Filmfolk had page-one Google rankings for "corporate videographers London." Strong backlink profile. Well-optimized content. By every conventional SEO metric, they were doing well.
When we ran their baseline AI audit, they were cited in zero of 63 tracked prompts across ChatGPT, Claude, Perplexity, and Gemini. Not once.
This isn't unusual. It's the norm. And it points to a fundamental misunderstanding about how AI systems retrieve information — one that most SEO practitioners haven't fully internalized yet.
AI Quick Summary: SEO optimizes web pages for a document ranking system (Google). AI visibility requires engineering a brand's entity representation in a retrieval system (ChatGPT, Claude, Perplexity, Gemini). These are different problems with different signals, different metrics, and different methodologies. A brand can rank #1 on Google and be cited in 0% of AI answers — as Filmfolk demonstrated. The fix is not "better SEO." It's structured Brand Memory, entity confidence engineering, and source trust reinforcement.
Two Different Retrieval Systems
Google's search engine is a document ranking system. It crawls pages, indexes content, evaluates authority signals (backlinks, E-E-A-T, technical quality), and returns a ranked list of URLs. The goal is to surface the most relevant document for a given query.
AI language models with retrieval capabilities — ChatGPT, Claude, Perplexity, Gemini — work differently. They don't return documents. They generate answers. And to generate accurate answers, they retrieve from a knowledge layer that is structured around entities, not pages.
The distinction matters enormously:
- Google ranks pages. AI systems retrieve entities.
- Google rewards backlinks and content volume. AI systems reward entity clarity and source trust.
- Google measures clicks and rankings. AI systems measure citation rate and brand mention frequency.
- Google's signals are relatively stable. AI retrieval signals decay as models update.
What AI Systems Actually Look For
When a buyer asks Perplexity "who are the best corporate video agencies in London," the model doesn't crawl Google's index. It draws on its training data and, in real-time retrieval systems, on a curated set of sources it has been configured to trust.
What determines whether your brand gets cited? Three things:
1. Entity Confidence
The model needs to understand who you are, what category you belong to, and what role you play. This is entity confidence — the clarity with which an AI system can place your brand in the right context. If your brand-role-category relationship is ambiguous or inconsistent across sources, the model won't cite you with confidence. It will cite a competitor whose entity signals are cleaner.
2. Source Trust Hierarchy
Not all sources are weighted equally in AI retrieval. Owned content (your website) carries some weight. Earned coverage (press, reviews, third-party mentions) carries more. Structured data (schema markup, knowledge graph entries) carries significant weight. Expert citations (academic papers, industry reports) carry the most.
Traditional SEO focuses heavily on owned content and backlinks. AI retrieval rewards a weighted hierarchy of source types — and most brands have significant gaps in the higher-trust tiers.
3. Freshness and Consistency
AI models weight newer signals more heavily than older ones. A brand that was well-represented in training data two years ago may have decayed in retrieval priority as the model has been updated with newer information. Maintaining consistent, fresh signals across all source types is an ongoing requirement — not a one-time optimization.
The core insight: SEO optimizes for a document ranking system. AI visibility requires engineering a brand's entity representation in a retrieval system. These are different problems that require different solutions.
Why Good SEO Can Actually Mislead You
There's a dangerous assumption that strong SEO performance implies strong AI visibility. It doesn't — and in some cases, SEO-optimized content can actively work against AI retrieval.
SEO content is often written to rank for keywords. It's structured around search intent, not entity clarity. It may be technically excellent but entity-ambiguous — meaning an AI model reading it can't confidently extract a clear brand-role-category relationship.
Filmfolk's website was well-optimized for Google. But the content didn't clearly establish: Filmfolk is a London-based video production agency specializing in corporate event coverage and brand storytelling for enterprise clients. That level of entity specificity — consistently reinforced across multiple trusted source types — is what AI retrieval requires.
What Actually Works
AI visibility engineering requires a different methodology:
- Prompt Reality Audit — test the actual prompts your buyers use and measure your current citation rate across all major AI models
- Entity Signal Mapping — identify gaps in brand-role-category clarity across all source types
- Brand Memory Construction — build a curated, weighted knowledge base that AI retrieval systems can read and trust
- Source Trust Reinforcement — strengthen signals in the higher-trust tiers (earned coverage, structured data, expert citations)
- Continuous Measurement — rerun the same prompts, measure citation delta, adapt as models update
After applying this methodology to Filmfolk, their citation rate went from 0% to 81% across 63 tracked prompts in 90 days. Their Google rankings didn't change. Their AI presence transformed completely.
The Practical Implication
If you're investing in SEO and assuming it covers AI visibility, it doesn't. The two disciplines can coexist and reinforce each other — but they require separate, deliberate work.
The good news: AI visibility is an engineering problem, not a content volume problem. It's measurable, improvable, and — unlike Google rankings — not subject to opaque algorithm updates. The signals are more transparent, the methodology is more precise, and the results are directly attributable. Google's own AI Optimization Guide (2026) confirms this shift toward Grounding and entity-based retrieval.
The Technical Anatomy of the Filmfolk Case
Let's open the Filmfolk case with a technical eye, because the "good SEO but zero citations" picture is more common than assumed. The site audit showed: page titles were keyword-focused ("corporate videographers London" variations), content was clustered by search intent, the backlink profile was solid, Core Web Vitals were green. A flawless picture — for Google.
