How to track your brand in AI search.

AI visibility is not one number. It is a handful of things worth separating, because they fail in different ways:

• Mentions: whether the assistant names you at all for a question you care about.
• Citations: whether it links to your pages as a source. You can be mentioned without being cited, and cited without being recommended.
• Position: whether you are named first or fourth, behind your competitors.
• Accuracy: whether what it says about you is right: your services, your prices, where you work.
• Company you keep: which competitors and which third-party sources appear in the same answers.

For twenty years, the deal with search was simple. Someone searched, they clicked a blue link, and that click landed in your analytics with a referral source attached. You could see it and count it.

AI answers broke that loop. When an assistant answers in a tidy paragraph, there is often no click at all. A Pew Research Center study of real browsing behavior found that when Google showed an AI summary, users clicked a traditional search result in about 8% of visits, against about 15% when no summary appeared, and clicked a source cited inside the summary in 1% of visits.

Some of it does show up. OpenAI says ChatGPT adds utm_source=chatgpt.com to the links it sends, according to its publishers and developers FAQ, and Google counts traffic from AI Overviews and AI Mode inside the Search Console Performance report. But the recommendation that happens without a click leaves no footprint. That is the part you have to go and look for.

Before you buy any software, become your own customer and ask the assistants directly: ChatGPT, Perplexity, Google's AI Mode and AI Overviews, and Gemini. The goal is a record you can repeat, not a screenshot.

1. Write a fixed question set

Ten to twenty questions, in the buyer's words, not your brand's. Mix four kinds: category questions (“best commercial plumber in Tampa”), comparisons (“[you] vs [competitor]”), problem questions (“how do I fix…”) and brand questions (“is [you] any good?”). Keep the wording identical from check to check.

2. Fix the conditions

Answers change with the platform, the mode, whether you are signed in and where you are. Write those down every time, so a change in the answer is not really a change in the conditions.

3. Run each question more than once

AI answers are not deterministic. Phrasing a question differently, or asking it again an hour later, can surface a different set of names. One run tells you almost nothing; several start to show a pattern.

4. Record five things per answer

Whether you were mentioned, where you appeared, which of your pages were cited, which competitors and sources appeared, and whether anything said about you was wrong.

5. Repeat monthly, and compare

Compare each round with the last before changing anything. Patterns across months are evidence. A single answer is an anecdote.

Mention rate and accuracy are usually the first to fix. A business named often but described wrongly has a problem that more content will not solve.

The comforting assumption is that if you rank well on Google, AI answers will simply mirror that. They do not, reliably. Google itself says AI Overviews and AI Mode may run several related searches across subtopics to build a response, and show a wider and more diverse set of links than a classic results page. Other assistants draw on their own indexes and sources.

So you can rank first and still be missing from the answer the assistant gives. AI visibility leans on many of the same signals as SEO, but it is a different scoreboard, and checking one does not tell you the other.

A whole category of AI visibility tools has appeared. Under the marketing, most automate the manual audit above: they send a library of prompts to the assistants, over and over, and count how often your brand appears, where, with which links and in what tone. That can be genuinely useful for trends. We are not naming products we have not tested ourselves; this is how to judge any of them.

• Which platforms and modes does it cover? ChatGPT search, Perplexity, Google AI Overviews and AI Mode, Gemini and others behave differently. Check what is actually measured, not what the homepage lists.
• Whose questions are they? Can you load your own question set, or only use the tool's library? Your buyers' questions matter more than a generic set.
• How does it handle randomness? Ask how many times each prompt runs, and how often. A single run per prompt is a screenshot with a chart.
• Does it keep the full answer and the citations? A score without the underlying answers cannot be checked.
• Does it separate mentions from citations? They mean different things, as above.
• Can you export the data? Your record should not live only inside a subscription.
• Is the method published? If a vendor will not explain how a number is produced, do not base decisions on it.

Keep one thing straight: these tools estimate visibility from their own prompts. They do not see inside the platforms, and they do not see the conversations your customers actually had. Treat their numbers as directional.

If you cannot control the output, you can still shape the conditions. The levers overlap heavily with SEO fundamentals; Search Engine Land's guide to generative engine optimization covers them in more depth.

• Be a clear, consistent entity. The assistant has to understand what you are, what category you are in and where you operate, and find that same story everywhere. If your website, your Google Business Profile and your directory listings describe you three ways, you become harder to recommend with confidence.
• Write content a machine can lift cleanly. A paragraph that fully answers one question on its own is far easier to quote than a meandering wall of prose. Clear headings and direct answers beat clever throat-clearing.
• Earn mentions on places you do not own. What others say about you matters more here than what you say about yourself: reviews, press, industry roundups and communities all feed the answers.
• Show real authority. Named authors, genuine credentials and accurate specifics are the signals that make a source safe to cite.

On the research side, the original GEO paper (Aggarwal and colleagues, KDD 2024) found that adding quotations, statistics and citations improved how visible a source was in generative answers on the authors' own benchmark, by up to 40% on one measure. It is a strong hint about direction, tested in a lab setup, not a promise for any website.

• Measurement is imperfect. Every AI visibility number, manual or tooled, is an estimate. Anyone presenting it with the false precision of a bank statement is overselling.
• It changes constantly. Models are retrained and interfaces redesigned. Build a habit of rechecking, not a one-time report.
• There are no guarantees. You do not own the output, and you cannot pay for placement in an organic AI answer.
• The work compounds anyway. Clear content, real authority and being talked about by others have always built durable visibility. None of it is wasted as the platforms churn.

You will not get a clean dashboard that proves exactly how AI is shaping your pipeline. What you can do is real: audit the assistants by hand every month, track the handful of metrics that matter, consider a tool once the decisions justify the cost, and do the durable work that makes a mention more likely.

At Mining Wells, AI visibility sits inside our AI search optimization work, because it is the same muscle pointed at a newer surface. We will not promise you a seat in ChatGPT's next answer, because nobody honest can. We will tell you what we see when we check, and help you earn the conditions that make a mention more likely.