Why Rankings Alone No Longer Win – What Actually Decides Digital Visibility Today
Many websites rank at position 6. Many have solid backlink profiles. And yet they are never mentioned by AI systems like ChatGPT or Perplexity — not even once. This article explains why, and what counts from here on out.
"The Fan-Out Effect: What Happens Between a Query and a Citation"
AirOps partnered with Kevin Indig (Growth Memo) to map ChatGPT's complete retrieval pipeline – from user query to cited source. The study provides the most precise data yet on the on-page factors that drive AI citation. It is the starting point for this article – and equally, the reason to go further: what the study does not measure is often just as decisive for real-world project decisions as what it does.
→ Read the original studyWhat is the difference between SEO and GEO – and why does this question cut to the heart of the problem?
SEO determines whether a website gets into the game (gets found). GEO determines whether it gets chosen – by AI systems that synthesise answers rather than listing links.
The distinction sounds simple. In practice, it separates two entirely different disciplines – with different metrics, different measures, and different success indicators.
"Getting into the game"
- Build rankings
- Backlinks & mentions
- Brand / entity strength
- Technical cleanliness
"Getting chosen"
- Headings as precise questions
- Clear, focused answers
- Freshness & structure
- Depth over breadth
Both layers operate simultaneously – and optimising one without the other is the most common strategic mistake we identify in audits.
Why are websites with good rankings still not cited in AI-generated answers?
AI systems do not cite whoever ranks highest – they cite whoever delivers the most precise, most clearly structured answer to a specific question.
This is the central paradigm shift: search engines have historically rewarded authority. Generative AI systems reward answer quality and structure.
The AirOps study backs this up with concrete numbers: a page at retrieval position 1 has a 58% probability of being cited – at position 10, that drops to 14%. A factor of four. Pages whose headings closely match the query are cited 41% of the time. Pages with a weak heading match: just 29%.
Classic SEO texts – long, broadly written, keyword-dense – are poorly suited for machine answer synthesis. The study confirms: the word-count sweet spot for AI citation is 500–2,000 words. Pages exceeding 5,000 words underperform against pages with fewer than 500.
"The old rule was: whoever ranks, wins. The new rule is: whoever ranks and delivers the best answer structure, wins."
What is the biggest strategic mistake in GEO optimisation in practice?
Either only SEO is optimised (good rankings, no AI visibility) or only GEO (perfect content that nobody finds). Both one-sided approaches produce measurably poor results.
In our project experience since 1998, we see two distinct camps:
Traditional SEOs think exclusively in rankings, backlinks, and technical hygiene. AI visibility is not part of their control logic.
GEO enthusiasts produce structured, question-oriented content – without the authority base that AI systems need to include that content in the retrieval process at all.
The outcome in both cases: wasted budget with measurable visibility loss.
This is also the central misreading of the AirOps study. The data shows that domain authority and backlinks have no positive correlation with AI citation rate. But the study exclusively measures what happens inside retrieval – not how a page gets there in the first place. That is precisely where backlinks and authority remain decisive.
- ✕How pages enter retrieval at all (authority, backlinks as prerequisite – not as on-page signal)
- ✕E-E-A-T signals and brand recognition as trust factors
- ✕Visibility across other AI systems: Perplexity, Claude, Gemini, Microsoft Copilot
- ✕UX, funnel, and conversion – what happens after the citation
- ✕Multilingual and international visibility dimensions
- ✕Only heading-level similarity measured – no body-text scoring
How can a URL be clearly classified – and what is the right measure to take?
Every URL can be assigned to one of three types. The type determines the priority – not gut feeling.
| Status | Diagnosis | Core Measure | GEO Impact |
|---|---|---|---|
| Invisible Position >10 | No retrieval – content is not processed by AI | Backlinks, Digital PR, entity building, internal linking | Minimal while outside retrieval |
| Visible, not chosen Top 10, no AI citation | Authority present, answer structure missing | Headings as questions, focused answer blocks, FAQ + Schema | Very high – GEO is maximally effective here |
| Traffic without impact Citations present, no conversion | UX or funnel breaks user behaviour | UX optimisation, trust elements, clear offers | Not relevant – problem lies in the funnel |
This three-type matrix is the operative core tool in our URL-level analysis. It prevents measures being applied to the wrong lever.
