Key Takeaways
- Monitoring means running the same frozen set of 10 to 15 customer-style queries across ChatGPT, Perplexity, Gemini, and Google AI Overviews every month, then scoring how often your business appears.
- Single checks are snapshots, not answers. AI models return different lists run to run, so score by frequency across two or three runs per query, not by one result.
- The full manual protocol takes about 45 minutes a month and costs $0. Paid trackers are a monthly subscription and only earn their cost at multiple locations.
- Test both 'near me' phrasing and named-city phrasing. They pull from different data layers and a business regularly wins one while losing the other.
- Each AI platform recommends from different sources, with only about 31% overlap in cited businesses in our 200-query test, so monitor each platform separately.
- Act when the trend says act: three flat months at low citation rate justifies foundation work, and 60 to 90 days is the honest window to see changes register.
Why Monitoring Beats Guessing
Most local business owners find out AI assistants are recommending their competitors the same way they find out about a bad review. A customer mentions it.
That is a terrible detection system. By the time a customer tells you "I asked ChatGPT and it sent me to the other guys first," the recommendation pattern has been running for months. BrightLocal's consumer survey found 78% of consumers now discover local businesses through AI-generated recommendations rather than traditional search results. The recommendations are happening whether you watch them or not.
Monitoring is not complicated. It is a frozen query set, a monthly calendar block, and a spreadsheet. What it buys you is the ability to answer three questions with data instead of vibes: Do AI assistants recommend me at all? Is my visibility trending up or down? Did the work I paid for actually move anything?
That last question is the one that matters most if you are paying anyone for AI search work. We cover the diagnostic version of this in how to tell if ChatGPT recommends your business. This post is the ongoing version: not a one-time check, a system.
What to Track: Platforms and Queries
Four platforms cover the real market: ChatGPT, Google AI Overviews, Perplexity, and Gemini. ChatGPT drives the large majority of local AI inquiry volume, but the platforms do not agree with each other. When we ran 200 local queries across 12 verticals, only 31% of cited businesses appeared across all platforms. Each assistant reads different sources. ChatGPT leans on Foursquare and the citation network. AI Overviews leans on Google Business Profile and Google-ranked pages. Perplexity reads the open web broadly. Being visible on one tells you almost nothing about the other three.
Your query set should mimic real customers, not marketers. Nobody types "top-rated licensed plumber [city] 2026" into ChatGPT. They type the situation: "my basement flooded, who do I call in Milwaukee" or "dentist near me that takes new patients." Build 10 to 15 queries in three buckets:
- Direct: "best [service] in [city]", "[service] near me"
- Situational: the emergency or need phrased the way a stressed customer phrases it
- Comparison: "how do I choose a [service] company in [city]"
Then freeze the set. The wording never changes month to month. You are building a trend line, and a trend line requires a constant instrument. If you want to see what a mature recommendation pattern looks like on the other end, our analysis of which plumbers ChatGPT recommends breaks down why the same handful of businesses keep winning these queries.

The 45-Minute Monthly Protocol
Here is the exact run order. Same day each month, same conditions.
- Clean sessions only. Incognito browser, logged out where possible. Logged-in accounts carry history that personalizes answers, and you want to see what a stranger sees.
- Run each query 2 to 3 times per platform. These models are probabilistic. Positions three through six reshuffle constantly. Frequency of appearance is the signal, a single ranking is noise.
- Screenshot everything. Answers change and there is no history to go back to. A dated screenshot folder is your audit trail.
- Log four things per query: did you appear (yes/no), position in the list, what the AI said about you, and which competitors appeared. The competitor column matters more than owners expect. It tells you who is winning the data sources you are losing.
- Note the sources. Perplexity and AI Overviews show citations. When a competitor gets recommended, the cited source is a map of exactly which directory, review site, or article earned them the slot.
First run takes closer to 90 minutes because you are building the spreadsheet. After that, 45 minutes is realistic. If even that is too much, a staff member can run it from the written protocol. The thinking happens in the scoring, not the querying.
