Some customers now ask ChatGPT who to call rather than searching Google. Kerbrank asks it the same kind of questions your customers would, and reports how often it recommends you, who it recommends instead, and which websites it was reading when it answered.
You'll find the results in the When people ask ChatGPT section of your report, and as "Recommended by ChatGPT" on your dashboard.
What we ask
The questions are built from your Google category, your town and the services you track. For a plumber in Portsmouth they might be "best plumber in Portsmouth" or "who should I call for a boiler repair near Southsea". They are written the way a customer would ask, not the way an SEO person would.
Because they come from your tracked searches, changing your searches changes the questions too. Choosing which searches to track is worth reading first if your list isn't settled.
Why each question is asked three times
ChatGPT rarely gives the same answer twice. Ask it the same thing a minute later and you may get a different list of businesses. A single answer is noise, so we ask each question three times and count how many of the answers recommend you by name.
The table in your report shows each question, how many times you were recommended (for example, "2 of 3"), and which other businesses it recommended most.
Sometimes ChatGPT shows you on its map panel without recommending you in its answer. That's counted separately. If the gap between the two is large, the report says so: it can see you, and it often names others instead.
Reading the range
The headline figure comes with a 95% range, for example "22%, likely between 14% and 32%". The first number is the share of answers that recommended you. The range is where the true figure most likely sits, given how many answers we sampled.
Read it this way:
- A wide range means the answers were mixed. Don't read much into a small change from one week to the next.
- If this week's range and last week's range don't overlap, something has probably changed.
This is also why we don't show a bare figure that jumps about every week. The range tells you how much weight the number can take.
The websites ChatGPT was reading
Under the questions is a list headed The websites ChatGPT was reading, with a count next to each. This list often matters more than the headline.
When ChatGPT answers a local question, it tends to lean on directories and review sites. Which ones depends on your trade: often Fresha or Treatwell for salons and clinics, and Yell or Checkatrade for trades. If ChatGPT keeps reading one directory for your kind of business, being missing from that directory, or having a thin listing there, is likely to be the real problem.
What to do
- Open your report and scroll to When people ask ChatGPT.
- Look at Others it recommended most. Check whether the same businesses hold the top three on your map in the competitor comparison.
- Look at The websites ChatGPT was reading. For each directory near the top of the list, check whether you're listed there.
- If you're not, create a listing. If you are, make sure it's complete, with your services, opening hours and the same name, address and phone number as your Google profile. NAP consistency explains why matching details matter.
- Keep your Google reviews coming, and ask customers to mention the service they had.
You won't need to work this out alone. When a directory ChatGPT cites is one you're missing from, it appears on your job list with the steps written out. Press Mark as done once you've sorted it, and the next scan checks it.
What we can't tell you
We can tell you how often ChatGPT recommends you and what it was reading. We can't tell you exactly why it picked one business over another, because it doesn't say. Treat the directories list as the strongest clue and work from there.
The ChatGPT check runs as part of every scan on every plan, with no extra charge, and the same results appear on the report you share with clients (see sharing reports).