Run Your AI Visibility Baseline

Why you start with a number

It is tempting to skip straight to rewriting your website. Resist that. Without a baseline you will have no idea whether anything you do works, and two months from now you'll be guessing. Ten minutes of measuring today gives you a score you can watch move. It also usually delivers a jolt of motivation, because most owners discover the AI is either ignoring them entirely or saying something about them that isn't true.

The four kinds of questions to test

• •Brand questions: “What is [your business] and what do they do?” This tests whether the engines know you exist and describe you correctly.
• •Category questions: “Who offers [what you sell] in [your city]?” This is the big one. It's how a customer who has never heard of you would ask.
• •Comparison questions: “[Your business] vs [a competitor]: which is better for [your type of customer]?” This shows how you're positioned against the people you compete with.
• •Trust questions: “Is [your business] legitimate? Where are they based and what do they charge?” This is what a cautious buyer asks before calling, and it's where wrong facts do the most damage.

Step by step

• •Run the AI Visibility Baseline prompt below in any AI chatbot. Because your business profile is filled in, your details are already in place. It gives you back a personalized set of about ten test questions.
• •Open a fresh chat in each engine you're testing: ChatGPT, Perplexity, Gemini, and a normal Google search to see whether an AI Overview appears. Use a fresh session with no earlier conversation, or the AI will just repeat what you already told it.
• •Ask the same questions in each one, word for word. Copy each answer into a document as you go.
• •Come back to the baseline prompt, paste the answers in, and let it score them and calculate your share of voice.
• •Save the whole thing to a document named “AI Visibility Baseline - [Your Business] - [Date].” You'll re-run it next month against the same questions.

Run each question more than once

AI answers are not fixed. Ask the same question twice and you can get two different sets of businesses. That means a single run tells you very little, and it's the most common mistake people make when they check their AI visibility. Ask each question two or three times, in separate chats, and record what shows up consistently. What you're after is the pattern, not one lucky or unlucky answer.

Your score: AI share of voice

Share of voice is simply your mentions divided by all business mentions across your test set. If ten questions produce forty business mentions and five of them are you, your share of voice is 12.5%. The absolute number matters far less than the direction. Watching it go from 8% to 14% over two months is real progress and proof the work is paying off. Practitioners often aim for roughly 30%, or parity with the other names in your category, but start by beating your own last month.

Pay attention to who wins the answers you lose, too. Note the competitor named instead of you and, where the engine shows its sources, which page fed that mention. That list of winning sources is a free roadmap: it tells you exactly which directories, listicles, and community threads the engines trust in your category.

Keep a wrong-facts log

As you read the answers, you will almost certainly find things that are simply false: an old address, a price you haven't charged in years, a service you don't offer, or a claim you're closed. This is normal and widespread. When Columbia University's Tow Center tested 1,600 queries across eight AI search tools, the tools answered incorrectly more than 60% of the time. Write every error down with the engine that said it and the exact wording. The final lesson of this course is a workflow for hunting down where each wrong claim came from and fixing it at the source.