How Bay Area customers ask AI

From ten links to three names

For twenty years, getting found online meant ranking on page one of Google. A searcher typed a few words, scanned a list and clicked. Bay Area AEO describes what has changed in its own words: a growing share of customers skip the list entirely. They ask ChatGPT, Gemini, Perplexity or Google's AI Overviews a question and get a short answer with two or three names in it.

That answer is a recommendation. It usually comes with a reason attached (“known for gentle care with nervous patients”, “open on Sundays”, “handles the rebate paperwork”). The customer may never see a list of ten options. If the business is missing from the paragraph, it is missing from the decision.

The Bay Area is where this habit is most advanced. Bay Area AEO's San Francisco page estimates that of roughly 730,000 adults in the city, about 358,000 use AI chatbots such as ChatGPT or Gemini, about 175,000 use one every day, and about 438,000 read Google's AI Overviews. Across its location pages the agency applies the same published rates: 49% of adults use AI chatbots like ChatGPT or Gemini, 24% of adults use an AI chatbot every day, and 60% of adults read Google's AI Overviews. Estimates published by Bay Area AEO according to Pew Research (June 2026); population from the California Department of Finance (January 2026); adult share from the U.S. Census Bureau's American Community Survey.

The anatomy of an AI question

Search queries were short because typing into a search box rewarded short phrases. Questions to an assistant are long because the assistant rewards detail. Take a question from Bay Area AEO's dentist page:

“Can you recommend a dentist in San Mateo who's good with anxious patients and takes Delta Dental?”

That one sentence carries five separate requirements. It names a service (a dentist), a place (San Mateo), a quality the customer cares about (good with anxious patients), a practical condition (takes Delta Dental) and, implicitly, an expectation that the practice is open to new patients. An assistant answering it has to find businesses that satisfy all five, and it can only do that if each fact is written down somewhere it can read.

Real questions from four trades

Bay Area AEO publishes sample questions on each of its industry pages. Reading them side by side shows how much detail customers now put into a single request:

• “My water heater is leaking. Is there a plumber in Hayward who can come today?”
• “What's a good restaurant in Palo Alto for a birthday dinner for six people?”
• “CPA in Cupertino who understands RSUs and ESPP for tech employees?”
• “Coffee shop in Sunnyvale with good Wi-Fi where I can work for a few hours?”

The plumbing question is urgent and mobile: someone standing in a wet garage wants a name and a phone number in one answer. The restaurant question is a planning question with a group size. The CPA question uses vocabulary (RSUs, ESPP) that only a specialist's own pages are likely to contain. None of them could be answered well by a business whose website says only “quality service since day one”.

How Bay Area customers name places

Customers in this region rarely stop at the city name. They name neighborhoods, landmarks and the streets people walk: Temescal and Rockridge in Oakland, Willow Glen and Japantown in San Jose, the Mission and the Sunset in San Francisco, Railroad Square in Santa Rosa. Bay Area AEO's city pages list the neighborhoods customers mention, and its San Francisco page includes questions such as “Best burrito in the Mission District” and “Plumber in the Sunset for an older home”.

A place word in a question works as a filter. When someone asks for a plumber “in the Sunset for an older home”, an assistant will favor a business that has said, somewhere it can read, that it works in the Sunset and works on older homes. Chapter 21 covers how to say that without writing thin copies of the same page for every district.

Spoken questions and follow-ups

Some of these questions are spoken. Siri and the other voice assistants draw on the same listing data as the map apps, so a question asked aloud in a car is answered from Apple Maps and similar sources (see chapter 11). Spoken questions tend to be shorter and more urgent; typed questions to ChatGPT or Claude tend to be longer and more considered.

Many customers also ask a follow-up. “Which of those is open now?” or “Which one is cheapest?” narrows the original answer, and the assistant goes back to its sources to check. A business whose hours or prices are written down clearly survives the follow-up; one whose hours differ between Google and Yelp may be dropped at that point.

Questions that never name a business type

Some of the most useful questions do not name a category at all. Bay Area AEO's food and drink page opens with one: “Where should I take my parents for dinner in Walnut Creek? Somewhere quiet with good wine.” The customer has described an evening. The assistant has to work out that this means a restaurant, that “quiet” rules out the loud places, and that “good wine” points toward a serious list. Then it has to find restaurants in Walnut Creek that someone, somewhere, has described in those terms.

Questions like this reward businesses whose descriptions are specific. A restaurant whose own site mentions its wine program, whose reviews mention conversation-friendly dining rooms, and whose listings carry the right attributes gives the assistant three separate places to find the match. A restaurant described only as “delicious food, great vibes” gives it nothing to work with, however good the food.

The same pattern shows up in other trades. “Someone who doesn't just crack and send you home” is a chiropractic question about treatment style. “Not a big chain” is a gym question about scale. “Natural-looking results” is a med spa question about taste. Each is answerable only from words customers and writers have used about specific businesses.

Questions at different stages of a decision

Customers do not ask one question and buy. They often start broad (“is solar still worth it in San Jose?”), move to comparison (“who's a good local installer for solar plus battery?”) and end with logistics (“which of them can start in March?”). An assistant may name different businesses at each stage, because different sources answer each question.

A business that publishes only a sales page shows up, at best, at the last stage. One that also explains its subject in plain language (how net billing works, what a battery adds, what the permit process looks like) can be cited at the first stage too, and an assistant that cited a business early in a conversation will often keep it in the shortlist later.

What this means for a business

Three practical consequences follow from the way people ask.

Collecting your own customers' questions

The most useful list of questions is the one a business already hears. Front-desk staff, the person who answers the phone and whoever reads the contact-form inbox know what customers ask before they book. Write those questions down for two weeks, word for word, including the conditions people attach. A dental office will hear about insurance and anxiety; a caterer will hear about guest counts and dietary needs; a contractor will hear about permits and timelines.

That list becomes three things later in this handbook: the prompt list for the audit in chapter 7, the plan for service pages and FAQs in chapter 16, and the prompt panel that measures progress each month in chapter 46. Bay Area AEO's own method starts in the same place: a list of the real questions a business's customers ask AI, run across every major platform.

Related chapters3. How an assistant builds an answer · 4. The five signals · 7. The AI visibility audit · 21. Neighborhood relevance