How a SayHawk check works

The eleven questions

None of the eleven questions contains your business name. If it did, a mention would be guaranteed and the count would mean nothing. The wording is fixed. Only the trade, the place and the four services change. Here they are, filled in for a dentist in Little Rock, Arkansas, the city of one of our first test checks. Your check uses your own city and trade.

For a trade other than dentistry we swap in the trade, its plural, and four services people ask for by name. The structure stays the same, so every report is built from the same eleven kinds of question.

Two engines, and the app caveat

We ask ChatGPT through OpenAI’s official API and Perplexity through its official Sonar API. Both are asked with web search switched on and the location set to your city, so they answer the way a person in your town would be answered. We label every result “ChatGPT (via OpenAI API)” or “Perplexity (via Sonar API)”. We never scrape the ChatGPT or Perplexity apps.

The consumer apps are not the same product as the API. They can use different settings, remember earlier chats and answer differently, so what you see when you try it yourself may not match your report. We say so on every report. Gemini and Claude are planned, not live.

Why every question is asked twice

Ask an AI the same question twice and you can get two different lists. One answer is one roll of the dice. So each question goes to each engine twice. Eleven questions, two engines and two runs give 44 answers.

More answers help, but they are not 44 separate facts. Two runs of the same question look at nearly the same search results, so they tend to agree. That matters for how sure we can be, and it is why the next sections show a range and not a single score.

How we decide you were named

We read each answer and look for your business name. Capital letters, punctuation and suffixes such as “LLC” or “Inc” are ignored, and we also try the shorter name people actually use. A close spelling counts, but only if it contains a distinctive word from your name. “Family dental” alone is not enough to count as Hawthorne Family Dental.

The businesses named instead are found the same way. We group spellings of the same name across all 44 answers and list the three named most often, with how many answers named each.

How the range is worked out

We start with the share of answers that named you. In the dentist sample that is 9 of 44. If every answer were an independent test, 44 answers would pin that share down quite tightly. They are not independent, because answers to the same question are related. So we do not pretend 44 answers are 44 independent facts. We count how much the answers cluster by question and work out how many independent observations the 44 answers are really worth. It is never fewer than the number of questions and never more than the number of answers.

The result is a 90 percent range from 0 to 100. In plain terms: if the same check were repeated many times, we would expect the true rate to land inside the range about nine times out of ten. In the sample that range is 11–34. A wide range means the answers disagreed or there was little to go on. A narrow range means the answers agreed.

If fewer than 6 of the questions produced answers, or there are fewer than 12 answers in total, we do not show a range. The report says “not enough data”. That is a valid result, and we would rather say it than invent a number. The same answers always produce the same range, because nothing in the calculation is random.

How the wrong information check works

This part is separate from the score. We ask each engine questions that do name your business, for example “What is the phone number, address, and hours for Hawthorne Family Dental in Columbus, OH?” Those answers never count toward the range, because naming you is built into the question.

We compare what comes back with two sources: your Google listing and your own website. We look for three things: a phone number that matches neither, a street number that does not match, and a claim that you are permanently closed when Google shows you as open.

Call tracking numbers are handled. If the number on your website is not the one on your Google listing, both count as correct, so an engine quoting either is not flagged. We only flag a number that appears in a statement about your business, so a competitor’s phone number is never pinned on you. A person reads every finding before it goes into your report. If we find nothing, the report says so.

What we do not do

• We do not promise rankings, results or that an AI will recommend you.
• We do not show one made up score. You get a range, or “not enough data”.
• We do not scrape the ChatGPT or Perplexity apps. We use the official APIs.
• We do not name your business in the eleven scored questions.
• We do not sell or rent your details, and we do not call you.

Known limits

• A check is a snapshot. Answers move from day to day, and a range is not a forecast.
• Two engines are not all of AI. Gemini and Claude are planned, and your customers may use other tools.
• API answers can differ from the consumer apps, so the report will not always match what you see yourself.
• Name matching can miss an unusual spelling, which would make a range slightly low.
• Your Google listing and website are our reference for what is correct. If either is out of date, a finding may point the wrong way, and the report shows both sides so you can judge.
• We measure what AI says. We do not know why it says it, and we do not measure Google search or maps.

Ready to see your own? Ask for your free check.