How the index is measured

What the index measures

Two things: whether AI assistants recommend a business, and how well they understand it.

Visibility is the first question a customer's AI answers: "who should I go to?" We measure how often, and how prominently, each business shows up when people ask.

Depth is the second question: once AI names a business, can it actually help a customer choose it? Can it explain what the business does, what it costs, and who it's for? A business can rank first on visibility and still leave customers with wrong or missing answers. We grade both.

Every number is an observed measurement of what ChatGPT, Gemini, Perplexity, and Google AI Mode said on the dates measured. It is not Courtyard's rating of the business, and it is not an endorsement.

The assistants we measure

We put the same questions to four assistants every week: ChatGPT, Gemini, Perplexity, and Google AI Mode. These are the ones customers actually use to find and vet local businesses.

Each question is asked from the perspective of someone in the metro, so the answers reflect what a local customer would see. We read the assistant's written answer: the businesses it names and describes, not the map or local-pack widget beside it.

How the prompts are built

24 real customer questions per market, held fixed so the numbers stay comparable week to week.

For each industry we write 6 customer intents (the different reasons someone reaches for AI) and 4 natural phrasings of each, for 24 questions in all. The prompts never name a business. We ask an open question and watch which businesses the assistant chooses to name.

Each question is put to all four assistants 3 times, because a single answer is noisy. An assistant that names a business two times out of three is telling you something a one-shot ask would miss. That is 24 × 4 × 3 = 288 answers per market, every week.

Intents are weighted, because not every question matters equally. A "who's the best" discovery search counts for more than a niche budget query. The weights:

• Discovery (25%): Weighted at 25% of the visibility score. For example, "best restaurants in Austin, TX".
• Urgent need (25%): Weighted at 25% of the visibility score. For example, "who can fix my AC in Denver today".
• Comparison (20%): Weighted at 20% of the visibility score. For example, "top-rated dentists near Calgary, AB".
• Trust (15%): Weighted at 15% of the visibility score. For example, "most reputable med spa in Dallas, TX".
• Specific service (10%): Weighted at 10% of the visibility score. For example, "gym with a pool in Vancouver, BC".
• Budget (5%): Weighted at 5% of the visibility score. For example, "affordable HVAC repair in Denver, CO".

How share of voice is calculated

Share of voice combines how often a business is named with how prominently. A business named first, on its own, scores higher than one buried in a list of ten. We roll that up across all 24 questions and all four assistants, weighting each question by its intent.

The result is a trailing average, so one unusual week doesn't swing a ranking. As the index runs longer, the number smooths further and the week-over-week movement becomes the signal.

Each market's leaderboard shows the top 25 businesses by share of voice. The rank reflects what the assistants said, nothing else. Businesses can't submit entries or pay to rank, and we don't scrape the local-pack.

How depth is graded

For the top 8 businesses in each market, we run the questions a real buyer asks on the way to a decision.

Visibility tells you AI mentions a business. Depth tells you whether AI can carry a customer from "who's good" to "I'll book them." We ask each assistant a short buyer's journey about the specific business and grade how well it can answer, across five dimensions:

• Offerings (25%): What the business actually does: its services, menu, or specialties.
• Availability (20%): Whether AI can say how to book, when it's open, and how to get in.
• Pricing (25%): What it costs. We compare the price each assistant quotes; a wide spread means AI is guessing.
• Fit (15%): Who the business is right for: the customer, occasion, or need it suits.
• Differentiation (15%): What sets it apart, in concrete terms rather than generic praise.

Reading the depth score

Each dimension is scored, then combined using the weights above into a single depth score from 0 to 100. We surface the weakest dimension for each business, because that is where AI is most likely to send a customer the wrong way.

Pricing gets special treatment. We compare the price each of the four assistants quotes for the same business; when they disagree widely, at least one is fabricating a number, and we flag it. A confident wrong price is worse than no price.

Some dimensions turn on judgment, like whether a business's differentiation is concrete or just generic praise. For those we use a language model to grade the assistant's answer against a fixed rubric. It grades the AI's output, never our opinion of the business.

What this is, and what it isn't

• An observed measurement: Every ranking reflects the output of ChatGPT, Gemini, Perplexity, and Google AI Mode on the dates shown. It is not an endorsement, certification, or rating by Courtyard.
• A snapshot in time: AI output changes. A business can rise or fall as the assistants update. That's why we re-run weekly and stamp every page with its run date.
• Names from the AI's own words: Business names and descriptions come from the assistants' answers themselves. We publish no phone numbers, addresses, or hours.
• Frozen prompts: Prompts don't change without a public methodology-version note, so a change in a ranking reflects a change in AI, not a change in how we asked.