How to Measure AI Search Visibility for a Local Business

More people are asking an AI assistant for a local recommendation instead of scrolling a results page. They type "who does emergency plumbing near me and has good reviews" into a chat tool, or they see an AI-generated summary at the top of a Google search that names two or three companies. If your business is not in those answers, you are invisible to a growing slice of high-intent demand, and most owners have no idea whether they are in them or not.

This guide is about measuring that. Not the basics of optimizing for AI search, which are covered in what generative engine optimization is and how to get cited by ChatGPT, Perplexity, and AI Overviews. This is about building a repeatable way to check where you stand, track whether it is improving, compare against competitors, and report it honestly, without pretending a single AI answer is a reliable score. The method is deliberately low-tech and low-cost, because for most local businesses this is a channel to monitor and steer by, not one to build a large measurement operation around.

Why AI visibility needs its own measurement

Traditional rank tracking does not capture this. A tool that tells you "you rank 3 for plumber near me" says nothing about whether an AI Overview on that same search names you, or whether ChatGPT recommends you when someone asks it directly. Those are separate surfaces with separate mechanics.

How AI answers pick sources

Google has described its generative features as using retrieval-augmented generation: the system leans on its core search ranking to retrieve relevant, current pages, then draws specific information from those pages to build a response and show links that support it. It also uses a technique sometimes called query fan-out, where a single question spawns several related searches across subtopics before the answer is assembled. Chat-based tools like ChatGPT and Perplexity work somewhat differently, running their own retrieval and sometimes their own live search, but the pattern is similar: the answer is built from sources the system judged relevant and trustworthy for that specific question.

What that means for a local business

Your presence in an AI answer depends on being one of the sources the system pulls: your website, your Google Business Profile, review platforms, local directories, or news and community sites that mention you. If those sources consistently and clearly identify you as a strong option for the query, you are more likely to appear. If your web presence is thin or inconsistent, the system has less to work with. This is why AI visibility is downstream of the same fundamentals that drive local search, plus the entity and consistency signals covered in entity SEO for local businesses.

Why measuring it is worth the effort now

Owners sometimes ask whether AI search is big enough yet to bother tracking. Two reasons it is worth starting now, even at low volume. First, the share of local recommendation queries flowing through AI surfaces is climbing steadily, and the businesses that measure early can see the trend and act on it while competitors are still guessing. Second, a baseline only has value if you set it before you do the work. If you wait until AI search is unavoidably large and then start measuring, you have no "before," and you cannot tell whether your position is strong, weak, improving, or slipping. A light monthly measurement started now costs an hour and gives you a year of trend data when you need it.

It is not a separate discipline

The thing that trips people up is treating AI visibility as its own project with its own tricks. It is not. The inputs are the same ones that have always driven local search: a complete and accurate Google Business Profile, a steady flow of genuine reviews, a website that clearly explains what you do and where, consistent business information across the web, and real mentions from local sources. AI systems are just a new consumer of those signals. Measuring AI visibility tells you whether that foundation is strong enough to be picked up by the newest way customers search, and where the gaps are.

Why a single AI answer is not a score

The biggest measurement mistake is treating one AI response as a verdict. AI answers are not stable the way a ranking is.

• They vary by phrasing. "Best plumber in Denton" and "who should I call for a plumbing emergency in Denton" can return different businesses.
• They vary by session and account. Personalization, location signals, chat history, and preferred-source settings all shift results.
• They vary over time. The same prompt asked a week apart can name different companies as the models and their sources update.
• They vary by platform. Google AI Overviews, Google AI Mode, ChatGPT, Perplexity, Gemini, and others each have their own retrieval and their own answer.
• They sometimes decline to answer. For some local queries the AI will say it cannot make a recommendation, or will only describe how to choose, which is not a signal about your business at all.

Because of this, a useful measurement is not "did the AI name us today." It is "across a set of realistic prompts, checked consistently over time, on the platforms our customers use, how often and how prominently do we appear, compared to before and compared to competitors." That is a trend, built from many data points, not a single check.

The right mental model

Think of it the way a pollster thinks about approval ratings. Any single poll has a margin of error and can be off. What matters is the average of many polls and the direction it is moving. Your monthly prompt sweep is one poll. Six of them, in a line, is a trend you can act on. One of them, treated as fact, is how people convince themselves AI search is either a non-issue or a crisis, when it is neither, it is a channel that is growing and that responds to work over months.

