What did the scorecard measure?
The scorecard asked ChatGPT five questions. Your business name was never in them. They were the kind of question a customer asks before they know you exist. Then, for the main number, each answer was checked for one thing: does your name appear?
Why is your name not in the question? Because a real customer does not type it either. They type something like this:
- "I hurt my knee playing football. Is there a physio in London who works with athletes?"
- "I have sensitive skin. Do you know a cosmetics shop in Istanbul that sells fragrance-free products?"
These are examples, not the questions the scorecard asks. The five questions in your scorecard were written for the category you entered, and each one is listed there. If your name shows up in an answer like this, the assistant brought you up on its own. That is the visibility that matters.
The main number: Discovery Visibility
The headline percentage on the scorecard is Discovery Visibility. It answers one question: of the questions ChatGPT actually answered, in how many did your name appear?
Illustrative example: all five questions were answered and your name appeared in two of the answers. The scorecard shows that as 2/5 and 40%.
Three things do not go into this number:
- Where in the answer you appeared. Whether you came early or late is shown separately, but it does not change the number.
- The tone of the answer. Positive, neutral or negative is also shown separately.
- The technical state of your website. The site checks are a separate section, covered below.
Then there are questions without a usable answer. If ChatGPT produced no answer or a technical error occurred, that question is left out instead of being counted as zero. If no usable answer comes back at all, no scorecard is created. Instead of a zero, you see an error saying the scan could not run right now. That does not mean you are invisible. If fewer than three usable answers came back, the scorecard flags a low sample, so read the result as directional only.
Question by question: where you appeared and where you did not
The scorecard lists each of the five questions on its own line and marks whether your name was found. This is where the useful work starts. Look at the questions where you were absent. What was the customer asking for? Do you actually offer it? If not, being absent there is normal. If you do, do your own site and the sites that write about you say so clearly?
Who was named instead of you?
The scorecard shows the other names that appeared in the same answers, and in how many answers each one appeared. This is often the most useful part, but read it with two caveats.
- These names are pulled from the answers automatically and are not verified competitors. The scorecard says so. You decide which ones are your real competitors.
- The scorecard puts the marketplaces, directories and listing sites it recognises in a separate box. These are often not rivals but places the assistant sends the customer; if one keeps appearing, a complete and accurate profile there may be a way into that answer. If you see a large chain in the list, it may be a direct competitor if it has a branch in your area.
For the physio clinic in London, other clinics on this list would be no surprise. The cosmetics shop in Istanbul might see a large chain or a marketplace. Nobody can tell you in advance which; the scorecard only shows what appeared in those answers on that day.
Which sources did ChatGPT lean on?
When ChatGPT showed sources with an answer, the scorecard lists those websites and how many times each was used. If you entered your website, your own site is marked. The figure for your own site ("Owned-site citations" on the scorecard) looks only at answers that came with sources and shows how many of them cited your site; answers without sources are left out, not counted as zero.
Not every answer comes with sources. If no answer showed sources, the scorecard says so, and that is not your fault. If sources were shown, that list tells you where your business needs to be described accurately: these are the sites the assistant relied on.
Do not skip the scope line
The scorecard states what it measured in one line: sales profile (for example local business or online store), channel (local, online or general), the area and target country if you entered them, and the language the questions were asked in. The form guesses your sales profile and shows it to you; if you did not change it, the guess was used.
If this line is wrong, the result answers the wrong question. If an online shop was measured as a local business, its customers' questions were not the ones asked. If the area was left empty for the clinic in London, the questions were not about London. Language matters too: a customer writing in another language may see different names. If you have not scanned yet, check these fields before you start. A new free scan for the same business can only be run after 30 days.
The website checks are a separate thing
If you also entered your website, the scorecard checks a few technical basics, in a section called "AI Site Readiness": structured data (schema, code that tells machines what a page is about), your page title and description, whether your robots.txt (the file that tells bots what they may visit) blocks AI bots, links to your social profiles, and llms.txt (a newer, proposed summary file for AI tools). It treats llms.txt as optional, because no effect on visibility has been proven.
These checks do not change the main number. Fixing all of them is no guarantee that your name will appear in answers; it only makes your site easier for AI tools to read and understand.
If your number is low
A low or even zero result on the first scan is not unusual. It is a single snapshot, not a verdict. Here is why:
- One engine. The free scan asks only ChatGPT. Your customers may use Gemini, Claude, Perplexity or Google AI Overviews, and being absent in one says nothing about the others.
- Five questions. That is a small sample. With five answers, each one is a fifth of the result, so a single answer moves the number a lot.
- Answers change. Ask the same question twice and the list can change, so a single scan is a snapshot, not a trend. The free scan can be run once every 30 days for the same website (or, without a website, the same brand, category and area), and it does not rerun on its own. If you try again sooner, no new scan runs and you get a link to your existing scorecard.
- A hard test. Your name is not in the question, so the assistant has to bring you up on its own. Showing up when someone types your name is much easier, and proves very little.
Our own sample looked the same. In our field note on local businesses, none of the 25 answers to non-branded questions across 4 sectors named the business being measured. That is a small sample, not a market statistic. But it shows that starting at zero is not unusual.
The scorecard number is not your SOV Score
Both numbers are written as percentages, so they get mixed up. They answer different questions.
The SOV Score is your share of mentions against the competitors you chose: your mentions divided by the total mentions of you and those competitors. It needs the same questions, the same period and a defined competitor list. The free scorecard has no such list, so it shows no SOV Score, and the list of names on the scorecard is not SOV either.
Illustrative example: if you were mentioned 2 times and the competitors you chose 6 times in total, your SOV Score is 2 / 8 = 25%.
For the full calculation, see how the SOV Score is calculated and our methodology page.
What to do next
The scorecard is a starting point. Here is what to do with it:
- Start with the three priority actions on your scorecard. They are drawn from this scan's results.
- Keep the scorecard. It has its own link, which you can share with your team or your agency.
- Make one change at a time. For example, complete your profile on a directory or marketplace that keeps appearing, or write a clear page that describes one service in your customer's words. None of these is a sure thing; make the change, then measure it.
- Do not buy reviews, and do not pay anyone who promises to "get you into ChatGPT". Nobody can guarantee that ChatGPT will recommend you.
A single free scan cannot tell you whether a change worked. The paid plans ask your chosen questions (prompts) every week on all five engines: ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews. They track your SOV Score against the businesses you choose as competitors, and the sources each answer relied on, over time. The details are on the pricing page.