For twenty-five years, a search produced a list of links and you competed for a position on the list. AI search produces a composed answer. The answer names a few brands, cites a few pages, and most people read it and stop. Your brand is either in the answer or it is invisible to that person.
The scale of the shift is measurable. ChatGPT reached 700 million weekly users sending 18 billion messages a week by July 2025, roughly 10% of the world's adult population.The NBER working paper "How People Use ChatGPT", written with OpenAI's own usage data, documents adoption from launch through July 2025. And inside Google itself, the answer is displacing the list: when Pew tracked real browsing in March 2025, 18% of Google searches produced an AI summary, users clicked a traditional result on only 8% of those visits versus 15% when no summary appeared, and clicked a source cited inside the summary on 1% of visits.Pew Research Center measured actual click behavior from panelists' browsing data, not survey answers.
The clicks are not coming back. Cloudflare's network data shows AI platforms crawling thousands of HTML pages for every visitor they refer; in one June 2025 week, Anthropic's ratio was 70,900 pages crawled per referral.Cloudflare Radar's crawl-to-refer ratios, measured across sites on Cloudflare's network. The ratios move week to week; the imbalance does not. The value your content earns increasingly arrives as presence inside an answer, not as a session in your analytics.
What an AI answer is made of
Every AI answer draws on two sources. The first is model memory: what the model learned about your brand and your category during training, frozen at its training cutoff. The second is live retrieval: the product turns the prompt into web searches, runs them on a search engine like Google, Bing, or Brave, and puts the results into the model's context before it writes. Model Memory and Live Retrieval covers the split in depth, and Retrieval and Grounding: How an AI Answer Gets Assembled walks through the retrieval pipeline stage by stage.
Which source produced the answer decides what work moves it. Memory work is slow and compounds: coverage, reputation, entity presence. Retrieval work is faster and looks a lot like SEO: crawlable pages, extractable passages, rankings on the search engines underneath.
The outcomes that count
Five things are worth measuring in an AI answer, and the rest of this book uses these terms consistently:
- Mention: your brand's name appears anywhere in the answer.
- Recommendation: the answer presents your brand as a pick, not a passing reference.
- Citation: the answer links or credits one of your pages as a source.
- Prominence: where you appear. First-named carries more weight than tenth-named, the same way position one beat position ten.
- Accuracy: whether what the answer says about you is true and current. A model can recommend you while quoting two-year-old pricing.
Together these make up your AI search visibility: the outcome you measure. AI search optimization is the work you do to change that outcome, and it includes classic SEO and paid placement along with the newer work. The two terms are never interchangeable in this book.
Why visibility replaces ranking
A ranking was a position you held in an index. You could check it, report it, and watch it move. AI answers have no index to read a position from. An LLM generates its answer one token at a time with deliberate randomness added at each step, so the same prompt on the same model names different brands on different runs. Any single answer is one sample from a distribution.
The unit of success becomes share of answers: across many runs of the prompts that matter to your business, in what percentage do you appear, and how prominently. Measuring AI Search Visibility builds the full metric stack; the point here is only that "where do we rank" stops being an answerable question, and "how often are we in the answer" replaces it.
The winner-take-all numbers
Absence costs more in AI answers than it ever did in rankings. In OppAlerts' study across 100 industries, the top five brands held 49% of all AI answer share, against 24% for the top five brands in Google organic results in the same industries.From the OppAlerts ranking-factors study: 403,000 prompts, 100 industries, 13 models. Method and tables are in The Research Behind This Guide, part of the AI Search Visibility research. AI answers are roughly twice as concentrated as the rankings they replace.
A results page with ten links gave position eight some traffic. An answer naming three brands gives the fourth brand nothing.
How much this matters depends on where your demand comes from. If buyers reach you through category questions, the kind that produce a shortlist ("best CRM for a small agency", "most reliable midsize SUV"), concentration is the whole game, because one composed answer now stands where ten links stood. If your demand is mostly branded, people asking about you by name, the risk shifts to accuracy: what the answer says about you, not whether you appear. Most businesses carry both exposures in different proportions, and Building Your AI Search Strategy starts from that split.
Where to go from here
The rest of this Part maps the terrain: The AI Search Ecosystem tours the platforms and the search indexes underneath them, The Name Mess: SEO, GEO, AEO, LLMO, and What This Book Calls Things sorts the industry's acronyms, and What Carries Over from SEO, and What Breaks maps existing SEO skill onto this work. If you hold one idea from this chapter, hold the two sources: every answer is memory, retrieval, or both, and everything in this book works through one of them.