What "prominent" means, concretely

When you record how prominently you appear, be consistent about the categories. First named or explicitly recommended is the top tier. Included in a short list of two to four, with a neutral or positive description, is the second. Buried in a long list, or mentioned once in passing, is the third. Cited only as a link supporting a fact, without being recommended as a provider, is its own category, useful for awareness but different from a recommendation. Absent is absent. Write the definitions down so that whoever runs the check next month scores it the same way you did.

Step one: choose realistic customer prompts

The prompts you test with determine whether your measurement means anything. Pick prompts a real customer would actually type.

Build a prompt set from real intent

• Direct recommendation requests: "who is the best [trade] in [city]," "recommend a [trade] near [neighborhood]," "I need an emergency [trade] in [city] tonight."
• Comparison and criteria requests: "which [trade] companies in [city] have the best reviews," "find a licensed and insured [trade] in [city]."
• Problem-first requests: "my [specific problem], who do I call in [city]," which is how a lot of people actually ask.
• Research requests where a citation matters: "how much does [service] cost in [city]," "do I need a permit for [project] in [city]," where being the cited source drives brand awareness even without a direct recommendation.

Keep the set small and stable

Ten to twenty prompts is enough. The point is to check the same set every time so the comparison over time is valid. Resist the urge to keep changing them, since a moving prompt set gives you a moving baseline and no real trend.

Cover your real service area

If you serve several towns, include prompts for the two or three that matter most to your revenue, not just your home city. AI answers are location-sensitive, and you may appear strongly in one market and not at all in another.

An example prompt set for a plumbing company

Here is what a stable set of a dozen prompts might look like for a plumber serving one metro plus two outlying towns:

• "Who is the best plumber in [main city]?"
• "Recommend a plumber near [neighborhood]."
• "I have a burst pipe in [main city] right now, who do I call?"
• "Which plumbing companies in [main city] have the best reviews?"
• "Find a licensed and insured plumber in [main city]."
• "Best plumber for water heater replacement in [main city]."
• "Who does emergency plumbing in [town 2]?"
• "Plumber in [town 3] with good reviews."
• "My water heater is leaking, who should I call in [main city]?"
• "How much does it cost to replace a sewer line in [main city]?"
• "Do I need a permit to replace a water heater in [main city]?"
• "Trusted plumber for an older home in [main city]."

That set covers direct recommendations, comparison queries, problem-first phrasing, service-specific searches, secondary markets, and research questions where being the cited source builds awareness. It is the kind of set you check every month without changing.

Step two: establish a baseline

Before any optimization work, run your prompt set and record exactly what you see. This is the "before" you will measure against.

What to record for each prompt, on each platform

• Did an AI answer appear at all? Some queries do not trigger one.
• Was your business mentioned by name? Yes or no.
• How prominently? First named, in a list, a passing mention, or a linked citation only.
• Which of your sources was cited? Your website, your profile, a review site, a directory.
• Which competitors were named, and how prominently.
• The date, the exact prompt, the platform, and a screenshot.

Turn it into a simple score

A basic scoring scheme keeps the trend readable: 3 points if you are the first or a clearly recommended option, 2 if you are named in a list or comparison, 1 if you are only a linked citation or passing mention, 0 if absent. Sum across the prompt set for a total, and track that total over time. It is rough, but a rough number that moves consistently is more useful than a precise number that does not exist.

Do it on the platforms that matter

For most local businesses, that means Google AI Overviews and AI Mode, since that is where the volume is, plus ChatGPT and Perplexity, since those are the tools a growing group of people use for recommendations. You do not need every platform, but pick two or three and stay consistent.

A sample baseline table

Your recorded baseline for one platform might look like this, simplified:

Extend that across all twelve prompts and you have a baseline total, say 11 out of a possible 36, plus a clear picture of where you are strong (review-driven queries), where you are weak (secondary towns, generic "best" queries), and which of your assets the AI actually uses.

Step three: run the checks

You can do this manually, with tools, or both.

Manual checks

Set a recurring block, monthly is usually enough, to run the prompt set yourself. Use a clean browser session where possible, so personalization is reduced, and note that you cannot fully remove location and account effects. Record the results in your tracking sheet. Manual checking is slow but it shows you exactly what a customer sees, including the wording and the surrounding context, which a tool summary can miss.

