Video (YouTube). Among the most-cited domains in AI search overall. Transcribed, well-titled explainer videos create citable assets in a channel most professional-services competitors still neglect.

Review platforms. For any "best X for Y" query — the commercial gold of AI search — models lean on G2, Capterra, Trustpilot, Yelp, and vertical equivalents. Reviews shape entity sentiment: the qualitative attributes AI associates with your brand.

Editorial & digital PR. Forbes and Business Insider rank among ChatGPT's most-cited sources; Claude prefers legacy journalism. The most reliable way in: become the statistic. Publish original research, then pitch the data. Journalists cite data; AI cites the journalists; AI cites you.

Professional platforms (LinkedIn). Top-five cited across several engines, with particular weight in B2B queries. Consistent expert publishing under named authors ties personal authority to your brand entity.

Chapter 06

Know Your Platforms — One Strategy, Five Dialects

The foundational work applies everywhere, but each engine has a measurable accent, and a mature program allocates effort accordingly.

Vertical patterns cut across platforms too: SaaS citations skew toward G2 and Reddit; health defers to government and hospital domains; finance rewards Bloomberg and SEC filings. Run your top 20 customer questions through each engine, log the sources, and target those specific watering holes.

Chapter 07

Technical Access — Letting the Machines In

A brief but non-negotiable chapter. Every retrieval-based system depends on a crawler reaching your content — and an astonishing number of sites block AI crawlers by accident, through stale robots.txt rules, aggressive CDN bot protection, or inherited firewall settings.

Beyond access, the checklist is the familiar one: server-side rendering or render-safe JavaScript (several AI crawlers execute little or no JavaScript, so critical content must exist in the initial HTML), fast load times, clean semantic HTML with logical heading hierarchy, XML sitemaps, and accurate canonicals. If content is locked inside images or scripts without HTML equivalents, assume the machines cannot read it.

Chapter 08

Measuring What Was Never Measurable Before

Traditional rank tracking cannot see inside a synthesized paragraph, so GEO requires its own scoreboard, built in four layers.

Mention rate. Assemble a prompt panel — 50 to 200 realistic questions your customers actually ask, spanning discovery ("how do I…"), comparison ("X vs Y"), and recommendation ("best [category] for [audience]") intents. Run it across the major engines on a recurring schedule, logging separately whether your brand is mentioned and whether your site is cited. Because outputs are non-deterministic, run multiple trials and track the frequency, not any single result. Dedicated platforms (Profound, Evertune, Peec AI, Semrush's AI toolkit) automate this — but a disciplined manual panel in a spreadsheet is a perfectly credible start.

Sentiment and accuracy. It is not enough to be mentioned; audit how you're described. Models sometimes carry outdated pricing or a competitor's attribute mistakenly attached to your brand. Every inaccuracy is a content assignment: publish the correction prominently and get it corroborated on third-party sources.

Referral traffic and conversion. Segment AI-platform referrals in GA4 (chatgpt.com, perplexity.ai, and peers). Expect small volumes and high intent — measure this channel on conversion rate and revenue per visit, not raw sessions. Google's own data points the same way: clicks from pages with AI Overviews are higher quality, with users spending more time on site.

Citation velocity. Track how quickly new content enters AI answers after publication — a leading indicator of your domain's standing with retrieval systems. Review monthly to see whether authority is compounding.

Chapter 09

What Not to Do

Every visibility gold rush attracts snake oil. Four warnings, each grounded in the sources referenced in this guide.

Chapter 10

The 90-Day Action Plan

The whole guidebook, compressed into a quarter's work.

Then repeat. GEO is not a project with an end date; it is the new shape of the discipline. The brands showing up in AI answers a year from now are the ones doing this ordinary, verifiable, well-sourced work today — while their competitors are still asking whether AI search is real.

It's real. Get cited.