Search just split in two. Most businesses are still measuring the wrong half.

For thirty years, the word search has meant one thing in every marketing department: a list of blue links, ranked, on a Google results page. That assumption is now obsolete, and it costs the average business several inquiries a week without anyone in the building noticing.

I want to walk through what actually changed, because the framing matters more than the tactics.

The surface moved

When a buyer types a category question into Google in 2026, what comes back at the top is no longer ten organic results. It’s an AI Overview — a paragraph or two of synthesized prose, with three or four brand names embedded inline as recommendations, and a row of citation chips below. The list of blue links still exists, but it sits beneath the fold on most queries, beneath an answer that already gave the buyer everything they needed to act.

The same thing has happened, with even more force, inside the chat surfaces. A buyer who opens ChatGPT and asks “who’s the best [your category] in [your city]” gets a single paragraph back. Three names. A short reason for each. A button to continue the conversation. The buyer reads that paragraph, picks a name, and dials. They never opened a search engine.

These two surfaces — AI Overviews on Google, and the standalone chat assistants — now mediate somewhere between thirty and sixty percent of category-defining commercial queries, depending on the industry. The number is moving up every quarter.

The tools didn’t move

Here is the part most marketing teams haven’t caught up to. The tooling that measures search still measures the SERP. Position one on Google. Click-through rate. Impressions. Domain authority. Every dashboard in the average marketing department is still pointed at the ten blue links — at the half of the surface that buyers increasingly skip.

This isn’t a criticism of the tools. They do what they were built to do. The problem is that the question has changed and the tools haven’t been retired.

A practical example: I audited a regional services business last quarter that had spent four years climbing the SERP for its core terms. By every traditional metric, the program was a success. Position one for three of its top five keywords. Forty-percent year-over-year traffic growth.

Then I ran the same queries inside ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews. The brand appeared in exactly zero of the responses. Three competitors — none of whom had stronger SEO by traditional measure — appeared in roughly seventy percent of the responses.

The brand had won the half of the surface that mattered five years ago and missed the half that matters now.

What “appearing in the answer” actually means

When a generative model produces a response to a category question, it draws from two pools.

The first is its training data — a snapshot of the internet that the model was trained on, plus or minus some recency. If your brand’s name appears repeatedly in trustworthy sources in that snapshot — Wikipedia, trade press, schema-rich pages on your own domain, structured entity surfaces like Wikidata and Crunchbase — the model will surface you when the prompt matches.

The second pool is live retrieval. Models that have browsing capability fetch content in real time when a question is asked. Perplexity does this almost exclusively. ChatGPT does this for many query types. Google AI Overviews are essentially built on it. Whether you appear here depends on whether the live page being retrieved cites you, links to you, or names you.

Both pools reward the same underlying signals: pages with clear structure, authoritative outbound citations, schema markup that tells the model what kind of entity you are, and entity surfaces that let the model triangulate your identity across the web. None of these are SEO tactics. Several of them sit outside the discipline of SEO entirely.

The instrument problem

The hardest part of explaining this to a marketing director isn’t the strategic shift. It’s the absence of an instrument.

A traditional SEO program reports against rank, impressions, click-through. Those numbers exist because Google publishes them in Search Console. There is no equivalent dashboard published by OpenAI or Anthropic or Perplexity. You cannot log into ChatGPT and see how often your brand was recommended last quarter.

The only way to know is to ask the models, methodically, on a defined prompt set, every month, and record what comes back. This is what we mean when we say citation rate. It’s not a metric the platforms publish. It’s a metric you generate by running the platform as a buyer would, hundreds of times, and counting.

Once you have that measurement, the strategy reveals itself. The pages that need rewriting are obvious. The entities that need filing are obvious. The competitors that are pulling ahead are obvious. Without the measurement, every quarter of “AI search work” is faith-based.

Where this goes

By 2028, more than half of commercial-intent buyer research is going to happen inside a conversation, not on a results page. The businesses that figure this out in 2026 will own their categories for the next decade. The ones that wait will spend the rest of the decade dislodging incumbents who built citation rate while they were still measuring impressions.

The single best move a marketing team can make right now is to stop assuming search means ten blue links. Measure both surfaces. Optimize for both. Report on both. The cost of the additional discipline is small. The cost of skipping it is the next ten years of buyers asking the question and hearing someone else’s name in the answer.

---

Luke LaFave is the founder of LaFave Consulting. He works with four brands a month on answer-engine optimization across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews. The studio runs from Wisconsin.