A method for measuring AI search visibility

One answer is a snapshot

Typing your business name into an AI assistant shows you one answer on one day. It tells you little about how often you are named.

Ask the same question tomorrow and the list of businesses can change. So measure a rate across many answers, never a single screenshot.

Track two kinds of prompt

A branded prompt contains your business name. An unbranded prompt asks about what you sell and names nobody.

Keep the two apart in every report. A prompt that contains your name has an obvious reason to repeat it.

The unbranded rate shows whether you are named while a customer is still choosing.

Find the prompts your customers ask

Start with your own enquiries. Write down the questions people ask on the phone, by email and in person before they buy.

Keep their wording, including the place and the problem they describe. Then add the questions that lead to a choice, such as who is best for a given need.

You won't know how often each question is asked. Pick by how close it sits to a sale.

Choose the assistants to track

List the AI assistants your customers mention, then track those.

Each assistant is a separate measurement. A few assistants checked every day tell you more than many checked once.

Read four numbers together

Visibility score is the share of answers that name your business. Share of voice is your mentions as a share of all business mentions in those answers.

Average position is where you sit when an answer lists businesses in order. Sentiment is whether the answer describes you in positive, neutral or negative terms.

A citation is separate from all four. It is a link to your website in the answer, where a mention is your name in the text.

Notice the answers you are missing from

Share of voice and average position only describe answers that already name you. A business can lead both and still be absent from most answers.

The visibility score is the number that counts the gaps. Read it first, then read the other two.

Compare yourself with every business the answers name, including ones you did not expect.

Check each assistant on its own

One blended score can hide the assistant where you are losing. Split the score by assistant and compare each one with its own earlier weeks.

Then look at the sources each assistant cites for your prompts. They can differ, and the difference shows where your business needs to appear.

If an assistant leans on community sites, a useful reply in the right thread may help. If it leans on business websites, a clearer page of your own may help.

Judge progress by the month

Record a baseline in the first month. Then review the same prompts on the same assistants once a month.

Progress looks like visibility score and share of voice rising together while sentiment holds. If visibility rises and share of voice stays flat, competitors rose with you.

Read monthly averages. One bad day is no reason to change a plan.

When a score moves and you changed nothing

Check the prompt list first. A reworded prompt is a new question, so adding, removing or rewording prompts changes the test.

Check the assistant next. If one assistant moved and the others held, the assistant itself may have changed. Wait a week before you react.

Check the place last. The same prompt asked from two towns can name different businesses. Keep your locations fixed to where your customers are.

Whatever is left after those three checks is worth acting on.

How this differs from search rankings

A search ranking report shows where one page sits in a list of links for a keyword. AI visibility asks whether an answer names your business at all.

The two can disagree. A page can rank well in search while AI answers to the same question name other businesses.

Sources and method

A Carqo measurement method. The example figures are illustrative.

Suggest a correction