Stop Writing for Google. Start Writing to Be Quoted.

The content strategy that ranked you on Google is not the content strategy that gets you cited by AI. The gap between the two is structural — not cosmetic. This is how to close it without throwing out everything you’ve already built.

In the last post I made the case that AI search has fundamentally disrupted B2B inbound — that the click-driven traffic model most demand gen programmes depended on is being eroded by AI Overviews and zero-click search.

The obvious follow-up question is: so what do you actually do about it? How do you write content that doesn’t just rank, but gets cited inside the AI answers your buyers are now reading instead of clicking through to your site?

That’s what this post is about. Not the abstract principle — “write better content” — but the specific structural changes that determine whether your content gets extracted and cited by AI systems, or passed over in favour of content that was built the right way.

I want to start with what’s actually happening mechanically when an AI generates a search response, because understanding the mechanism changes how you think about the solution.

What AI Systems Actually Do When They Generate an Answer

Traditional SEO optimises for a ranking algorithm that looks at your entire page — its domain authority, backlink profile, keyword relevance, technical health — and assigns it a position in a results list. The page wins or loses as a whole unit.

AI citation works differently. When ChatGPT, Perplexity, or Google’s AI Mode generates a response to a query, it doesn’t rank your page. It extracts passages from your page. The underlying system — Retrieval-Augmented Generation, or RAG — does four things in sequence: it retrieves a set of candidate passages based on what the user asked, scores each passage for how well it answers the query on its own, evaluates the trustworthiness of the source, and synthesises a response citing the most useful extracts.

The critical word there is “passage.” Not page. Not domain. Individual passages — usually 100 to 200 words — that can be extracted from your content and understood completely without any surrounding context.

“Google ranks pages. AI cites passages. That single difference changes everything about how you structure, write, and format B2B content.”

This is why a page that ranks number three on Google for a relevant keyword might not appear in a single AI Overview for the same query — while a page with lower domain authority, fewer backlinks, and lower traditional search performance gets cited repeatedly. The traditional ranking signals and the AI citation signals overlap but they’re not the same. They reward different things. And most B2B content was built entirely for the first set.

 

40% Higher citation rate for quantitative claims vs qualitative statements

134 Optimal words per AI-cited passage (sweet spot: 120–180)

44% Of all LLM citations come from the first 30% of a page’s text

2–3× Higher conversion rate from AI-referred traffic vs organic

 

The Real Problem with Most B2B Content

Most B2B content — including most of the content on well-run demand gen blogs, including content I’ve produced for clients — was written in a way that made sense for the old model. Long, flowing articles. Arguments that build across multiple paragraphs. Insights embedded in the middle of a section that assume the reader has read everything before them.

That writing style produces engaging, readable content for a human who’s sitting down to read a whole piece. It produces poor AI citation performance, because the AI can’t extract a clean, self-contained answer from it.

Here’s a concrete example of the problem. Suppose you’ve written a section of a blog post arguing that marketing teams are structurally misaligned from pipeline generation. The argument builds across six paragraphs. The core insight — the specific, citable claim — is in paragraph four. But paragraph four uses pronouns referencing paragraph two (“this misalignment”) and assumes the reader has absorbed the setup in paragraphs one through three.

An AI system extracting that passage gets an incomplete answer that depends on surrounding context it can’t include in a 150-word citation block. So it doesn’t cite it. It moves on to content where the insight is front-loaded, self-contained, and extractable as a standalone passage.

The common mistake: The most common B2B content mistake in 2026 is restructuring for AI visibility after the fact — adding FAQ sections at the bottom of articles, inserting bullet lists between existing paragraphs — without changing the fundamental writing pattern. Surface-level structural changes on top of an argument-first writing style don’t fix the passage-level extractability problem. The restructuring has to happen at the sentence and paragraph level, not just at the section level.

