How to create content LLMs actually surface: 800+ links audited

Introduction

For years, inbound strategy centred around one dominant goal: ranking on the search engine results page (SERP). But that world is disappearing. Buyers no longer rely solely on Google to learn, compare, or validate products. Discovery is increasingly happening off-site, inside AI assistants, chat-based search tools, summaries, social feeds, communities, YouTube explainers, and analyst-style GPTs.

To understand how LLMs actually retrieve, cite, and recommend content in these environments, we ran a comprehensive audit of Cognism’s entire content library, which included hundreds of URLs spanning definitions, outbound guides, demand gen frameworks, explainers, and evaluative content.

Rather than relying on assumptions or generic “LLM content optimisation” theories, we analysed how our real pages perform within these systems, supported by external studies that confirm or challenge our findings. Using Ahrefs, Screaming Frog, and Dreamdata, we reviewed more than 800 links across AI citations, traffic value, internal links, backlinks, authorship, schema, topic clusters, and core update behaviour.

Our goal was simple: What types of content do large language models trust enough to surface during real B2B buying journeys and why?

Below are the key findings from the audit.

Data methodology

Rather than relying on assumptions or generic “LLM optimisation” theories, we sought hard data on how our content actually performs within these systems. Backed by other studies out there, too, to support or challenge our findings.

Proprietary data sources:

• Ahrefs: AI citations, traffic value, estimated monthly organic traffic, and backlinks.
• Screaming Frog: Internal links, topic clusters, and page format.
• Dreamdata: Conversion performance, influenced MQLs, pipeline, and revenue.

The audit analysed a wide range of signals, including:

• AI citations: how often our pages are used as source material in model responses.
• Organic traffic & value: whether AI-visible pages also perform with human searchers.
• Backlinks & internal links: how external authority and site structure influence retrievability.
• Page format: video usage, schema, authorship, depth, and content freshness.
• Topic clusters: which categories LLMs consider authoritative.
• Google Core Update performance: how different content types behaved across the March, June, and August algorithm updates.
• Conversion alignment: whether highly cited pages also convert better.

Because Cognism produces a large volume of definition-style, educational, and evaluative content, our library is rich with the kinds of pages that AI systems use to explain complex topics. 

This provides us with unusually clear visibility into what modern search and LLM ecosystems value.

Ultimately, this wasn’t a theoretical exercise. It was a practical investigation into a critical question: What types of content do large language models trust enough to surface during real B2B buying journeys, and why?

By analysing our own pages, we can now see distinct patterns emerging around:

• What LLMs prioritise.
• Which authority signals matter most.
• How retrievability is determined.
• And why traditional SEO signals alone no longer tell the whole story.

These findings form the foundation of the playbook that follows, one rooted entirely in Cognism’s real-world data, not guesswork, observed across thousands of sessions, pages, clusters, and citations.

Key findings about how LLMs surface content

1. AI citations are now a leading indicator of in-model authority

Our pages are frequently cited by LLMs when they:

• Drive traffic.
• Generate higher traffic value.
• Also rank well in Google.
• AI citations correlate more strongly with performance than any other factor.

Recent studies show a growing divergence between Google rankings and LLM citations, meaning certain LLMs prioritise different signals. For example, analysis from Search Engine Journal highlights that some models surface sources Google ranks lower, creating a visibility gap marketers must account for.

What this means: AI citations shouldn’t be viewed in isolation, but as part of a broader signal set that includes referral traffic, log-file activity, and search visibility. Together, they show how often and confidently LLMs draw from your site.

2. Educational, definition-style content is the top performer

Our most AI-visible and highest-traffic pages are:

• Evergreen explainers.
• “What is…” definitions.
• Category primers and fundamentals.

These pages benefit from EEAT updates and serve as a source material for LLM responses.

AI is increasingly answering definition-style queries directly, which reduces some TOFU traffic. But that doesn’t diminish the value of creating these pages - it changes why you make them. There’s no point in creating them, for example, if you don’t bring anything new to the conversation. New perspectives, data-backed studies, and big ideas are crucial for the new world of TOFU content.

