How to track your brand in AI search.

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. You could count it. You could argue about it in meetings. That visibility is what made search engine optimization a measurable discipline rather than a vibe.

AI answers broke that loop. When someone asks ChatGPT for a recommendation and it simply tells them the answer in a tidy paragraph, there is often no click at all. Even when the assistant cites sources, most people read the summary and move on. They do not click through. A Pew Research Center study of real browsing behavior found that when Google showed an AI summary, users clicked a traditional search result in only about 8 percent of visits, compared to roughly 15 percent when no summary appeared. They clicked a source cited inside the summary in just 1 percent of visits. One percent.

So the recommendation got made. The buying decision got nudged. And almost none of it left a footprint in the tools you have been staring at for years. That is the pain. Not that AI is naming your competitors, though it might be, but that you genuinely cannot see it happening.

Before you buy any software, do what every honest practitioner does first. Become your own customer and ask the machines directly. Open the major assistants, ChatGPT, Perplexity, Google's AI Mode, and Gemini, and type the exact questions a buyer would type. Not your brand name. The buyer's question.

• Use buying language, not branding language. "Best commercial plumber in Tampa," "top CRM for a small law firm," "who should I hire to redo my kitchen in Austin." Match how a stranger talks, not how your brochure talks.
• Ask the same question several ways. AI answers are not deterministic. Phrasing it differently, or asking again an hour later, can surface a completely different set of names. One run tells you almost nothing. Five runs start to show a pattern.
• Watch for three outcomes. Are you named at all? If so, are you named first, or buried fourth behind your rivals? And is what it says about you actually accurate, or is it quietly confusing you with someone else?
• Log it like a grown-up. A plain spreadsheet does the job: the prompt, the date, the platform, whether you appeared, your position, and who beat you. Patterns only emerge once you have a record to compare against.

Yes, this is manual and a little tedious. It is also the most honest signal you can get, because you are reading the actual output a real customer would see rather than a vendor's estimate of it. Semrush, in its own guide to AI visibility, recommends exactly this manual spot-check as the starting point before any tool enters the picture. Do it once a month. It costs nothing but your coffee break.

The comforting assumption is that if you already rank well on Google, the AI answers will simply mirror that. They do not, reliably. The same Semrush analysis noted that fewer than half of the pages ranking in Google's top ten for a query also showed up in the AI-generated answer for that same query. Roughly 44 percent overlap. The other half ranked beautifully in the old world and were invisible in the new one.

Read that twice if you are paying an agency for rankings. You can be number one on the page and still be absent from the answer the assistant reads aloud. AI visibility is related to traditional SEO, it leans on many of the same authority signals, but it is not the same scoreboard. Checking one does not tell you the other.

A whole product category sprang up to solve the blindness problem, and it is worth understanding what these tools really are under the marketing. They are not a clear window into the AI's brain. They are, essentially, automation of the manual spot-check you just learned, run at enormous scale.

Here is the mechanism. The tool fires a huge library of buyer-style prompts at the major assistants, over and over, then parses the responses to count how often your brand surfaces, where you land relative to competitors, whether you were cited with a link, and whether the tone around you reads positive or neutral. Semrush's AI visibility toolkit describes drawing on a database of well over a hundred million real prompts to estimate this share of voice across ChatGPT, Perplexity, Gemini, and Google's AI features. Ahrefs Brand Radar works in the same family, tracking brand mentions across AI Overviews, ChatGPT, Perplexity, Copilot, Gemini, and Grok, and pointing to the third-party sources (think Reddit, YouTube) that seem to feed those mentions.

Useful? Genuinely, yes, especially if you want a trend line over months instead of a hand-built spreadsheet. But keep your expectations sober:

• It is a sample, not a census. No tool sees the literal conversations your actual customers had. It infers visibility from its own prompt set. The numbers are directional, not gospel.
• The answers move under everyone's feet. Because AI output shifts run to run and model to model, two tools (or the same tool on two days) can report different figures. Treat any single number with a healthy squint.
• Coverage varies by platform. Some assistants are easier to query at scale than others, so a tool may track ChatGPT thoroughly and Gemini more loosely. Read the fine print on what is actually being measured.
• It costs real money. The serious tiers run into the hundreds of dollars a month. For a small local business, disciplined manual checking may be plenty until you have a reason to spend.

If you cannot control the output, you can still stack the deck. The factors that influence whether assistants pull you into an answer turn out to be remarkably down-to-earth, and they overlap heavily with the SEO fundamentals you may already half-know. Search Engine Land's guide to generative engine optimization lays out the core levers, and they are not complicated.

• Be a clear, consistent entity. The assistant has to understand exactly what you are, what category you are in, and where you operate, and it has to find that same story everywhere. If your website, your Google Business Profile, and your directory listings describe you three slightly different ways, you become harder to confidently recommend.
• Write content a machine can lift cleanly. AI assembles answers from self-contained passages. A paragraph that fully answers one question, on its own, without needing the rest of the page for context, is far more liftable than a meandering wall of prose. Plain 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, Reddit threads, industry roundups, and YouTube all feed the models. Being the business other people reference is the strongest position there is.
• Show real authority and trust. Named authors, genuine credentials, accurate specifics, and a credible track record all act as proxies the models use to decide who is safe to cite. This is the familiar experience, expertise, authority, trust idea, wearing a new hat.

There is actual peer-reviewed work here, not just agency blog confidence. The original GEO research paper, published by a team out of Princeton and presented at the KDD 2024 conference, ran controlled experiments on what changes to a page improved its visibility inside generative answers. They found certain tactics lifted visibility by up to 40 percent on their benchmark.

The standouts were unglamorous and, frankly, just good writing. Adding relevant expert quotes was the strongest single move in their tests, with statistics and clear inline citations close behind. The intuition makes sense: a model treats quotation marks, hard numbers, and references to sources as signals of credibility, the same shortcuts a hurried human reader uses.

The honest caveat, which the authors themselves stress, is that the gains varied a lot by industry and topic, and this was a study on specific models at a specific moment. Treat it as a strong hint about direction, not a recipe with guaranteed yields. The web is full of people quoting that 40 percent figure as if it were a button you press. It is not.

A few things are simply true about this space, and pretending otherwise will only cost you money and patience.

• Measurement is genuinely imperfect. This field is a couple of years old. Every number you see, manual or tooled, is an estimate. Anyone presenting AI visibility with the false precision of a bank statement is overselling.
• It changes constantly. Models get retrained, interfaces get redesigned, and what worked last quarter may shift without warning. Build a habit of rechecking, not a one-time report you file and forget.
• There are no guarantees, full stop. You do not own the output, you cannot pay for placement in an organic AI answer, and the day someone guarantees you a mention is the day to keep your wallet closed.
• The work compounds anyway. The reassuring part: the moves that help here, clear content, real authority, being talked about by others, are the same moves that have always built durable visibility. None of it is wasted, even as the platforms churn.

You will not get a clean dashboard that proves exactly how AI is shaping your pipeline. That tool does not exist yet, and anyone who claims theirs does is decorating the truth. What you can do is real and worth doing: spot-check the assistants by hand every month, consider a tracking tool once the spend is justified, and quietly do the durable work that makes a mention more likely, namely being clear about who you are, being genuinely useful on the page, and being worth talking about elsewhere.

At Mining Wells, AI search visibility sits inside the same SEO and generative engine optimization work we already do, because frankly it is the same muscle, just pointed at a newer surface. We are not going to promise you a seat in ChatGPT's next answer, because nobody honest can. We will tell you what we actually see when we check, and help you earn the conditions that tilt the odds your way. In a space this young, that candor is the most valuable thing on offer.