Courtyard · AI Visibility Index · July 2026

How AI finds local businesses

AI doesn’t invent an answer about a business, it reads sources first. We looked at every source ChatGPT, Gemini, Perplexity, and Google AI Mode cited across hundreds of thousands of answers, to see where they learn who to recommend, and where they learn about you.

When the question gets specific, AI reads your own site.

Cite the business’s own site72%
Cite it as the only source42%

Share of answers about a specific business, by source.

4AI assistants
390,000+sources cited
≈99%of answers cite sources
7–12sources per answer
01

Discovery vs depth

AI reads different sources depending on the question

When someone asks AI who to go to, it leans on the reputation layer, the sites that talk about businesses: Google, Reddit, review directories, and local critics. Reddit lands second on that list, ahead of every review site, and almost none of it is yours to write.

Ask about a specific business and the mix flips. The business’s own site, invisible while AI is deciding who to recommend, becomes a top source: from 0% of discovery citations to 39% in depth. The reputation layer recedes as it rises, Google from 14% to 5% and Reddit from 4.6% to 1.7%. The one source that’s genuinely yours is the one AI reaches for to answer about you.

Discovery · who to recommendDepth · about a specific business
The business's own site
0%
39%
Individual sites across the web
66%
38%
Search (Google)
14%
5%
Directories & booking (Yelp, OpenTable, Zocdoc…)
10%
11%
Community (Reddit)
4.6%
1.7%
Editorial (Michelin, TimeOut, The Infatuation)
3.5%
1.5%
Social (Instagram, Facebook)
1.5%
3.6%
Share of citations by source type, discovery vs depth.AI Visibility Index · week of July 10, 2026
02

By industry

Every industry has its own source stack

The reputation layer AI reads is different for a dentist than for a restaurant. Health directories decide who gets recommended in one; critics and reservation platforms in the other. What holds across all of them: when the question gets specific, AI comes back to the business’s own site.

Dentists

Discovery · who to recommend

Google · Reddit · Zocdoc · Opencare · Healthgrades · RateMDs · WebMD

Depth · about this business

Google · Zocdoc · Yelp · the practice's own site · Opencare

Discovery runs on health directories; the specifics come from the practice's own site.

Restaurants

Discovery · who to recommend

Google · OpenTable · Reddit · The Infatuation · TimeOut · Michelin · TripAdvisor

Depth · about this business

OpenTable · Google · the restaurant's own site · Yelp · TripAdvisor

The outlier: discovery is critic- and editorial-driven; depth shifts to OpenTable and the restaurant's own site.

Home services (HVAC)

Discovery · who to recommend

Google · Reddit · HomeStars · Angi · ConsumerAffairs · HousecallPro

Depth · about this business

Google · BBB · the company's own site · Yelp · manufacturer sites (ex: Carrier)

Discovery via review aggregators; depth leans hard on BBB and the company's own site.

03

The catch

Reaching your site isn't the same as reading it

AI falls back on your website for the specifics, but a crawler isn’t a customer. It can land on your page and still come away with nothing, for two reasons that have nothing to do with how good the business is.

Formats AI can’t read

  • Menus, price lists, and hours saved as an image or PDF, not text
  • Prices and availability locked inside a booking widget (Mindbody, Vagaro, OpenTable) loaded from another site
  • Content that only appears after JavaScript runs, which many AI crawlers never execute
  • Key details trapped in a video, an image caption, or a click-to-expand section a crawler won't open

Sites that fall out of date

  • Holiday hours, seasonal menus, and new services that never make it onto the site
  • A price that changed months ago but still shows the old number
  • Sites built once and rarely touched, with no one keeping them up to date
  • AI's cached copy of your site running weeks behind the live version

It shows up in the data. In our depth analysis, understanding of pricing, the question that most often decides who gets called, barely registers. The source AI leans on hardest is the one most businesses never keep machine-readable or current. See the full depth breakdown.

What it means

You can’t control the reputation layer. You can control your own source.

Discovery runs on sites you don’t own, Google, Reddit, the directories and critics for your industry. You can’t rewrite those on demand. But the moment AI answers a real question about your business, it comes back to your own website, and that you can control. In most specific answers your own site is one of the sources; in many, it’s the only one.

If your site is thin, or a machine can’t read it, AI has nothing to work with. Courtyard publishes and maintains a live, machine-readable knowledge base of a business, its services, hours, pricing, and availability, in a form ChatGPT, Gemini, Perplexity, and Claude can read. No tool can promise a specific answer, but the businesses AI describes well are the ones that give it something accurate to read.

Methodology

How this was measured

The corpus

Every source cited by ChatGPT, Gemini, Perplexity, and Google AI Mode across the Index's weekly runs, hundreds of thousands of citations attached to real customer answers.

Discovery vs depth

We split citations by question type: business-agnostic discovery questions (“who should I go to”) versus depth questions about a named business (“what does this place offer”).

Source types

Each cited domain is bucketed (search, community, directory, editorial, social, or the business's own footprint). Own-site rate is a conservative match of the business name against cited domains.

This edition

A point-in-time snapshot from the week of July 10, 2026. The live, continuously updated view is the AI Visibility Index.

These figures are an observed measurement of what third-party AI assistants cite, gathered by Courtyard. They are not a rating, certification, or endorsement of any business. Full methodology on the Index.