Tool-assisted checks

A category of tools now monitors brand mentions in AI answers across platforms, running prompt sets on a schedule and flagging when you appear, when you drop, and which sources are cited. These save time and add consistency, and they are worth considering if AI visibility is a real priority for your business. Treat their numbers as directional, since they face the same variability you do, and check their findings against your own manual spot checks.

When evaluating a tool, look for a few things: can you set your own prompts rather than only using its defaults, can it check the specific platforms your customers use, does it let you set the location or market you care about, and does it show you the actual answer text and cited sources rather than just a yes-or-no. A tool that only reports "mentioned: yes" without the context is not much better than a coin flip, because it cannot tell you whether you were recommended or buried. For a small local business, a lightweight tool plus a monthly manual run is usually the right balance. For a multi-location company, tool automation becomes more necessary because the manual work does not scale across markets.

Do not over-invest in the measurement

The measurement should cost a fraction of the work it informs. An hour a month of checking, or a modest tool subscription, is proportionate. Spending several hours a week or a large monthly fee to track a channel that is still a minority of your leads is out of balance. The point of measuring is to know whether the fundamentals work is paying off in this new surface, so most of your effort belongs on the fundamentals, not on the scoreboard.

Reducing personalization in manual checks

You cannot fully escape personalization, but you can reduce it. Use a fresh browser profile or a private window, sign out of accounts where the platform allows it, and be aware that your device's location still influences local results. If your business serves a market you are not physically in, you will get a more representative read from someone in that area running the same prompts, or from a tool that can set location. Note in your records that your manual checks reflect your own location, so month-to-month comparisons stay apples to apples.

Documenting what you see

Screenshot every result, even the misses. The wording of an AI answer is data: how it describes you, how it describes competitors, what criteria it says matter. Over a few months those screenshots show you not just whether you appear but how the systems are characterizing your market, which tells you what to emphasize on your site and profile.

What to watch between checks

Some signals show up faster than a monthly prompt sweep. A rise in branded searches for your business, people googling your name directly, often follows increased AI visibility, because someone got recommended and then looked you up. A rise in direct traffic to your site can mean the same. Some AI platforms also pass referral traffic that your analytics can attribute to them as a source, so segment and watch that. Watch these as leading indicators between your formal monthly checks.

Judging citation quality, not just presence

Being mentioned is not the same as being mentioned well. Assess the quality of how you show up.

The recommendation ladder

• Named as the top or clearly preferred choice. The strongest position. The AI is effectively recommending you.
• Included in a shortlist with a positive or neutral description. Good. You are in consideration.
• Listed among many options with no distinction. Weak. You appear but do not stand out.
• Cited only as a source for a fact, for example your cost guide is quoted, but you are not recommended as a provider. Still valuable for awareness, but a different kind of win.
• Mentioned negatively or with a caveat. Rare, but worth catching. Usually traces to a review pattern or an inconsistency the AI picked up.

Which source gets cited

If the AI cites your own website, that is ideal, because you control that page. If it cites your Google Business Profile, that is strong too. If it only ever cites a third-party directory's page about you, your own web presence is not doing enough work, and improving your site and profile content is the fix. The structured data guide covers how clear, well-marked-up pages become easier for these systems to use as sources.

Track the source mix over time as its own small metric. A shift from "mostly directory citations" to "mostly your own site and profile" is a real improvement even if your overall score is flat, because it means the systems are increasingly treating your first-party content as the authority on your business. That is a more durable position, since you control what those pages say and can keep them current, whereas a directory page about you might be outdated or thin and you cannot fix it.

The description matters

Note how the AI describes you when it names you. "A highly rated local plumbing company known for fast emergency response" is a description built from your reviews and content. "A plumbing company in Denton" is generic. Over time you want the description to reflect your actual strengths, which happens by making those strengths clear and consistent across your website, profile, and the places that mention you.

Comparing against competitors

Your AI visibility only matters relative to the businesses you compete with.

Track a small competitor set

Pick three or four direct competitors and record their appearances alongside yours on every prompt check. Now your report can say "we appeared in 6 of 12 recommendation prompts, competitor A appeared in 9, competitor B in 4." That comparison is far more actionable than your number alone.

Look at what the winners have in common

When a competitor consistently appears above you, look at their sources: a stronger review profile, a more complete website, more third-party mentions, clearer service and area information, a more established entity presence. The AI is assembling its answer from those signals, so closing the gap means closing it on the underlying fundamentals, not on some AI-specific trick.