What does a realistic budget allocation for modern visibility management look like?
Based on current project experience: 40–60% SEO/Authority, 20–40% GEO/Content, 10–20% UX/Conversion – shifted according to the URL's starting position.
The distribution is not a rigid model but a starting point. What matters is the URL-level diagnosis: a website with a thin backlink profile needs more authority investment. A website with solid domain strength but weak content formatting needs more GEO investment.
The complete system in three layers
Foundation (Authority): Backlinks, mentions, brand search, entity building – brings the website into retrieval. Often ignored by AI analysis reports, but it is the necessary precondition.
Engine (GEO/Content): Headings as questions, clear answer structure, freshness, content depth – determines whether the site gets cited. This is the layer most severely underestimated right now.
Output (UX/Funnel): Offer, messaging, conversion elements – determines whether visibility generates revenue. Almost always excluded from visibility audits.
What specific measures lead to a page being cited by AI systems?
Three measures work demonstrably: formulate headings as precise questions, place direct answers in the first two sentences after each heading, prioritise content depth over content breadth.
AI systems like ChatGPT, Perplexity, and Claude operate with Retrieval-Augmented Generation (RAG). They identify text passages that answer a user query – and cite those passages. What matters is therefore the local relevance of the text block, not the global strength of the domain.
The AirOps study provides precise benchmarks: pages covering 26–50% of fan-out sub-queries are cited more frequently than pages covering 100%. The optimal heading structure sits at 7–20 subheadings (H2–H4). Fewer than 4 subheadings underperforms against pages with no structure at all.
Checklist: GEO optimisation at text level
Heading = exact user query formulated as a question. First paragraph = direct, precise answer without preamble. Structure = one heading, one topic, one answer block. Length = 500–2,000 words (AirOps benchmark). Subheadings = 7–20 structural H2–H4s. Schema = FAQ and Article schema implemented. Freshness = date visible, regularly updated.
ERP Software for the Glass Industry
Starting situation: Position 6 in Google – solid. AI visibility in ChatGPT and Perplexity: zero. The page had not a single heading formulated as a question.
Measures taken: H1 reformulated as "What does ERP software do specifically for the glass industry?" · H2s as sub-questions · Answer blocks placed directly below each heading · FAQ schema implemented.
Result: Same rankings. Clearly measurable AI visibility for product-relevant queries – without a single new backlink.
Have backlinks become less important because of GEO and AI optimisation?
No. Backlinks are not less important – they are no longer the final step. They are the first: without authority, no page enters the retrieval process that GEO optimisation requires to have any effect at all.
The AirOps study is frequently misquoted in this debate. Its finding that domain authority shows no positive correlation with citation rate applies exclusively to pages that are already in retrieval. The study's measurement frame begins after the gating process. What happens before retrieval – authority, brand, entity building – lies entirely outside its scope.
What has changed is the sequence of impact. Authority gets the page into the game. Structure and answer quality determine whether it is chosen. Previously, optimisation work ended with the ranking. Today, a second game begins there.
Conclusion: What actually decides digital visibility today?
Not SEO alone. Not GEO alone. But the interplay of both layers – with a clear view of the starting position of each individual URL.
- Authority brings pages into retrieval (SEO / Backlinks)
- Answer structure lets pages win (GEO / Content)
- UX and funnel turn visibility into revenue (Conversion)
Anyone optimising only one of these three layers is paying for visibility that does not convert – or for quality that nobody sees. The AirOps study is a valuable data point for layer two. Layers one and three remain the responsibility of whoever is making the decisions.
AirOps / Kevin Indig: "The Fan-Out Effect: What Happens Between a Query and a Citation", April 2025. Analysis basis: 16,851 queries, 353,799 pages, ChatGPT retrieval pipeline. airops.com/report/the-fan-out-effect
This analysis extends the study's measurement frame to cover the dimensions it does not measure: authority as a retrieval prerequisite, multi-LLM visibility (Perplexity, Claude, Gemini), E-E-A-T, funnel, and conversion.