Here’s the nuance:

• Even when AI “eats” clicks, these pages still drive strong brand visibility inside LLMs, which is crucial in a zero-click environment.
• We haven’t seen a collapse in traffic to these pages this year; instead, a shift in where attribution is visible (more via citations/referrals, less via classic organic).
• These TOFU assets act as the training data models pull from. If you don’t create them, someone else supplies the definition — and owns the category.
• We expect AI-driven reduction in TOFU clicks to continue, but the brands with the best educational content will dominate AI discovery, not just SEO.

What this means: TOFU still matters, not for volume, but for authority, category ownership, and in-model visibility. You’re no longer optimising for clicks alone; you’re optimising to become the source LLMs rely on.

3. Video meaningfully boosts visibility and authority

Pages containing video saw:

• 3.1× more AI citations.
• 2.4× higher traffic value.
• 1.8× more organic traffic.

Ilse Van Rensburg, Senior SEO and Content Manager at Cognism, shared her findings on LinkedIn:

External research shows a similar pattern:

• Wistia’s 2024 Video Index Report found that pages with video drove up to 2.3× longer time-on-page, which is a signal both LLMs and search engines use as a proxy for content quality.
• BrightEdge (2024) reported that video-led pages appeared in 65% more universal search results, increasing multimodal visibility across SERPs.
• Google’s own documentation emphasises that multimodal content (text and video) provides stronger contextual grounding for ranking systems and AI overviews.
• Studies analysing Gemini and ChatGPT retrieval behaviour show that pages containing mixed media, especially video, are surfaced more often in LLM answers because they signal higher authoritativeness and clearer instructional quality.

Why this matters: Video isn’t just improving human engagement, it’s improving machine understanding. LLMs are treating video as a reliability signal, and Google is increasingly weighting multimodal results in both traditional rankings and AI Overviews.

4. Backlinks remain the strongest signal of external authority on the open web

Highly AI-cited pages have nearly six times more backlinks than poorly performing pages. 

External authority still outweighs internal structure in predictive power.

5. Internal linking is an important secondary signal

Pages with strong internal links:

• Are more likely to receive AI citations.
• Perform better in EEAT-related updates.

Top pages averaged 35-45 internal links, versus a median of 20-25.

Internal linking helps both Google and LLMs map the relationships between your pages, essentially teaching them which concepts belong together and which pages carry the most authority.

According to Search Engine Land’s internal linking guide, strong internal linking helps by:

• Clarifying topical relationships, making your content easier for models to interpret
• Improving crawlability, increasing the likelihood that high-value pages are indexed and retrieved
• Passing authority across your site, reinforcing priority pages
• Reducing orphan content, which LLMs are less likely to surface

What this means: Internal links aren’t just navigational; they’re structural signals that improve retrieval, authority distribution, and in-model visibility.

Ilse added:

“Backlinks and internal linking have always been essential components of a good SEO strategy, but our data now shows that they’re integral for LLM visibility too. Backlinks build authority. Internal links build context. And both increase your chances of becoming AI-visible and human-relevant.”

6. E-E-A-T–aligned content grew the most across core updates

June’s E-E-A-T update drove:

• +52% traffic increase.
• +113% more AI citations.

Winners were defined as pages with definition sections, data-backed explainers, video-supported articles, and pages featuring strong authorship and schema markup.

When we reference “strong authorship,” we’re talking about more than simply having an author bio page, which Cognism already does. In the context of Google’s E-E-A-T systems and LLM retrieval, “strong authorship” means:

• Clear, visible author attribution on the page (name, role, expertise).
• Evidence of real-world experience, e.g., domain-specific job titles, lived expertise, professional background.
• Author pages that demonstrate credibility, including links to other articles, LinkedIn profiles, and relevant work.
• Consistent authorship across related content, which helps both Google and LLMs connect expertise to topic clusters.
• Schema markup (Author, Article, and Organisation) to reinforce who created the content and why they are qualified.
• A track record of publishing authoritative content, which models use as a signal of trustworthiness.

Ilse said:

“Every core update is a feedback loop. Monitor them closely, and you’ll see exactly what’s performing and what you need to work on. For Cognism, I found that our informative content is what’s driving performance. However, we have a few pages that need work, especially as we enter this next era of search.”

