The average CMO inherited a marketing system built on twenty years of SEO. It worked. In 2026, something has changed: a growing share of queries no longer leads to a blue link, but to a synthesized answer. ChatGPT has passed 400M weekly active users (OpenAI, Jan 2026). Perplexity serves more than a billion queries a month. Google has generalized AI Overviews. This article explains what that means, operationally, for anyone deciding where to put tomorrow’s organic budget.
What happened to search
Search stayed roughly stable for two decades: keyword, ten blue links, click. Three shifts broke that equilibrium in the last eighteen months.
First: users talk to machines. ChatGPT, Gemini, Claude, Perplexity. They do not type «best B2B CRM Italy» — they write «Which CRM would you recommend for a 40-person company selling to Italian SMBs?» and receive a synthesized answer with three proper names and a justification. Different query, different intent, different outcome.
Second: Google itself integrated AI into the SERP. AI Overviews — available in Italy since 2025 — absorb up to 38% of clicks for certain informational queries (Similarweb, Q1 2026). Whoever gets cited in the overview gets visibility without a click. Whoever does not get cited disappears.
Third: AI agents are starting to buy. ChatGPT rolled out agentic commerce support via MCP. Amazon launched Rufus. Perplexity Shopping is generally available. For an e-commerce, this means the end customer in 24 months might be a machine querying your catalog in a structured way.
What GEO is, operationally
Generative Engine Optimization is the discipline that makes a brand extractable, citable, preferred by generative engines. It is not an SEO rebrand: it has three concrete differences.
First difference: the target is not ranking, it is citation.
A model like GPT-5 does not show you position 3 of a SERP. It shows you the answer directly, citing — or not citing — sources. The ranking-equivalent metric is Citation Rate: how often your brand is mentioned in generative answers for relevant queries.
Second difference: content must be extractable, not just indexable.
An LLM does not read a two-thousand-word article narratively. It extracts blocks: definitions, lists, factual statements. A GEO-ready article has a TL;DR, FAQ, clear definitions, sourced data. A classic SEO article has a long intro, storytelling, a conclusion: useful for humans, useless for machines.
Third difference: entity matters more than page.
In Google’s Knowledge Graph, in ChatGPT’s training set, in Perplexity, your brand exists as an entity. If your entity is ambiguous — two companies with the same name, disconnected subsidiaries, undisambiguated founders — the LLM describes you badly or not at all. GEO includes systematic Entity SEO work: Schema markup, sameAs cross-referencing, coherence across Wikidata, LinkedIn, Crunchbase, public registries.
What changes concretely for a CMO
Three priority interventions, ranked by impact × effort:
1. A curated llms.txt
The /llms.txt file is the modern equivalent of a machine-readable business card. LLMs read it instead of your sitemap to understand what your company does, what you sell, who you are. Commit one day to it: it should be written, not auto-generated. It helps you decide how you want to be described by machines.
2. GEO-ready content architecture
Existing editorial content — blog articles, product pages, service pages — needs rethinking with extractable blocks. Two concrete criteria: every page must have a FAQ with FAQPage schema; every answer should fit in two or three factual sentences. This is editorial work, not technical, and your best writer can learn it in a week.
3. A measurement baseline
Before any GEO intervention, measure where you are today. Run fifty relevant queries on the three main engines (ChatGPT, Gemini, Perplexity) and count: how often you are cited, how many competitors beat you, which intents you are absent on. Without this baseline, every intervention is blind and every dashboard is fiction.
How you measure GEO
Classic SEO metrics — organic traffic, ranking, CTR — capture a smaller and smaller share of reality. The four GEO metrics that matter in 2026:
- Share of AI Voice — percentage of generative answers that cite you, over a defined query pool.
- Citation Rate — absolute count of weekly citations across generative engines.
- Citation Quality — is the brand described accurately? Is the link to the right page?
- Intent Coverage — of the search intents relevant to your market, how many are covered by content of yours that LLMs can extract?
None of these is native to GA4 or Search Console. You need dedicated dashboards that periodically query generative engines and aggregate the results.
When not to do GEO
GEO is not urgent for everyone. Two cases where the return is low today:
- Brands with dominant branded traffic: if 90% of organic comes from branded queries, priority is defending the entity, not expanding generative visibility.
- Hyper-local markets where LLMs have not penetrated intent yet: some very specialist B2B niches or local retail still live on classic Google. Measure, do not assume.
Essential glossary
- GEO — Generative Engine Optimization. The discipline of optimizing a brand for extraction and citation by generative engines.
- Generative engine — A system that answers human queries producing synthesized text instead of a list of links (ChatGPT, Gemini, Perplexity, Copilot).
- AI Overview — Synthesized answer shown at the top of Google’s SERP, generated by AI from indexed sources.
- Citation Rate — Count of citations per time period over a defined query pool.
- Share of AI Voice — Percentage of generative answers that cite a brand on a predefined query pool.
- llms.txt — Public text file describing a site in a machine-readable way, read by AI agents and crawlers.
- Entity SEO — Discipline of making a brand recognizable as an entity in Knowledge Graphs and LLM training.
- MCP — Model Context Protocol. Open standard for LLMs to talk to external systems (catalogs, inventory, APIs).
- Citability — Property of content that can be extracted and cited by an LLM without loss of meaning.