But the retrieval-side picture differed. First, the brand-role-category triple was never established in one clear sentence anywhere. The site said "video production," one directory wrote "media agency," another said "event filming." Instead of reconciling three definitions, the model turned to the competitor with one. Second, page content wasn't question-and-answer shaped; long paragraphs suited the search engine, but quotable passages the model could cite didn't exist. Third, third-party verification was weak: press mentions, sector reports, structured data — the layers retrieval trusts — stood empty.
The fix sat on the same three axes: the definition was fixed in one sentence and spread across all channels; pages were rewritten with question-and-answer passages; high-trust layers were built. The result went from 0% to 81% in 90 days — ChatGPT 84%, Perplexity 86%, Claude 78%, Gemini 76% by model. Google rankings barely moved in the same period. The two systems' independence can't be proven more clearly.
SEO Content vs Retrieval Content: Side by Side
To make the difference concrete, compare two writings of the same topic:
| Dimension | SEO content | Retrieval content |
|---|---|---|
| Primary goal | Rank for the query | Get quoted in the answer |
| Heading logic | Keyword + click | Question + clear answer |
| Paragraph shape | Long, flowing, intent-covering | Short, standalone, single-claim |
| Number usage | Supporting, scattered | Extractable, in one sentence |
| Definition sentence | Usually absent | The same sentence on every page |
| Success measure | Rankings, clicks, traffic | Citation rate, mention frequency |
Don't misread the table: the two columns aren't enemies. A good page does both — a question heading works in search and in answers, a clear passage reads and gets quoted. The problem is writing only the left column and assuming the right one follows on its own. It doesn't; the right column is a separate writing discipline.
In practice the most efficient approach isn't rewriting existing SEO pages from scratch but adding a retrieval layer: a 2–3 sentence passage answering the question at each page's top, critical claims stated with numbers in single sentences, the definition sentence added to the page. This addition doesn't harm SEO; in most cases it even improves time on page.
How the Two Disciplines Work Together
Frame the SEO and AI visibility teams' relationship as division of labor, not rivalry. The SEO team serves search demand: keyword research, technical health, link profile, content calendar. AI visibility work manages the answer layer: prompt set, entity consistency, citation measurement, adaptation to model updates. Both teams use the same content pool but ask different questions: one asks "will this page rank?", the other "will this page get quoted?"
- Content planning: every new piece gets planned with both a target query and a target prompt. Writers get two briefs: queries to rank for, questions to answer.
- Page templates: retrieval elements join the templates — question heading, clear answer passage, definition sentence, numbered claim blocks. Writers do no extra work; the template guides them.
- Reporting: the monthly report splits in two: search performance (rankings, traffic) and answer performance (citation rate, model breakdown). The two curves read side by side; when one falls while the other rises, the cause gets investigated.
- Model updates: after major model updates the citation round refreshes. What an algorithm update is on the SEO side, a model update is here: part of the routine, not a crisis.
The typical result in teams that build this model: the citation curve rises in the first 90 days while the SEO curve holds; after month 6, retrieval-ready pages start earning natural links too and both curves climb together. Independent in the short term, mutually feeding in the medium term.
To learn where you stand, start with a Citation Baseline; concepts are in the glossary, the full method in our Adaptive RAG methodology, the evidence in the Filmfolk case study. General questions are answered on the FAQ page. Strong SEO means a starting point worth celebrating — now carry the same discipline to the answer layer.
FAQ
Why doesn't traditional SEO fix AI visibility?
Traditional SEO optimizes for a document ranking system (Google). AI systems like ChatGPT, Claude, and Perplexity use entity retrieval — they cite brands based on entity confidence, source trust hierarchy, and factual proof points, not backlinks or keyword density. These are fundamentally different retrieval systems that require different optimization strategies.
What do AI systems look for instead of SEO signals?
AI systems prioritize three things: (1) Entity Confidence — how clearly the model understands who you are and what category you belong to, (2) Source Trust Hierarchy — the diversity and authority of sources that confirm your entity data, and (3) Freshness and Consistency — whether your entity signals are current and aligned across all sources.
Can SEO and AI visibility work together?
Yes. SEO and AI visibility are complementary disciplines that can reinforce each other. Strong entity clarity benefits both. However, they require separate, deliberate work — assuming SEO investment covers AI visibility is a common and costly mistake. Filmfolk maintained #1 Google rankings while achieving 81% AI citation rate by investing in both.
Related Articles
Why Ranking #1 on Google Doesn't Get You Cited in ChatGPT
Google rankings and AI citations are fundamentally different systems — here's what signals actually matter.
AEO vs SEO: The Complete 2026 Comparison
What's different, which matters more, and how to optimize for both AI citations and Google rankings.
What is Brand Memory?
How AI decides which brands to remember — entity relationships and source trust.
Entity Confidence for AI Citations
How AI models decide which entities to trust — and the 5 dimensions you can engineer.
About the author
Bora Kurum is the founder of RAG Signal and a Ph.D. researcher at Istanbul Bilgi University, where his work focuses on LLM retrieval behavior and brand representation in generative AI systems. Read more →

