When Andreessen Horowitz (a16z) published "How Generative Engine Optimization Rewrites the Rules of Search", the industry took notice. And for good reason. The piece validates what practitioners of Generative Engine Optimization have been building toward: ChatGPT's 810 million daily users, Google AI Overviews reaching 1.5 billion monthly users, and a fundamental shift from "ranking in results" to "being cited in answers."
They're right. The shift is real, it's accelerating, and brands that ignore it will become invisible. But here's what the a16z framework misses: being cited by AI is not the same as being chosen by AI. And the distance between those two outcomes — what AIVO Standard calls the Decision Gap — is where most brands are quietly losing.
What Does "GEO Replaces SEO" Actually Mean?
The a16z argument rests on a structural observation: AI search engines don't return lists of links. They return synthesized answers. If your brand isn't inside the answer, it doesn't exist for that user. This is undeniably true.
The data backs it up:
- 93% of AI Mode searches end without a single click to any website, according to Semrush data released under the UK CMA order (Digital Arka, July 2026).
- Only 16.7% of AI Overview citations come from pages in Google's traditional top 10 results (BrightEdge, September 2025).
- The overlap between AI-cited pages and Google top-10 rankings dropped from 76% to 38% in just seven months (Ahrefs, 2026).
- AI citation URLs have a 28-day persistence rate of only 10.6% — meaning citations are a flow, not a stock (Digital Authority Partners, 2026).
These numbers confirm the a16z thesis: the old playbook is failing. Traditional SEO signals — backlinks, domain authority, keyword density — are increasingly weak predictors of whether AI will cite you.
What Happens After You Get Cited? The Part a16z Doesn't Cover
Here's where the conversation needs to go further. AIVO Standard, an independent research body focused on AI brand measurement, has published findings that reframe the problem entirely.
In their audit of 68 major consumer brands, AIVO found that 87% had near-zero win rates at the recommendation turn despite strong first-prompt visibility (AIVO Journal, June 2026). The model knew them. It could describe them. But when a user asked "which one should I choose?", the model picked someone else.
This is the Decision Gap: the structural distance between what a model knows about your brand and what it deploys at the moment of commercial recommendation.
Why Does the Decision Gap Exist?
Every AI-generated answer moves through at least two phases:
- Retrieval and assembly — the model gathers what it knows, what it can find, and what appears in its context window.
- The decision turn — the moment a user asks the model to choose, compare, or recommend. The model runs a compression pass across everything it has, summarizing candidates against constraints, and defaulting to whichever brand most cleanly resolves the query.
These are not the same process. At the decision turn, the model is not re-reading your content. It is compressing across all candidates — and the brand whose evidence is structured around outcomes rather than attributes consistently wins.
AIVO's decomposed analysis found a consistent pattern: brands with the highest citation frequency often accumulated the most conflicted evidence in the retrieval layer — strong on awareness signals, weak on the specific proof anchors a model needs to resolve a direct comparison. At the compression pass, the model defaulted to the competitor whose evidence was cleaner.
In plain terms: cited more, chosen less.
The Content Farming Risk: When GEO Makes Things Worse
This is where the current GEO conversation becomes actively dangerous.
The dominant response to GEO measurement has been volume: more content optimized for citation eligibility, more prompts engineered to surface brand mentions, more structured data, more schema, more material fed into the retrieval layer.
AIVO's evidence suggests this approach can backfire. Unanchored content — volume without evidence structure — adds noise to the compression pass. The model cannot tie intent to outcome when the content it has consumed conflates them. The brands that performed worst in AIVO's audits were not the least visible. They were brands whose retrieval layer had been infected by content farming: high citation frequency, degraded evidence structure, near-zero decision-stage deployment.
This is the central risk of treating GEO as "more SEO but for AI." It's not. The optimization target is different, and the failure mode is different.
How Do Different AI Engines Make Decisions?
Compounding the problem, each AI engine has its own "personality" in how it selects and recommends sources:
- ChatGPT and Claude: Extremely reliant on earned media (independent third-party sources like reviews, news, and professional evaluations). Brand websites and social media are cited far less. Over 80% of citations come from earned media in categories like automotive and electronics.
- Perplexity: A hybrid — earned media plus significant social media (especially YouTube) and e-commerce sources. The most diverse citation profile.
- Gemini: The most "brand-friendly" AI engine, with noticeably higher citation rates for brand websites alongside earned media.
Research from the University of Toronto (Chen et al., EDBT/ICDT 2026) found that each engine has 55-67% unique domains that no other engine cites in the same category. One strategy does not fit all.