How to Score and Read the Results
One number rules the sheet: citation rate. Appearances divided by total query runs, per platform. If you ran 12 queries three times each on ChatGPT and appeared in 14 of those 36 runs, your ChatGPT citation rate is 39% for the month.
Rough benchmarks from the local businesses we track: under 10% means AI assistants effectively do not know you exist. 10 to 30% means you are in the data but weak, usually a citations-and-reviews depth problem. 30 to 50% is solidly competitive. Above 50% means you are the default recommendation in your market, which is where the phone starts ringing from sources you cannot see in Google Analytics.
Read platform gaps as diagnoses, not mysteries. Strong on AI Overviews but invisible on ChatGPT almost always means your Google presence is fine and your Foursquare and citation-network presence is not. Losing 'near me' queries while winning named-city queries points at map-layer completeness. Losing comparison queries points at content: you have no page that answers "how do I choose," so the assistants quote someone who does. The structured-data side of that diagnosis is covered in which schema types actually move AI citations.
And read reviews the way the models do. AI systems mine review text for clues about your services, your location, and the jobs you are known for. A review that says "fixed our AC in Brookfield same day" is a signal. Fifty reviews that say "great service!" are almost nothing.
What Monitoring Costs
Three tiers, honestly priced:
- DIY: $0 and 45 minutes a month. Everything above. Right answer for most single-location businesses.
- Paid trackers: a monthly subscription. Automated querying, trend charts, multi-location rollups. Worth it at 3+ locations or when you need reports for someone else.
- Agency retainers. Monitoring included as the measurement layer. You are not paying for the monitoring, you are paying for the work that moves the numbers it reports.
The trap is paying retainer money for what is functionally just monitoring. If a monthly report shows you citation rates but nobody is fixing citations, publishing content, or building review velocity, you bought an expensive thermometer. The worth-it math for small local businesses only works when measurement and execution come together.

When the Numbers Say Act
Monitoring without a trigger is a hobby. Set the triggers up front:
- Citation rate under 10% for two consecutive months: do the foundation work now. Foursquare, Bing Places, Apple Maps, GBP completeness, LocalBusiness schema. 4 to 6 hours, $0 software.
- Foundation done, still flat after 90 days: the gap is depth, meaning review velocity, city-specific content, and directory presence. This is where paid help starts to pencil.
- A competitor jumps from occasional to constant: pull the sources the assistants cite for them and copy the mechanism, not the words.
- You made changes: tighten monitoring to every two weeks for 60 to 90 days. That is the honest window for changes to register in AI answers. Anyone promising movement in a week is selling something.
The whole loop is simple: measure monthly, diagnose the platform gaps, fix the specific weak layer, and give the fix 60 to 90 days before judging it. Businesses that run this loop compound. Businesses that check once, panic, and buy whatever gets pitched that week do not.
If you want the baseline done for you, we run a full AI visibility audit that queries every major assistant against your market and hands you the gap map. Book a call and we will run your baseline.
Want to know what AI assistants say about your business right now?
We will query ChatGPT, Perplexity, Gemini, and Google AI Overviews against your market and show you exactly where you stand.
Get Your AI Visibility BaselineFrequently Asked Questions
What is the fastest way to check if AI assistants recommend my local business?
Open ChatGPT in a logged-out or incognito session and ask the exact question a customer would ask: 'best [your service] in [your city]' and '[your service] near me, I'm in [neighborhood]'. Note whether your business appears, where it ranks in the list, and what the AI says about you. Repeat the same two queries in Perplexity, Google AI Overviews, and Gemini. That takes about 15 minutes and gives you a real baseline. One session is a snapshot, not a trend, so log the results somewhere and repeat monthly. A single check tells you where you stand today. The pattern over three months tells you whether anything you're doing is working.
How often should I run monitoring queries?
Monthly is the right cadence for most local businesses. AI answers shift when the underlying data sources update, which happens on a weeks-to-months cycle, not daily. Weekly checks mostly measure noise: the same model can return slightly different lists on back-to-back days. Monthly checks, run the same way each time with the same query set, produce a trend line you can trust. The exception is the 60 to 90 days after you make changes such as fixing citations, adding schema, or building review volume. During that window, check every two weeks so you catch movement early and can connect it to the specific work you did.