ChatGPT, Perplexity, and Google AI Overviews Cite Differently

Before getting into the practical changes, there’s an important nuance: not all AI engines select citations the same way. If you’re optimising for AI visibility, understanding which engine weights which signals is worth the five minutes it takes to internalise.

 

Engine	What it prioritises	B2B implication
Google AI Overviews	Traditional SEO signals + extractable answer blocks. Strong correlation with top-10 organic rankings.	Your SEO foundation still matters here — but content also needs passage-level extractability on top of it.
ChatGPT / OpenAI Search	Comprehensive, authoritative guides with clear entity definitions. Favours older, well-established sources on a topic.	Long-form, substantive content with named frameworks and defined terminology performs well. Recency matters less than depth.
Perplexity	Recency, structured data, domain authority on the specific topic. Treats citations as a first-class feature.	Most favourable engine for newer content and smaller domains. Structure and freshness are high-leverage here.
Claude (via Brave Search)	Structured, substantive content with clear sourcing. Conservative citation volume but high selectivity. Content acknowledging limitations earns 1.7x citation boost.	Intellectual honesty — naming trade-offs, acknowledging complexity — is specifically rewarded.

 

Only 11% of domains get cited by both ChatGPT and Perplexity. Treating all AI engines as a single system is a strategy for being optimised for none of them. The pragmatic approach: start with Perplexity (most accessible for newer domains, highest growth rate for B2B) and Google AI Overviews (most search volume, most pipeline impact), then layer in ChatGPT optimisation as your domain authority builds.

The 5 Structural Changes That Actually Move Citation Rates

These aren’t tactical tweaks. They’re changes to how you write. Once they’re built into your default approach, the incremental effort per piece is small. The hard part is the first ten pieces where you’re breaking old habits.

1. Front-load the answer in every section

The most important structural shift: put your clearest, most complete answer to the section’s implied question in the first two sentences. Not after three sentences of setup. Not at the end of the paragraph as a conclusion. In the first 40 to 60 words.

The research is unambiguous on this: 44% of all LLM citations come from the first 30% of a page’s text. The opening of every section is prime citation real estate. Your instinct as a writer is to build to the point. AI citation rewards getting to the point first and supporting it after.

Practically: before you write any section, write the one-sentence answer to the question that section addresses. That sentence goes first. Everything after it is elaboration and support.

Before and after example: BEFORE (argument-first): “There are several reasons why in-house marketing teams struggle to generate pipeline. The first relates to how they’re measured. Marketing teams are typically held to MQL targets, which creates an incentive to optimise for volume over quality…”AFTER (answer-first): “In-house marketing teams structurally struggle to generate pipeline because they’re measured on metrics that don’t map to revenue — MQL volume, website sessions, impressions — rather than on pipeline contribution. This measurement misalignment means teams optimise for the wrong output…”

2. Write in self-contained passage units of 120 to 180 words

AI systems extract content in passage-sized chunks — typically 120 to 180 words. Pages structured into sections that fall within this range receive 70% more ChatGPT citations than pages with shorter, fragmented sections. The sweet spot identified across multiple 2025 citation studies is 134 to 167 words per key answer section.

This doesn’t mean every paragraph needs to be exactly 150 words. It means each subsection — the content between two H3 headings — should function as a complete, standalone answer to an implied question. If you extract that 150-word block and show it to someone without any surrounding context, they should be able to understand it completely.

Test your own content against this: take any section from your best-ranking post, copy just that section, and ask whether someone who hasn’t read the rest of the article can understand the full point. If the answer is no — if the passage depends on setup from earlier, uses pronouns referencing prior content, or ends with an implication rather than a conclusion — it’s not passage-ready.

3. Replace qualitative claims with quantified, sourced statements

Quantitative claims receive 40% higher citation rates than qualitative statements across AI systems. This is the highest-leverage single change most B2B content writers can make.

“Marketing teams often struggle to generate pipeline” is a qualitative claim. It might be true. It might be insightful. It will rarely be cited.

