How to Track Whether AI Recommends Your Business

How do I track whether AI recommends my business?

Track AI recommendations by writing down the real buying questions your customers ask, running each one through ChatGPT and Gemini several times, and recording whether you’re named. Tally your mention rate and share of voice against competitors, then re-run on a schedule and watch the trend — not any single result.

That last part is the whole game. AI assistants don’t give you a stable “rank.” Ask the same question twice and the businesses they name can change. So tracking AI recommendations isn’t about capturing one snapshot — it’s about measuring how often you show up across many prompts and runs, and whether that frequency climbs or slips over time. This guide walks the manual method first, so you can do it free, then shows where a tool takes over the tedious part.

Why tracking AI recommendations matters

Because a growing share of your customers now ask an assistant before they ask around. BrightLocal’s research found that 45% of consumers now use AI to find local businesses — up from just 6% a year earlier, one of the fastest behavior shifts in local search anyone has measured. If nearly half your prospects might type “best [your service] near me” into ChatGPT, whether you’re in that answer is no longer a curiosity. It’s pipeline.

The scale behind the shift is hard to ignore. ChatGPT alone crossed 900 million weekly active users in early 2026, and the traffic those assistants send tends to be unusually ready to buy — Similarweb reports that AI referral traffic converts at around 7.1%, second only to paid search. Being recommended by an assistant isn’t a vanity mention; it puts you in front of people who are close to a decision. The only way to know if that’s happening is to measure it deliberately.

“For months I assumed we were fine because we rank well on Google. Then I actually asked ChatGPT ‘who’s the best studio in our city’ ten times and we came up twice. Google ranking and AI recommendations turned out to be two different scoreboards,” — Priya Nair, boutique fitness owner (illustrative).

How to track whether AI recommends your business

You don’t need software to start. You need a repeatable routine and a spreadsheet. Here are the steps, in order.

1. Write down your real buyer prompts

Start with the questions a customer would actually type, not the ones you wish they’d ask. Aim for 8–15 buying-intent prompts: “best [service] in [city],” “who should I hire for [problem],” “affordable [service] near me,” “[competitor] alternatives,” and “is [your business] any good?” Mix broad category questions with a couple that name you directly. These prompts are the backbone of everything that follows — if they don’t match how buyers really talk, your measurement won’t mean much. Keep the list in a spreadsheet so you run the exact same prompts every time; consistency is what makes runs comparable.

2. Pick your engines — and run each prompt more than once

Choose the assistants your customers actually use. For most buyers that’s ChatGPT and Google’s Gemini; you can add Perplexity or Microsoft Copilot if they matter in your market, but two well-chosen engines beat a long list you can’t keep up with. The critical move most people skip: run each prompt three to five times per engine, ideally in a fresh chat each time. Because answers are non-deterministic, a single run is noise. Only by repeating do you see how often you’re named — which is the actual signal.

3. Record who gets mentioned, every run

For each prompt and each run, note whether your business appears, and write down every competitor the assistant names too. A simple grid works: prompts down the side, run numbers across the top, a checkmark where you’re mentioned, plus a list of the other businesses named. This is tedious — that’s the honest truth of the manual method — but it’s also where the insight lives. You’re not just tracking yourself; you’re capturing the full recommendation set so you can measure your slice of it in the next step. Note which answers cited live web sources versus answering from memory, too — grounded and ungrounded mentions are different signals.

4. Calculate your mention rate

Mention rate is the percentage of runs where you appeared. If you ran 12 prompts five times each on one engine — 60 runs — and you were named in 21 of them, your mention rate is 35%. That single number, tracked over time, tells you more than any one answer ever could. Do it per engine, because a web-grounded engine like ChatGPT and a memory-based probe like Gemini can disagree sharply, and the gap itself is useful: strong on grounded, weak on memory usually means your on-page presence is fine but your broader reputation hasn’t sunk in yet.

5. Measure your share of voice against competitors

Mention rate tells you how visible you are. Share of voice tells you how visible you are relative to the field. Count the total mentions across all businesses in your recorded answers, then divide your mentions by that total. If you were named 21 times and all businesses combined were named 140 times, your share of voice is 15%. This is the number that reframes the whole exercise from “am I there?” to “how much of the AI conversation do I own versus my rivals?” Our deep-dive on how to measure AI share of voice walks through the math and the traps in more detail.

6. Re-run on a schedule and watch the trend

One measurement is a baseline, not a verdict. Put a recurring block on your calendar — monthly is a sensible cadence for most small businesses — and run the exact same prompt set again. Add an extra check whenever you do something that should move the needle: publish a batch of service pages, earn a wave of reviews, or land a press mention. Then compare mention rate and share of voice run over run. A rising trend means your efforts are landing; a flat or falling one means it’s time to diagnose. The trend is the product. A single number is just a starting pixel. If you’re still establishing whether AI knows you exist at all, start with our guide on whether AI knows your brand.

Common mistakes when tracking AI recommendations

A few habits quietly wreck an otherwise good tracking routine:

• Trusting a single run. The most common mistake. One answer where you’re absent isn’t a verdict, and one where you’re first isn’t a win. Only the pattern across runs counts.
• Chasing a “rank.” There is no stable AI ranking to climb. If a dashboard hands you a confident position number off one query, it’s dressing up noise. Measure frequency and trend instead.
• Changing your prompts between checks. If the questions drift, your runs aren’t comparable and the trend is meaningless. Lock the prompt set; only add prompts deliberately and note when you did.
• Ignoring the competitors named. Who does get recommended is half the data. It shows you who the assistant considers authoritative — and who to study.
• Confusing grounded and memory answers. An assistant citing live sources and one reciting training data are measuring different things. Track which is which, and don’t treat a memory-based mention as proof of current visibility.

Manual tracking vs. automating it

The manual method is genuinely useful and completely free — you should try it at least once, because doing it by hand teaches you what the numbers mean. What it isn’t is sustainable. Running 12 prompts five times across two engines every month, logging every mention, and doing the share-of-voice math is a real chunk of an afternoon, and the temptation to cut corners (fewer runs, skipped months) is exactly what makes the data unreliable.

That’s the gap tools fill. ScoutRival (full disclosure: it’s our tool) runs your buyer prompts through two engines — ChatGPT’s web-grounded answers and a Google Gemini “model-memory” probe — and reports your mention rate, share of voice against competitors, sentiment, and the citations behind grounded answers. Checks run on demand, so you re-check any day rather than waiting on a schedule, and every number is measured through official access, never by scraping the ChatGPT app — there’s no legitimate public API for the consumer app, so anyone claiming to track exactly what a shopper sees in it is scraping. It’s built for service businesses with no marketing team, and it turns the gaps it finds into one-click content fixes. Plans start free. If you’d rather compare the whole category first, see our roundup of the best AI visibility tools, and for the strategic picture start with our AI visibility guide or browse all our AI & content guides.

Whichever route you pick, the honesty rule is the same: measure mentions and trends across many runs, through official access, and never present a single answer as a fixed rank.