What does Verdict ask the assistants?
The buyer's question, word for word, the way a person types it into ChatGPT.
You choose the questions, and Verdict suggests some to start with. We don't dress them up with instructions like "list the top ten", because that changes the answer a real buyer would see.
Which assistants, and with search on?
ChatGPT, Claude, Gemini, Grok, Perplexity and Google AI Overviews, each searching the web the way it does for real users.
When an assistant gives no answer, such as Google showing no AI Overview for a question, we record "no answer" and leave it out of every average, rather than counting it as a miss.
From which country?
The client's. Each client is set to one of 16 markets, and its checks run for it.
A dental group in Dublin is measured on the answers Irish buyers get, not American ones.
How often, and why not daily?
Weekly by default, with every-two-weeks and monthly as options. Each prompt is checked up to four times a month.
Answers vary from one asking to the next, so a daily line of single answers mostly measures noise. We confirm a change before we report it, so a trend line moves when the answers move. If an assistant fails on a check, it's marked as failed and the rest of the check still counts.
What counts as being named?
The company's name in the answer text, read the way a person would read it, once per answer.
Different spellings of one company count as one company. Competitors the answers name that you haven't added are listed separately, so you can decide whether to track them.
How are the numbers calculated?
Visibility is the share of answers that name the client. Position is where it's named. Share of voice compares every tracked company, weighted by position.
- Visibility
- Answers naming the client ÷ answers given. Named in 73 of 192 answers is 38%.
- Average position
- Where the client appears when it's named. 1 means first.
- Share of voice
- The client's share of all mentions of tracked companies. Being named first counts most.
Which sources count?
Only pages the answer cites, each counted once however it was linked.
Every site is labeled by type: the client's own site, a competitor's, review sites, forums, publishers, video, social and reference. A page the answers cite for competitors but not for the client is a gap, and those gaps are where the off-site fixes come from.
What does the website audit check?
Whether the crawlers behind AI answers can reach the client's site, read it, and quote it.
That covers the search crawlers that decide what appears in answers, not only the training bots, along with whether pages show their text without JavaScript, respond quickly, and carry structured data an assistant can use. It runs again on its own as the site changes.
What can't Verdict tell you?
Exactly what every buyer sees. Assistants personalize answers, and we measure the answer a new visitor gets.
Logged-in users with long chat histories may see different answers. Verdict measures what a new buyer in the client's country is told, consistently, week after week, which is what you need to see whether your work is moving the answer. We don't estimate traffic from AI answers, and we never change a client's website.