The Retrieval Bottleneck: Why Your AI Strategy is Still a Search Strategy

Nom du conférencier | Société | Fonction The Retrieval Bottleneck
...Why Your AI Strategy is Still a Search Strategy

SEO & GEO Summit 2026 Dawn Anderson | Bertey |
Founder & SEO Consultant Every company now claims an AI strategy. Billions invested in foundation models Endless discussions about: • Tokens • Model size • Prompts • Temperatures The AI Goldrush

Founder & SEO Consultant What Everyone Talks About Most AI conversations focus on: • Prompt engineering • Model parameters • Fine-tuning • LLM benchmarks But very few people talk about… RETRIEVAL

Founder & SEO Consultant ...is the true engine behind AI answers Understanding Why Information Retrieval (IR)

Founder & SEO Consultant Who is Dawn Anderson? • SEO practitioner for almost 20 years • International SEO conference speaker since 2017 • Boutique agency owner • Enterprise SEO consultant • Information retrieval & AI search world interloper • Pomeranian lover ;P

Founder & SEO Consultant • an index • a retrieval engine • a ranking system Under every AI answer lies:

Founder & SEO Consultant LLMs cannot generate what they cannot retrieve. Failures usually come from: • missing documents • outdated data • poor structure • weak relevance signals The Retrieval Bottleneck

Founder & SEO Consultant The Hallucination Reality Common explanation: “LLMs hallucinate.” Reality: Often a retrieval failure.

Founder & SEO Consultant 'The Hallucination Narrative' Just filling in the probability gaps as the model sees fit in the face of uncertainty.

Founder & SEO Consultant A Bit Like 'GuesSEO' and the Misinformation Super Highway

Founder & SEO Consultant Open Book Analogy LLMs are like students in an open-book exam But the answers depend on: Which pages were in the book.

Founder & SEO Consultant What is RAG? Retrieval- Augmented Generation combines: • Search • Indexing • Language models

Founder & SEO Consultant Why RAG Exists LLMs alone have limitations • Static knowledge • Outdated training data • Hallucination risk RAG adds external retrieval

Founder & SEO Consultant The Return of IR Information Retrieval research now powers AI search.

Founder & SEO Consultant The RAG Pipeline

Founder & SEO Consultant Where Errors Actually Start Typical failure chain: • Missing documents • Weak retrieval • Incomplete context • Model guesses (educated guesstimating)

Founder & SEO Consultant Generation is the Last Step

Founder & SEO Consultant Retrieval Determines Truth AI answers depend on: • What was retrieved? • What was ranked highest (importance and consensus for LLMs)? • What entered context?

Founder & SEO Consultant The Retrieval Bottleneck AI systems fail when: • content isn’t indexed • context isn’t retrieved • relevance signals are weak

Founder & SEO Consultant Where SEO Fits SEO influences: • Crawlability • Indexing • Structure • Retrievability SEO is becoming even more foundational

Founder & SEO Consultant Sparse vs Dense Retrieval Two retrieval approaches with many models: • sparse (includes BM25) • dense (includes vector)

Founder & SEO Consultant Reality of Modern Hybrid Retrieval Modern search uses: hybrid retrieval BM25 + vector search

Founder & SEO Consultant 'Query Fan Out' AI systems expand queries into multiple searches. Works with the notion of 'recall' improvement to move from 'sparse' to 'dense' retrieval

Founder & SEO Consultant Why 'Fan Out' Exists Fan-out improves: • coverage • recall • context

Founder & SEO Consultant Example 'Fan Out' Query: “How does AI search work?” Subqueries: • RAG architecture • vector search • ranking models

Founder & SEO Consultant AI systems process content as: • chunks • embeddings • vectors NOT PAGES Documents Become Data

Founder & SEO Consultant Documents are broken into chunks Each chunk becomes a retrieval unit Poor chunking can cause: • fragmented context • incomplete answers Chunking

Founder & SEO Consultant LLMs have limited context windows Only a small number of chunks can be included.

Founder & SEO Consultant Ranking Still Matters Retrieval → Ranking → Context Ranking decides: • which chunks are used • which are ignored

Founder & SEO Consultant Semantic Collapse Large models risk semantic collapse... Meaning distinctions blur.

Founder & SEO Consultant Semantic Metadata As Navigation Semantic Metadata helps AI understand: • hierarchy • topic boundaries • relationships

Founder & SEO Consultant Traditional SEO Goal Traditional objective: Rank #1 Focus on: • Keywords • Backlinks • SERP positions

Founder & SEO Consultant The New Objective BE RETRIEVED... BE EXTRACTED... BE CITED... BE TRUSTED...