Do this analysis in detail once, in your baseline month. For each competitor that consistently outranks you in AI answers, note their Google review count and rating versus yours, roughly how many pages their site has and how deep the service content goes, whether they have location or city pages you lack, and whether they show up in local news, association directories, or supplier locators. The list of gaps is your work plan. Usually two or three items explain most of the difference, and they are the same items that would help your traditional local ranking, which means the effort is not AI-specific overhead, it is core local SEO that also improves your AI position.

Watch for the businesses that are not real competitors

Sometimes an AI answer names a lead-generation front or a national aggregator instead of local businesses. That is worth noting in your report, and it is a case where cleaning up map spam and strengthening genuine local signals, covered in reporting fake listings and building local prominence, helps the AI surface real local providers.

Common measurement mistakes

The ways this goes wrong are consistent.

Checking once and drawing a conclusion

A single run is a snapshot in a system that varies constantly. "I asked ChatGPT and it did not mention us" is not a finding, it is one data point. Only a consistent set over time means anything.

Changing the prompts every check

If the prompt set moves, the trend is meaningless. Lock the set. Add a prompt only if a genuinely new type of customer query emerges, and note the change so the history stays interpretable.

Testing prompts no customer would use

"Tricky Soft Tech reviews" tells you the AI can find you when someone already knows your name. That is not visibility, it is a branded check. Measure the prompts where a customer does not yet know who to call.

Ignoring the platform mix

Being strong in ChatGPT and absent in Google AI Overviews, when most of your customers use Google, is a weak position that a ChatGPT-only check would miss. Weight your attention toward where your customers actually are.

Treating a "no recommendation" answer as a loss

Some prompts return "I cannot recommend a specific business, but here is how to choose one." That is the AI declining, not ranking you last. Log it as no-answer, not as a zero for your business specifically.

Chasing AI-specific hacks

There is a lot of noise about special formatting, prompt-injection-style tricks, and content written to flatter language models. Most of it does not work, some of it violates guidelines, and none of it substitutes for the fundamentals. If your measurement shows you are weak, the fix is almost always better reviews, a clearer site, consistent data, and real local mentions, not a clever trick.

Not connecting it to anything

An AI visibility score that lives in its own spreadsheet, disconnected from branded search, traffic, and leads, becomes a vanity metric. It has to feed the same question as everything else: is qualified demand growing and where is it coming from.

Tracking change over time

The whole point is the trend. Here is how to make it readable.

A monthly cadence

Run the full prompt set once a month, on the same platforms, recording the same fields. Monthly is frequent enough to catch real movement and infrequent enough to be sustainable. Weekly checking mostly captures noise, and quarterly is too slow to connect a change you made to a result. Pick a date, the first business day of the month works well, and treat it like any other recurring report.

Chart the score, not the anecdotes

Plot your total score, and your appearance rate, month over month, with the competitor scores alongside. One good month is not a trend. Three months of movement in the same direction is. Annotate the chart with what you changed, a big batch of reviews, a website rebuild, a citation cleanup, so you can connect cause and effect.

Expect lag and volatility

AI visibility responds to the same slow fundamentals as local SEO, so give changes a quarter or two to show up. And expect month-to-month bounce even when the underlying trend is positive, because the systems themselves change. Judge by the direction over six months, not the wobble between any two checks.

Connecting AI visibility to branded search and leads

AI visibility is worth measuring because it should eventually show up in business results. Connect the two.

Branded search is the clearest link

When an AI recommends a business, a lot of people then search that business by name to check it out. So a rising trend in branded search volume, and in the "direct" versus discovery split in your Google Business Profile's data, is a reasonable proxy for AI and other word-of-mouth visibility working. Track branded search alongside your AI score.

The mechanism is worth spelling out because it is how a lot of AI-driven business actually arrives. A person asks an assistant for a plumber recommendation, gets two or three names, and does not call from the chat. They search the top name on Google, land on the Google Business Profile, read a few reviews, glance at the website, and then call. To your call tracking, that looks like a branded search that turned into a profile call. The AI recommendation that started it is invisible in the attribution. This is why a steady climb in branded searches, especially when your total marketing effort has been steady, is one of the better signs that your visibility in AI answers and other recommendation sources is improving.

Ask new customers

Add "an AI assistant or chatbot" and "the summary at the top of a Google search" as options when you ask new customers how they found you. The numbers will be small at first, but they are direct evidence, and they grow. Train whoever handles intake to note it, and review the tallies quarterly. Even a handful of customers a quarter saying "ChatGPT recommended you" is a real signal that the channel is producing bookings, and it is the kind of evidence that justifies continuing to invest in the fundamentals behind it.