“79% of marketing-qualified leads never convert to sales opportunities, according to HubSpot’s 2025 State of Marketing report” is a quantified, sourced statement. It makes the same point. It has specific numbers. It has a named source. It tells the AI system exactly what kind of claim this is and how verifiable it is. Citation rate goes up substantially.

For demand.consulting specifically, this means building a habit of: (a) citing specific data points when making factual claims, (b) sourcing them with the organisation and year, and (c) developing original data where possible — client case study results, programme benchmarks, diagnostic findings — since proprietary data is the highest-citation-probability content that exists. You can’t be the second source if you’re the only source.

4. Use entity-rich, self-defining terminology

AI systems use entity recognition to understand what a passage is about and assess whether it’s authoritative on a specific topic. Pages with 15 or more recognised entities — named concepts, tools, frameworks, organisations, techniques — show a 4.8x citation boost compared to pages with sparse entity presence.

For B2B demand gen content this means: name your frameworks. Name the tools. Reference specific methodologies. Define technical terms inline rather than assuming the reader knows them. “Signal-triggered outbound” is better than “outreach triggered by intent signals.” “Clay enrichment waterfall” is better than “using a tool to find email addresses.” Specific named entities are what AI systems use to understand the topical authority of the content they’re extracting from.

It also means creating named frameworks of your own. If Lewis consistently refers to the “4-channel ABM system” in his content, and that phrase appears across multiple posts with consistent definition, AI systems begin to associate that named framework with demand.consulting as the entity that defines it. That association compounds — each time the framework is cited, the authority signal on that term strengthens.

5. Acknowledge limitations and trade-offs explicitly

This one is counterintuitive. Most B2B content is written to be persuasive — to make the strongest possible case for a position, without giving ground to alternatives or complications. That’s good marketing instinct. It’s poor AI citation behaviour.

Claude in particular — whose user base skews heavily toward professionals and enterprise decision-makers, making it disproportionately valuable for B2B — specifically rewards content that acknowledges limitations. Content that explicitly names trade-offs, caveats, or contexts where an approach doesn’t work receives a 1.7x citation boost from Claude compared to equivalent content that doesn’t.

More broadly, AI systems are trained to provide balanced, nuanced answers. Content that mirrors that quality — that says “this approach works well in X context but has limitations in Y context” — is structurally better matched to what AI citation systems are trying to cite. Write like you want to be quoted by someone who values intellectual honesty. That’s exactly who’s doing the quoting.

The Practical Rewrite Framework: Where to Start

If you’re looking at an existing content library and wondering where to apply this, here’s the sequence I’d recommend.

Step 1: Audit your top 10 traffic pages first

Start with the pages that already have organic traction — they have the domain authority signals that AI systems use as a baseline trust filter. Restructuring these pages for passage-level extractability is higher leverage than creating new content from scratch, because you’re adding AI citation potential to pages that already have the prerequisite authority signals.

For each page, ask three questions: Does each section open with the direct answer? Are sections self-contained enough to be understood without surrounding context? Does the content make quantified, sourced claims or qualitative assertions?

Step 2: Identify your “invisible” passages

Run your 10 most important queries through ChatGPT, Perplexity, and Google AI Mode manually. Note where you do and don’t appear. The queries where competitors are being cited but you aren’t are your highest-priority rewrite targets — those are queries where your content exists and ranks, but isn’t structured well enough to be extracted.

This audit takes about two hours and produces a clear prioritised list. It’s more useful than any tool-based citation tracking because it shows you the exact queries where the structural changes will have immediate impact.

Step 3: Rebuild, don’t add on

The temptation when retrofitting content for AI citation is to add things: an FAQ section at the bottom, some bullet lists between existing paragraphs, a table summary of the key points. These surface-level changes help at the margins but don’t fix the underlying problem if the body content is still written in an argument-first, context-dependent style.