7. Thin, templated, or outdated content was penalised

The March and August updates heavily impacted:

• Short blogs.
• Generic sales/prospecting content.
• Keyword-stuffed articles.
• Pages with weak linking or high outbound link ratios.

These clusters saw traffic and visibility decline.

8. Some topic clusters are commercially stronger than others

Top-performing clusters across traffic, value, and citations:

• Prospecting.
• Lead generation.
• Email.
• Demand generation.
• Data.

Lowest-performing clusters:

• Compliance.
• RevOps.
• GTM.
• AI and automation.
• Weakly developed demand generation pages.

9. AI visibility correlates with conversion rate

There is a consistent (though modest) correlation:

• r = 0.24 between AI citations and conversion rate.
• r = 0.36 between traffic value and conversion.

This suggests that pages trusted by AI systems tend to convert more effectively.

Ilse said:

“AI visibility is fast becoming a soft proxy for authority. When it sits alongside clear content intent and the right CTAs, it may be one of the strongest predictors of future performance.”

10. If your overall site health is good volatility is isolated

Pages unaffected by recent core updates held steady, showing:

• No structural issues.
• Strong baseline quality.
• Volatility is limited to thin or weak clusters.

What does this data tell us?

Taken together, the audit findings point to a larger shift happening beyond our own content performance: buyers are no longer learning on websites first; they’re learning off-site, within AI ecosystems. In 2026, zero-click search will continue to become the norm.

The patterns we observed, from the dominance of definition pages, to the rise of AI citations, to the outsized impact of video and internal linking, are not random SEO anomalies. They are symptoms of a fundamental change in how buyers educate themselves.

Today, early-stage discovery is happening through:

• AI chat tools that explain concepts, map categories, and compare vendors.
• Summaries and overviews that collapse whole SERP journeys into a single answer.
• YouTube, TikTok, and podcasts are where foundational knowledge is consumed.
• Slack groups, LinkedIn posts, and communities where peers share shortcuts and recommendations.
• Analyst-style GPTs that interpret frameworks, product types, and buyer considerations.

Instead of searching via just keywords, buyers now ask longer, more conversational questions like:

• “Explain demand generation at a high level.”
• “What does good intent data look like?”
• “Who are the top GDPR-compliant data providers?”
• “How does Cognism compare for EMEA coverage?”

This behaviour aligns directly with what our audit revealed.

1. AI has become the front door of education

Models draw from the most structured and authoritative sources available, which explains why pages with strong schema, entities, authorship, and internal linking tend to outperform.

2. Discovery no longer happens page-by-page

LLMs break your content into smaller “knowledge chunks”: definitions, frameworks, examples, and relationships. Pages that make these clear stand out.

3. Authority now comes from context, not keywords

LLMs prefer content that is complete, clear, deeply interlinked, and enriched with human expertise, which is exactly what the top-performing pages in our audit delivered.

4. AI citations are emerging as a new trust signal

Our strongest pages in organic search are also the ones most frequently cited by AI systems. In other words, the content AI trusts is increasingly the same content that Google trusts.

Ilse said:

“AI visibility is becoming a critical part of any modern SEO strategy, but it doesn’t replace the fundamentals. While LLMs increasingly shape early-stage discovery, users still click, read, and evaluate content on real pages. That means the basics still matter: write for humans, optimise for experience, and treat AI visibility as an extension of good SEO, not a shortcut around it.”

The takeaway is simple but critical: The buyer journey now begins before the website visit, inside the systems trained on your content. Your pages are no longer competing only in Google’s index. They’re competing for trust inside LLMs, where buyers form their first impressions and foundational understanding long before they land on your site.

How to write content that LLMs will surface in decision journeys

Writing for LLMs isn’t about gaming the system. It’s about creating content that is so structured, helpful, clear, contextually linked, and expert-driven that models consistently choose it over alternatives.

Here’s the emerging framework.

1. Start with the buyer’s question, not the keyword

LLMs answer natural language queries. They prioritise clarity and directness over keyword density. Each page should clearly address:

• What the buyer wants to understand.
• How the concept works.
• Why it matters.
• Real examples.
• Adjacent concepts the model might care about.

Pages that begin with a crisp definition or explanation tend to appear more frequently in AI answers.

