How to Track AI Search Visibility and Measure Your Progress

Open ChatGPT right now and type your service and your city if you want to track AI search visibility. “Best commercial electrician in Auckland.” “Who should I use for bookkeeping in Hamilton?” See what comes back.

If your name is not in that answer, three competitors just got handed to a buyer who had no reason to keep looking. No search results page. No ads. No second position to fight for.

Most owners have no idea whether this is happening to them, because you cannot measure this from your analytics. There is often no click to count and no referrer to trace. The traffic does not arrive, so it never registers as lost.

Here is how to check whether AI is mentioning your business at all, the four numbers worth measuring, and what to do when the answer comes back blank.

Key Takeaways

• To track AI search visibility, run the same 20 buyer questions through ChatGPT, Claude, Perplexity and Google AI Overviews every month, logged out.
• Four numbers matter: mention rate, citation rate, sentiment and accuracy, and which competitor gets named when you do not.
• Pew Research found people clicked a search result on 8% of visits that showed an AI summary, against 15% of visits without one.
• AI answers get assembled from what other sites say about you, so third-party mentions usually move more than your own homepage copy.
• A blank result is still data. If no model names you for your core service, you have a source problem, not a ranking problem.
• Manual tracking takes about 30 minutes a month in a spreadsheet. Paid trackers only earn their keep once you know which prompts matter.
• Ranking well on Google does not guarantee an AI mention, because the models pull from reviews, directories, forums and press as well as web pages.

Why You Need to Track AI Search Visibility Before You Fix Anything

For twenty years the question was simple. Where do we rank for our main keyword? You could check it in a minute, and the number meant something, because position one meant traffic.

That link between position and traffic is coming apart. The Pew Research Centre studied real browsing behaviour and found users clicked a traditional search result on 8% of visits where Google showed an AI summary, compared with 15% of visits where no summary appeared. Their write-up is worth reading in full. Same query, roughly half the clicks.

Then there is the demand that never touches Google at all. Someone opens ChatGPT or Perplexity, describes their problem in a full sentence, and gets two or three suppliers named back with a line of reasoning for each. That is a shortlist, delivered before you knew a buyer existed.

Here is the part that catches people out. None of it shows up in your reporting. Analytics records a referral only when the model links out, and the person clicks. Most of the time, the answer was enough, so there is no visit, no bounce, no session. Your numbers look flat and stable while a slice of your market quietly gets routed somewhere else.

So the first job is not optimisation. It is measurement. You need a baseline that tells you, in plain terms, whether the machines that now sit between you and your buyers know you exist.

The good news is that this is genuinely cheap to check. Unlike rank tracking, which needs tools and history, you can get a usable read on AI search visibility in half an hour with a browser and a spreadsheet. The bad news is that almost nobody does it, which is also the opportunity.

The 30-Minute Manual Check Anyone Can Run

Start with the prompt list. Twenty questions, written the way a customer talks, not the way you write ad copy. Mix five types:

Category and location questions (“best plumber in Christchurch”). Problem-first questions (“my hot water cylinder keeps running out, who do I call in Wellington”). Comparison questions (“X vs Y for small business accounting”). Brand-direct questions (“is [your business] any good”, “what does [your business] charge”). And buying-criteria questions (“what should I look for when hiring a bookkeeper in NZ”).

Then run every prompt across four surfaces: ChatGPT, Google AI Overviews, Perplexity, and either Claude or Gemini. Use a logged-out browser or a temporary chat. If you run these from your own account, memory and history will feed you a flattering answer that no stranger will ever see.

Record six things per run in a simple sheet: the prompt, the model, the date, whether you were named, which link it cited, and which competitors appeared. Add a one-word sentiment note where you were mentioned. Positive, neutral, wrong.

That last column matters more than people expect. Being mentioned inaccurately is worse than not being mentioned. If a model tells buyers you only service one suburb, or lists a phone number you retired two years ago, that is an active leak.

Do the whole thing on the same day each month with the same twenty prompts. The absolute scores are noisy. The direction of travel is not.

Half an hour. No subscription. And at the end of it, you know something about your business that most of your competitors do not know about theirs.

The Four Numbers Worth Measuring

Twenty prompts across four models give you eighty data points. Reduce them to four numbers, or you will never look at the sheet twice.

Mention rate

Out of eighty runs, how many named your business? That is your headline figure. A new local service business often starts at zero, and zero is a perfectly useful reading. An established operator with reviews and press coverage might sit at 15 to 30%. What matters is the same number next month.