The sections that need the most work are the ones where your key insights are buried in the middle of paragraphs, depend on prior context, or are expressed qualitatively rather than with specific data. Those sections need to be rewritten from the first sentence — not decorated with a bullet list at the end.

What the New Content Brief Looks Like

For any new piece of B2B content going forward, the brief should include two things that most content briefs don’t currently include:

1. The specific query this content should be cited for — not the keyword, but the actual question a buyer would ask an AI that should surface this content. That question shapes every structural decision: the opening sentence of each section, the evidence that needs to be included, the length and format of key passages.
2. The “extracted passage” test — for every major section, a written version of what that section would look like as a 150-word standalone answer to that question. Writing this before writing the full section forces the answer-first structure, because you’ve already established what the key point is.

 

It’s a different writing discipline. It takes slightly longer at the brief stage. It saves significant time in rewrites because the structural decisions are made upfront rather than retrofitted after the fact.

The Honest Caveat: SEO Is Still the Foundation

I want to be clear about something that gets lost in “AI citation optimisation” content: traditional SEO is not irrelevant to AI visibility. It’s the prerequisite.

76% of Google AI Overview citations come from pages that rank in the top 10 organic results for that query. Strong organic SEO is what puts you in the candidate pool for AI citation. GEO — the structural and content changes covered in this post — is what converts that candidacy into an actual citation.

The practical implication: you can’t skip SEO and go straight to GEO. Domain authority, technical health, internal linking, and backlink quality all matter — not because they directly determine AI citation probability, but because they determine whether your content gets considered at all. The citation selection happens among a set of candidates that the retrieval system has already assembled. Getting into that candidate set is still largely an SEO problem.

The relationship between SEO and GEO: Think of traditional SEO as getting your content in the room. GEO is what determines whether it gets cited once it’s there. You need both. Companies that abandon SEO for GEO will find their citation rates declining over time as their pages fall out of the candidate pool. Companies that invest in SEO without GEO will rank well and still not appear in AI answers. The compound effect comes from doing both — building the authority that gets you considered, and structuring the content to earn the citation.

What This Means for How demand.consulting Content Is Built

The ABM series on this blog — the nine posts on the 4-channel system — was written for human readers. Engaging, argument-first, content that builds a case across sections. It ranks reasonably well. But most of it isn’t structured for AI citation.

That’s going to change. Starting with this series, every post is written with passage-level extractability built into the first draft. Every major claim is quantified and sourced. Every section opens with the direct answer. Named frameworks are used consistently and defined on first use.

The ABM series will be restructured over the next two months — the top-traffic posts first, then the rest. Not wholesale rewrites. Targeted structural changes at the passage level: front-loading insights, adding specific data points to qualitative claims, breaking long argument-sections into self-contained answer units.

The goal isn’t to optimise the content for AI systems at the expense of human readers. The goal is content that earns trust from both — because the qualities that make content AI-citable (specific, substantive, answer-first, intellectually honest) are also the qualities that make content genuinely useful to a senior B2B buyer who doesn’t have time for slow-building arguments.

“The content that gets cited by AI and the content that earns trust from a CRO are the same content. Specific, substantive, answer-first, honest about trade-offs. Write once, earn both.”

The Summary — Five Changes, One Direction

To get cited by AI systems, B2B content needs five structural changes. Every section should open with the direct answer — not after setup, not as a conclusion. Key sections should be self-contained passage units of 120 to 180 words. Qualitative claims should be replaced with quantified, sourced statements. Terminology should be entity-rich and self-defining. And trade-offs should be acknowledged explicitly, not glossed over in pursuit of a cleaner argument.

None of these changes are about gaming an algorithm. They’re about writing content that’s genuinely more useful — to a senior B2B buyer who’s time-poor and evaluating quickly, and to an AI system that’s trying to provide the most accurate, trustworthy answer to a research question.

The overlap between “content AI wants to cite” and “content a CRO finds worth reading” is larger than most B2B content strategies currently acknowledge. Closing that gap is the work.