Which queries should I include in a monitoring set?
Build a set of 10 to 15 queries across three intent levels. Direct: 'best [service] in [city]' and '[service] near me'. Situational: 'my water heater is leaking, who should I call in [city]' or 'emergency [service] [city] open now'. Comparison: 'how do I choose a [service] company in [city]' and '[competitor type] vs [your type]'. Add two or three neighborhood-level variants because AI assistants answer differently at metro scale versus neighborhood scale. Keep the exact wording frozen. If you change the queries every month you lose the ability to compare results, and comparison over time is the entire point of monitoring.
Do I need paid monitoring tools, and what do they cost?
Not at first. Manual monitoring costs $0 and roughly 45 minutes a month, and for a single-location business that is genuinely enough. Paid AI visibility trackers charge a monthly subscription and automate the querying, screenshotting, and trend charts across ChatGPT, Perplexity, Gemini, and AI Overviews. They earn their cost when you manage multiple locations, need client-facing reports, or want daily sampling to smooth out answer variance. Full-service agencies bundle monitoring into the retainer, where the monitoring is the measurement layer and the retainer pays for the work that moves the numbers.
Why do results change every time I ask ChatGPT the same question?
Two reasons. First, large language models are probabilistic. The same prompt can produce a slightly different list each run, especially among businesses ranked third through sixth. Second, personalization and session context matter: a logged-in account with chat history can get different recommendations than a clean session. That is why monitoring protocol matters more than any single result. Use incognito sessions, run each query two or three times, and score by frequency of appearance rather than a single ranking. A business that appears in 8 of 10 runs is genuinely recommended. A business that appears once in 10 runs got lucky, and treating that one appearance as a win will mislead you.
What does it cost to fix low AI visibility once monitoring finds a gap?
The foundation is cheap. Claiming and completing Foursquare, Bing Places, Apple Maps, and Google Business Profile, plus adding LocalBusiness schema, costs $0 in software and 4 to 6 hours of your time. Citation cleanup is a one-time project if your name, address, and phone data is messy across directories. Ongoing optimization, meaning review velocity work, content that answers buyer questions, and directory depth, is a monthly agency cost. Most single-location businesses should DIY the foundation first, re-monitor for 60 days, and only pay for ongoing help if the trend line stays flat.
Does 'near me' phrasing actually change what AI assistants recommend?
Yes, and it is worth testing both phrasings. When a user asks 'near me', assistants that have location access resolve it to a city or neighborhood and lean harder on map-layer data: Foursquare for ChatGPT, Google's local layer for AI Overviews, Apple Maps for Siri-connected queries. When a user names the city outright, the assistant leans more on web content, reviews, and citations that mention that city by name. A business can win one phrasing and lose the other. That is exactly the kind of gap monitoring exposes, and it usually points to a specific fix: map-layer completeness for 'near me' losses, city-specific content and review mentions for named-city losses.
Can I automate AI visibility monitoring myself without buying a tool?
Partially. The API versions of these models are not the same as the consumer products. ChatGPT's API, for example, does not use the same retrieval stack as the chat interface, so scripted API queries can miss what real users actually see. What you can automate cheaply: a monthly calendar block, a frozen query spreadsheet, and a scoring template so the manual run takes 45 minutes instead of two hours. Some owners have a staff member run the protocol and screenshot everything into a shared folder, which costs about 45 minutes of staff time a month. That is automation enough until you cross multiple locations, at which point a paid tracker starts making sense.

About the author
Matthew Johnson is the founder of Pleiades Consultancy. He previously scaled his own marketing agency to multiple six figures before serving as CMO of an Amazon agency, where the client base tripled from 15 to 45 active clients during his tenure. He worked with some of the largest names in e-commerce, including Ridge Wallet, HexClad, BK Beauty, The Woobles, Walkize, Lonely Planet, and Obvi. He now works with service businesses to maximize their client acquisition and visibility through AI search with ChatGPT, Claude, Gemini, Perplexity, and Bing Copilot.