What Should Brands Actually Do? A Practical Framework
Based on the research evidence, here's a framework that moves beyond "get cited" toward "get chosen":
1. Build Evidence Structure, Not Just Content Volume
Every piece of content should answer: What proof does this give a model to choose us at the decision turn? Specific statistics, named case results, direct expert quotes, and comparison tables are evidence. Generic descriptions, feature lists, and marketing language are attributes. Models default to evidence at the compression pass.
2. Adopt the brand.context Standard
AIVO Standard has published brand.context v2.0 (CC-BY 4.0), a machine-readable JSON-LD schema that structures brand evidence with confidence levels — distinguishing verified claims from self-declared ones. This gives AI systems the exact evidence anchors they need at the decision turn. Early adopters of this standard gain a structural advantage.
3. Optimize for the Decision Turn, Not Just Retrieval
Track whether your brand appears in AI recommendations, not just citations. Ask the models to choose, compare, and recommend — not just describe. Measure win rates at the decision turn, not just mention frequency. This is the metric that correlates with commercial outcomes.
4. Differentiate by Engine
Don't treat all AI platforms the same. For ChatGPT and Claude, invest in earned media and third-party authority. For Perplexity, expand into YouTube and community content. For Gemini, balance earned media with strong on-site content depth. Only 17% domain overlap exists between any two AI platforms — your visibility strategy must be platform-aware.
5. Monitor Citation Accuracy
The SourceCheckup study found that 50-90% of AI citations don't fully support their claims. Your brand may be cited with incorrect information. Regular auditing of what AI says about you — not just whether it mentions you — is essential.
What This Means for the GEO Industry
The a16z piece is an important milestone. It validates the market, legitimizes the discipline, and will bring significant investment and talent into GEO. That's good for everyone.
But the industry needs to evolve quickly beyond citation tracking. The next frontier is decision-stage optimization: understanding why models choose one brand over another at the moment of commercial recommendation, and building the evidence structures that win at that turn.
Brands that treat GEO as "SEO 2.0" — more content, more keywords, more schema — will find themselves in the worst possible position: visible but never chosen. The ones that invest in evidence structure, engine-specific strategy, and decision-turn measurement will build a durable advantage that compounds as AI search grows.
The question is no longer "how do I rank?" or even "how do I get cited?"
The question is: "What does the model deploy about my brand at the moment a buyer asks it to choose — and where does my case collapse?"
Frequently Asked Questions
What is the Decision Gap in AI search?
The Decision Gap is the structural distance between what an AI model knows about your brand and what it deploys at the moment a user asks for a recommendation. AIVO Standard's research found that 87% of brands with strong AI visibility still lose at the recommendation turn — the model can describe them but defaults to competitors when forced to choose.
Is GEO replacing SEO as a16z claims?
The shift from ranking to citation is real and accelerating. However, "GEO replaces SEO" oversimplifies the challenge. Being cited is necessary but insufficient — the real commercial value comes from being recommended at the decision turn. Strong technical SEO remains a foundation, but the content and authority tactics that win in each discipline are increasingly different.
How do I know if AI is recommending my brand or just mentioning it?
Test with decision-stage prompts. Instead of asking "What is [your brand]?", ask "Which [product category] should I choose for [specific use case]?" and "Compare [your brand] vs [competitor] for [scenario]." Track whether you appear in the recommendation, not just the description. Dedicated AI visibility tools like Profound, Peec AI, and Otterly can automate this at scale.
References
- AIVO Standard. "The Cannes AI Playbook Stops at the Wrong Metric." AIVO Journal, June 2026. https://www.aivojournal.org/the-cannes-ai-playbook-stops-at-the-wrong-metric/
- AIVO Standard. "brand.context: A Machine-Readable Evidence Standard for Closing the Linkage Gap in Agentic Commerce (v2.0)." Zenodo, July 2026. https://zenodo.org/records/21262005
- Digital Authority Partners. "AI Visibility Study." 2026. Cited in How AI Engines Cite the Web, Everything PR, June 2026.
- Semrush / UK CMA Order. AI Mode zero-click data. Reported in Google Search Console's New AI Reports, Digital Arka, July 2026.
- BrightEdge. AI Overview citation analysis. September 2025. Cited in Marco Diversi, How to Get Cited by ChatGPT, July 2026.
- Chen, M., Wang, X., Chen, K., & Koudas, N. "Navigating the Shift: A Comparative Analysis of Web Search and Generative AI Response Generation." EDBT/ICDT 2026 Workshops. Tampere, Finland.