Expect the attribution to be fuzzy

A customer who was recommended by an AI, then searched your name, read reviews, checked your site, and finally called, will usually tell you they found you "on Google" or "through a review site," because that is the last thing they remember. AI visibility often works upstream of the touch a customer names. This is the same multi-touch problem that affects all local marketing measurement, and the response is the same: watch the trend in branded search and total qualified leads, do not expect a clean line from one AI answer to one booked job, and use the "how did you hear about us" data as supporting color rather than a precise count.

Fold it into your overall reporting

AI visibility is one input among many, and it belongs in the same monthly review as calls, booked jobs, and rankings, covered in how to track local SEO ROI. Do not run it as a separate scoreboard disconnected from revenue. The question is always whether total qualified demand is growing and where it is coming from.

What the platforms show now

The official data is improving but incomplete, and you should understand its limits.

Google Search Console

Google has been adding reporting for its generative search features. Clicks and impressions from AI Overviews are generally reflected in the main Search performance data, and Google has introduced additional views for generative AI performance. What this gives you is aggregate trend data, whether clicks and impressions from AI surfaces are rising, not a prompt-by-prompt picture of where you appear. Use it as a macro signal alongside your manual prompt tracking.

Google Business Profile

As AI features surface business information, some of your visibility happens in contexts the profile's own metrics do not fully separate out. The profile performance data remains a strong directional signal for local visibility overall, but do not expect it to isolate "AI-driven" views.

The chat platforms

ChatGPT, Perplexity, and similar tools generally do not give businesses a dashboard of how often they are recommended. Some send referral traffic your analytics can see as a source, which is worth segmenting and watching, but the coverage is partial. For these, your manual prompt checks and third-party monitoring tools are the main measurement.

One practical step: in your website analytics, build a segment or filter for traffic from known AI platforms, and check it monthly. The volume will likely be small, but the trend is informative, and a visitor arriving from an AI tool is often a warm lead who was just given your name as a recommendation. If that segment is growing while your prompt-check score is also improving, the two data points reinforce each other.

Put the official data in context

The platform data tells you the macro story: are AI surfaces sending you more clicks and impressions over time. Your manual prompt tracking tells you the micro story: for the specific questions your customers ask, do you get named, and how well. You need both. The macro data without the prompt tracking cannot tell you where to improve. The prompt tracking without the macro data cannot tell you whether the surface is actually growing for your business. Read them together each month.

Reporting it without false certainty

How you present this matters, because it is easy to overclaim and lose credibility.

Report ranges and trends, not absolutes

Say "across our 12 recommendation prompts this month, we appeared in 7, up from 5 in the baseline, and ahead of two of our three tracked competitors." Do not say "we rank first in AI search," which is not a real thing.

Name the limitations in the report

A short standing note: results vary by phrasing, session, location, and platform; this is a sample checked consistently, not a complete measurement; AI systems change frequently and month-to-month movement includes noise. Owners trust a report more when it is honest about what it cannot do.

Never promise placement

No one can guarantee an AI will cite or recommend a business. The honest framing is that strengthening the fundamentals, reviews, a clear and consistent web presence, real local prominence, entity signals, improves the odds and tends to show up in the trend over a few months. If a vendor promises guaranteed AI citations, that is a reason to walk away.

A one-page report format

Keep the monthly report to a single page with four parts:

• The score: your total across the prompt set this month, versus baseline and versus last month, with the appearance rate.
• The competitor line: your score against your two or three tracked competitors.
• What moved and why: the prompts where you gained or lost, and the likely cause, tied to work you did.
• Leading indicators: branded search trend, AI-referral traffic if visible, and "how did you hear about us" tallies mentioning AI.

Same format every month, so the trend is easy to read and hard to spin.

Tie it back to the decision

The report should help answer one question: is our investment in web presence, reviews, and content making us more visible in the places customers increasingly look, and is that showing up in branded search and leads. If the trend is up over two quarters and branded search is rising with it, keep going. If it is flat despite real work, revisit whether the underlying fundamentals are actually as strong as you think. Measured this way, AI visibility becomes a normal part of running a local marketing program, not a mystery and not a marketing gimmick. It is one more surface where customers are trying to find a business like yours, and the businesses that measure whether they are being found there, and act on the answer, will be ahead of the ones still guessing.