To get cited by AI Search, the method Google gives is the same as for classic Search: build content you'd actually want to find yourself, then go one level beyond the surface. That's what Nick Fox, who runs Search, Maps, Commerce, and Ads at Google, told Semafor's Ben Smith at Google Marketing Live 2026. One sentence of advice. Then he moved on.
Now count what Google shipped the same week for the paid side: keywordless AI Max, cross-retailer Universal Cart, Direct Offers inside AI Mode, buy-now-pay-later via Affirm and Klarna, expansion into Canada, Australia, and the UK. One side got a slogan. The other got infrastructure. That gap isn't an oversight. It's the map.
The setup: two economies inside one answer
Something is splitting inside the AI answer, and most brands still treat the whole thing as one block they call "AI visibility."
There's the part you can buy: Shopping ads in AI Mode, Direct Offers, agentic checkout powered by the Universal Commerce Protocol that Google launched in January 2026 with Shopify, Target, and Walmart, and that Visa, Mastercard, and Stripe lined up behind. That part has an auction, a product feed, a support page.
And there's the part nobody sells: being named in the answer itself. Being the source the model reaches for when a buyer asks which option to choose. No auction there, no beta program, no interest form.
If you sell a $40 product, the first economy is where the money is. If you sell a mandate, a treatment plan, a school year, or a legal opinion, the first one doesn't exist for you: nobody adds a private bank to a cart. Your entire presence is in the second.
The 3 moves for high-consideration sectors
Once you see the split, the playbook gets concrete. Three moves, in this order.
1. Write the real questions, not keywords
Google says it in its own numbers: queries now stretch across two, three, four sentences. People type the whole problem, the way they'd say it to a colleague. Per Google Trends, queries starting with "which one" grew 40% faster than average over six months, and the biggest risers were the ones asking which option to choose.
Read that as buying behavior, not search behavior. "Which fiduciary handles a Swiss holding structure for a foreign resident" isn't research. It's a decision, outsourced. The first deliverable of a GEO strategy isn't a keyword list: it's the twenty to forty full-sentence questions a real buyer asks before choosing someone like you.
2. Check whether you're named, not whether you rank
Classic search produced ten links and ten winners. The buyer did the comparing, and ranking fourth was still visible. An AI answer names one to three sources, and the machine does the comparing. There is no fourth place.
Being cited isn't a gradual ranking, it's binary: either the assistant says your name or it doesn't. Most brands we audit rank perfectly well on their informational questions — the ones that don't convert — and are completely absent from their decision questions.
3. Go beyond the surface — for real
Fox advised going beyond the surface level. Let's apply that advice to Google's own headline number.
The "+27% conversions" attached to AI Max is real, and it isn't the number most advertisers should plan around. Google's broader benchmark is +14%. The 27% only applies to campaigns still leaning heavily on exact and phrase match — that is, the population with the most headroom by construction.
Then comes the independent read. Smarter Ecommerce analyzed more than 250 retail Search campaigns running AI Max: a median revenue lift of +13%, but a +16% cost-per-acquisition increase, and a ROAS spread running from +42% to −35%. Roughly one campaign in five hit its original target.
The average is honest. The variance is the story. And nobody who copies "+27%" into a slide has gone one level deeper than the slide. That's exactly the behavior models now reward on the content side: surface-level restatement is the one thing generative systems produce infinitely and for free. It has no citation value, because it has no scarcity.
Comparison: old Search / AI Search
| Criterion | Old Search | AI Search (post-GML 2026) |
|---|---|---|
| Unit of query | Keyword, 1 to 3 words | Situation, 2 to 4 sentences |
| Number of winners | 10 visible links | 1 to 3 named sources |
| Who compares? | The user | The machine, on their behalf |
| Nature of the result | Gradual ranking | Binary: cited or absent |
| What wins | Keyword optimization | Point of view and proprietary data |
Takeaways
- Google reinvented its search box to hold a paragraph; the answer shrank to hold three names.
- Two economies are forming inside the AI answer: one you can buy (ads, UCP), one you can't (being cited). Regulated sectors only live in the second.
- The deliverable that matters isn't a keyword list, it's your buyers' list of decision questions — testable today across ChatGPT, Perplexity, Gemini, and AI Mode.