Citation rate

Of the answers that mentioned you, how many linked to a page you control? Mentions build recognition. Citations bring people. If you are named often but never linked, the models trust the story about you more than they trust your own site, which usually means your service pages are too vague to quote.

Accuracy and sentiment

Of the mentions, how many were correct and positive? Track the errors specifically. Wrong service area, old pricing, a director who left, a product you discontinued. Each one is a fixable fact sitting in a source the models are reading.

Competitor share

When you were not mentioned, who was? Keep a running tally. After two months, a pattern shows up, and it is rarely the biggest player in your market. It is usually whoever has the most third-party coverage: directory listings with real descriptions, reviews with written comments, forum threads, a local news feature, a podcast appearance.

Treat these four numbers like any other business metric. Same sheet, same day, reviewed alongside leads and revenue. This is the same discipline behind a 90-day automation checkpoint, and it works for the same reason. Numbers you look at monthly get fixed. Numbers you look at once get forgotten.

Tools That Track AI Search Visibility Automatically

Once the manual check has run for two or three months, you will know which prompts actually matter to your business. That is the moment a tool starts earning its place, not before.

The category has filled out quickly. Profound, Peec AI and Otterly.ai are built specifically for AI answer monitoring. Semrush and Ahrefs have both added AI visibility modules onto their existing platforms, which is convenient if you already pay for one. They all do roughly the same job: run your prompt set across the major models on a schedule, log every mention and citation, and chart the trend.

What you get for the money is frequency and memory. A tool will run 200 prompts weekly instead of 20 monthly, catch shifts you would miss, and hold the history so you can point at a chart six months later and say the work moved something.

Two limits are worth knowing before you spend anything. First, these models are not deterministic. Ask the same question twice, and you can get different suppliers named. Any tool is sampling a moving target, so treat small week-to-week swings as noise. Second, the tools measure. They do not fix. Buying one without a plan to change what the models are reading is the same mistake as buying a rank tracker and never publishing anything.

My honest position: run manual for a quarter, then decide. If AI answers are already sending you enquiries, or you compete in a market where buyers research heavily before they call, a tracker pays for itself. If you have never once had a customer say “ChatGPT recommended you”, spend the money on getting mentioned instead of on watching that you are not.

What to Do When Nothing Mentions You

A blank sheet is not a failure. It is a diagnosis, and it is a specific one. Language models assemble answers about businesses from what other sources say, so if you are invisible in the answers, you are thin in the sources.

Start with the places models already trust. Industry directories and association listings, with a real description rather than a name and a phone number. Review platforms where customers write sentences, not just star ratings. Local business media. Comparison and “best of” roundups in your category. Forums and community threads where your kind of work gets discussed. One decent write-up on a site the models already crawl usually does more than three months of blog posts on your own domain.

Then fix your own pages so they can be quoted. Models lift plain, factual statements. A service page that says “servicing Auckland, Waikato and Bay of Plenty, specialising in commercial switchboard upgrades, 24-hour callout, licensed since 2011” is quotable. A page that says “we deliver quality solutions tailored to your needs” is not. Answer real questions in full sentences, put the answer first, and keep your name, address and phone identical everywhere they appear.

Then correct the errors your tracking surfaced. Chase down the source of each wrong fact and update it at origin. Old directory entries and stale listings are where most of them live.

Last, make it recurring. This is exactly the kind of task that gets done once, feels good, and never happens again, which is the same pattern that stalls most AI adoption in small businesses. Put the monthly run in the calendar, or hand the whole loop to a system that runs the prompts, updates the sheet and flags what changed while you sleep.

The Bigger Pattern Here

Tracking AI search visibility is a small job with an uncomfortable lesson attached. A channel opened up, it started routing buyers, and you had no instrument pointed at it. Not because it was hard to measure, but because nobody was watching for new things worth measuring.

That is the real constraint in most founder-led businesses. It is not the missing tactic. It is that every scan, every check, every “someone should look at this monthly” sits in one person’s head and competes with the day job. So it happens once, or never.

The fix is the same one that fixes missed calls, cold databases and slow follow-up. Give the business a brain that holds the context and a workforce that does the repeat work, so you can track AI search visibility and ensure the monthly check happens whether or not you remembered it. That is what AI automation for a business actually looks like in practice: not a clever tool, a system that keeps running.

Start with the thirty minute check this week. Twenty prompts, four models, one sheet. Whatever the number says, you will know more than your competitors do.

If you want a second set of eyes on what the results mean for your business, book a 30-minute Discovery Call. Bring your sheet. I will tell you straight whether AI visibility is your problem or a distraction from a bigger one.

Frequently Asked Questions

About Octavius