2. Make every page “entity-rich”

LLMs retrieve information based on concepts, relationships, and semantic clarity. To make content more retrievable, include:

• Key entities (e.g., SDR, ICP, GDPR, enrichment data).
• Related processes and frameworks.
• Step-by-step explanations.
• Adjacent topics.
• Real scenarios or use cases.

Think of entities as the hooks that AI systems use to match your page to a buyer’s question.

3. Add 3-5 contextual internal links per 300 words

This is one of the strongest findings in the audit. High-performing pages contain significantly more internal links than those with low AI visibility. The best links:

• Sit within the paragraph.
• Connect directly related concepts.
• Help establish topic clusters.
• Reflect on how buyers actually learn.

This structure helps LLMs understand the relationships between your pages and strengthens your content’s “retrievability graph.”

4. Prioritise authorship, quotes, stories, and real expertise

Models prefer content that clearly signals who wrote it and why they’re credible. Pages that include:

• Named authors.
• Examples drawn from real data.
• Commentary from SMEs.
• Industry-specific insights.

…are more likely to be chosen when a buyer asks, “How should I think about this?” or “Give me guidance.”

5. Add video wherever the topic supports it

If a page covers a high-intent topic, such as category definitions, evaluators, comparisons, decision frameworks, or tool overviews, adding a video can significantly enhance both AI visibility and organic performance.

Video is effectively an “expertise amplifier.”

6. Add schema markup across all pages

Schema helps LLMs interpret the structure of your page and extract the correct knowledge. FAQ schema, How-To schema, VideoObject schema, and Author schema are especially valuable.

7. Write with clarity, structure, and completeness

AI-visible content tends to share the same characteristics:

• It answers the core question in the first 2–3 sentences.
• Paragraphs are short and logical.
• Concepts are broken into steps, frameworks, or examples.
• The page builds a clear narrative or argument.
• There’s very little fluff.

This doesn’t just help users, it helps models understand and reuse your content.

8. Identify and refresh any pages that fit the “thin content” profile

Pages under ~800 words, with few internal links, minimal authorship, outdated information, or heavy reliance on generic keywords will become increasingly vulnerable. 

A simple refresh that adds depth, examples, FAQs, links, and video often restores performance fully.

Ilse said:

“The thing about writing for LLMs is that it isn’t really a new playbook; it’s just a sharper version of what good content has always been. If your pages are clear, structured, genuinely helpful, and linked in a way that reflects how people actually learn, models will naturally surface them. You don’t need tricks. You just need to make your content the easiest, most complete answer in the room for both humans and AI.”

What this means for content, SEO, and growth teams

The shift to LLM-first discovery changes the role of content in the entire go-to-market engine.

Every article, video, and definition contributes to the knowledge base that LLMs draw from when they answer queries about your category.

Your audit found a strong alignment between:

• AI citations and referral traffic.
• Organic traffic.
• Traffic value.
• Conversion rate.

These should become KPIs in your content and SEO dashboards.

3. Category leaders will take control of the narrative in LLMs

Winning search in 2026 is about taking control of the narrative for your brand. It’s not just about building authority for what you want to be known for on your own domain, but also off your domain, too. The complexity of LLMs vs traditional search is far greater, but this also presents an opportunity for brands to double down on what they do well. 

4. Search rankings become the downstream effect not the goal

Both Google and LLMs are rewarding similar signals: expertise, structure, authorship, depth, and multimedia evidence. Search rankings are far from irrelevant, especially when you consider the web feeds the answers you find in LLMs, and Google's market share of the search market is still huge.

5. Content teams need a new approach to briefs

Instead of starting with “What keyword are we targeting?”, a modern brief should begin with:

• What question are we answering?
• What entities must appear?
• Which internal pages should this connect to?
• What POV or data makes this uniquely ours?
• What video or visual should support the page?
• How do we signal authorship and expertise?

This is what inbound looks like now.

The last word

Your website is no longer just the place buyers come to learn. It’s the database that teaches the system’s users how to research.

Your content now serves two audiences:

1. Humans.
2. The LLMs humans trust to make decisions.

The companies that understand this and build content that is clear, expert-led, structured, and richly connected will become the dominant voices inside every model, search tool, and assistant.

Those who continue to publish thin, isolated, keyword-led content will gradually disappear from the new discovery ecosystems.