Founder & SEO Consultant The Retrieval Question As well as asking... "Can we rank this?" Ask... "Can we get this retrieved / extracted / cited?"

Founder & SEO Consultant Contextual Visibility Semantic Clarity Context Relevance Entity Relationships

Founder & SEO Consultant Contextual Retrievability Content must be: • structured • understandable • machine interpretable

Founder & SEO Consultant Structure is the Map Structure helps machines understand: • hierarchy • relationships • topic organisation

Founder & SEO Consultant Why Headings Matter Headings create: • semantic anchors • clear topical sections THOSE HEADINGS WERE NEVER JUST FOR DESIGN PURPOSES. Use them ALL down the hierarchy to h6

Founder & SEO Consultant Internal Links Internal links create: knowledge pathways & Hierarchical Navigable Small World Graphs (HNSW) 'Emulations of 'real world' entities and their connections

Founder & SEO Consultant Context Graphs These connected documents form 'context graphs' to aid to both relevance and disambiguation. "You shall know content by the pages it is connected to". Firthian approaches & improved retrieval reasoning

Founder & SEO Consultant Topic Networks Build topic clusters • Strengthen relationships and themes between pages • Semantic similarity reinforces via power of 'Firthian' linguistics

Founder & SEO Consultant Entity Based Content AI systems understand entities Content should clarify: • people • organisations • concepts

Founder & SEO Consultant Knowledge Graphs Knowledge graphs connect: • entities • documents • relationships

Founder & SEO Consultant AI Needs Fresh Data LLMs rely on retrieval for freshness

Founder & SEO Consultant The Freshness Problem Outdated content → outdated answers

Founder & SEO Consultant Retrieval and Time Retrieval systems evaluate: • relevance (precision) • authority (consensus / trust) • freshness (what can add to the closed book value?)

Founder & SEO Consultant The New SEO Mission Solid SEO as: Engineering retrievability (ranking optimisation and retrieval optimisation)

Founder & SEO Consultant Retrieval First Content Design content for: • human-first trust • focused-brevity • topical connections • structure • think 'inverted pyramid of journalism'

Founder & SEO Consultant Machine Readable Knowledge Content must be: • structured • semantically clear • connected

Founder & SEO Consultant Indexability Still Matters AI systems rely on: • crawlable pages • accessible content • stable URLs Watch out for 'crawl budget' comeback as a huge SEO topic

Founder & SEO Consultant AI Bots are Rampant Nosey Crows

Founder & SEO Consultant Retrieval Optimisation Focus on: • structure • metadata • entities • semantic similarity • topic focus • avoiding semantic drift

Founder & SEO Consultant Retrieval Friendly Content Include: • clear sections • descriptive headings • contextual explanations • multi-modal aspects

Founder & SEO Consultant Context Window Design Brevity and self-contained topics and sub-topics have never been more important. 'Chunking' matters but don't fall for what is potentially the modern day 'keyword stuffing'

Founder & SEO Consultant Don't 'Chunk' Spam Chunks and context windows are important for AI Search, but it's arguably not the job of SEOs to create tiny blocks of content.

Founder & SEO Consultant Solid Technical Foundation Everything rests on solid technical foundations: Signals include: • architecture • canonicalisation • linking • structured data

Founder & SEO Consultant The Retrieval Bottleneck AI systems fail when: retrieval fails 'Closed book' trained models are NOT enough Other knowledge / search MUST augment

Founder & SEO Consultant The New Visibility Model Retrievability → Context → Generation

Founder & SEO Consultant The Real AI Strategy A real AI strategy includes: • knowledge architecture • retrieval infrastructure • content design & structure

Founder & SEO Consultant Why SEOs Matter? SEOs already understand: • crawling • indexing • ranking • structure

Founder & SEO Consultant The Future of Search Search evolves from: documents → knowledge

Founder & SEO Consultant Key Takeaways 1 Retrieval is the biggest AI failure point 2 SEO shifts to retrievability 3 Knowledge architecture determines visibility

Founder & SEO Consultant The organisations that understand and optimise for retrieval will be the ones whose knowledge becomes visible in the AI search era.

Founder & SEO Consultant Thank you X - @dawnieando Bluesky - @dawnieando Linkedin – Ms Dawn Anderson Threads - @dawnieando Bertey.com

The Retrieval Bottleneck: Why Your AI Strategy ...

The Retrieval Bottleneck: Why Your AI Strategy is Still a Search Strategy

More Decks by Dawn Anderson

Other Decks in Marketing & SEO

Featured

Transcript