How Do You Check Your Business's AI Search Visibility Manually?
Updated: Sep 28
If someone asked ChatGPT, Perplexity, Claude or Google's AI Overviews about the service your business provides, would your name come up? For most business owners, the honest answer is "I don't know" and that gap matters more each year, because a growing share of buying research now happens inside an AI answer rather than a list of blue links.
The good news is you don't need specialist software to do a basic check of AI search visibility for your business, a manual AI visibility audit costs nothing but time. You play the role of your own customer inside of these AI chat platforms and record what comes back. Done properly, this kind of AI visibility testing tells you whether AI systems know who you are, associate you with the right things, and recommend you when it counts. It gives you a baseline you can test again in a few weeks to see if anything has changed.
Here's how to run that check properly, step by step.
The 6-step manual AI visibility audit
Build a prompt set around the questions real customers ask. Start with questions that never mention your business name.
Run it across every AI platform your customers use, under the same conditions each time.
Record what each answer actually says: mentioned, recommended, cited, and whether the description is accurate.
Run the same prompts for 3 to 5 competitors and work out your share of voice.
Turn the answers into a baseline: mention rate, citation rate and recommendation rate.
Repeat it in 4 to 8 weeks, with the same prompts and conditions, to see what's actually changed.
It takes an afternoon the first time. After that it's quicker, because you're comparing rather than starting from scratch.
Want to run this on your own business? Download the free AI Visibility Audit Template. It has the prompt set, the results log and a scorecard that calculates your three rates for you. Download the template below.
More on AI Visibility here
Step 1: Build an AI Search Visibility Prompt Set Around Real Customer Intent
The single biggest mistake in a manual AI visibility check is typing your own business name into ChatGPT and treating a decent answer as proof you're "visible." This proves almost nothing as a real customer researching a purchase rarely starts with a brand name - they start with a problem, a question or a comparison, and only look up a specific business once they're closer to deciding. Your prompt set needs to reflect that same journey, or the results won't tell you anything useful about how you actually get found.
Test Non-Branded Discovery Before Branded Verification
Start every audit with prompts that never mention your business at all, test category and problem-based questions a stranger would ask before they've heard of you.
Something like "who are the best physios in Auckland for sports injuries" or "what should I look for when choosing a chiropractor" tests whether AI systems surface you on their own, without being prompted.
Only after you've run the non-branded set should you move to branded-verification prompts like "what does [business name] do," "is [business name] reputable," "what are the pros and cons of [business name]", which check whether the AI understands and describes you correctly once your name is in front of it.
Discovery and verification are different problems, and mixing them together hides which one you actually have. This is the single most common mistake we see when someone runs their first audit: they jump straight to a branded search typing in their own business name instead of testing the educational or best-fit prompts a customer would actually use before they know who you are. It feels reassuring, because the AI almost always gets your own name right, but it tells you nothing about whether you'd ever be found in the first place.
Spread Your Prompts Across the Full Customer Journey
A handful of prompts around one keyword isn't enough to see a pattern, you need a spread across the stages a customer actually moves through. A useful set covers:
Problem prompts: "What causes lower back pain after playing sport?"
Educational prompts: "What's the difference between a physio and a chiropractor?"
Service discovery prompts: "Who provides sports injury treatment in South Auckland?"
Recommendation prompts: "What are the top-rated physio clinics near me?"
Comparison prompts: "Compare [Business A] and [Business B] for sports injury treatment."
Decision-stage prompts: "What should I look for when choosing a physio, and who fits that?"
Fifteen to twenty prompts spread across these categories is a realistic starting point, enough to reveal patterns without turning the audit into a full day's work.
Test Entity Relationships Such as Founder, Service, Location and Specialism
AI systems can technically "know" your business exists while still failing to understand how it connects to the things that actually drive referrals. Test those relationships directly, rather than assuming they're covered by your general prompts:
[Founder/Key people] + [Business]: does the AI connect your name to the business?
[Business] + [Service]: does it know what you actually do?
[Business] + [Location]: does it know where you operate?
[Business] + [Specialism]: does it associate you with your niche, not just your category?
[Founder] + [Qualification/framework/expertise]: does it credit you with the credentials that build trust?
A business can appear in general answers while these specific relationships stay weak and it's usually the relationship gap, not a total absence, that's costing the recommendations.
We saw this play out with a chiropractic clinic we worked with. Testing [Founder] + [Business], the AI had confused the founder with a completely different person - wrong background, wrong name attached to the business. After cleaning up the entity footprint across platforms and building clear, consistent links between the founder and the business, that confusion cleared. Four weeks later when asked who the founder was, the AI correctly linked them to their business and surfaced their actual specialism in the answer.
For more on this read What is an entity in AI search and why your business is invisible without one
Step 2: Run the Same Prompts Across Every AI Platform Your Customers Use
Each AI system pulls from different sources and weights signals differently, so a business can be strong in one and effectively invisible in another. If you only test business visibility in ChatGPT, you'll miss where the bigger gaps actually sit.
ChatGPT itself is useful for general brand awareness and unstructured recommendations, especially with web search enabled. Perplexity leans heavily on live web indexing and shows its citations openly, which makes it particularly useful for seeing exactly which URLs it's pulling from. Google AI Overviews and AI Mode matter most for local discovery because they draw heavily on Google Business Profiles and traditional search indexing. Gemini and Copilot round out the set and are worth including if your customers are likely to use them.
Claude is worth adding too, and it's the clearest example of why testing conditions matter. When we ran our own baseline, the platforms that searched the web live described our founder correctly. Claude, answering without live search, mixed her up with two unrelated people who share her name. Same question, different setup, very different answer.
Running the same prompt set across all of them, rather than just the one you personally use, is what turns this from a curiosity into a genuine audit.
Platform | How to get a clean session | Shows its sources? | Most useful for |
ChatGPT | Logged out, or a temporary chat. Check web search is on. | Yes, when it searches the web | General recommendations and "who should I use" comparisons |
Perplexity | Logged out or a private window | Yes, always, in a sources panel | Seeing exactly which pages and directories it pulls from |
Google AI Overviews | Private window, signed out of Google | Sometimes, as link cards | Local discovery. Leans heavily on Google Business Profiles |
Google AI Mode | Private window, signed out of Google | Yes, as source links | Longer, follow-up style research |
Gemini | Private window. If it asks you to sign in, use an account that isn't linked to your business | Sometimes | Customers who live in Google and Android |
Copilot | Private window in Edge or Bing | Yes, as numbered footnotes | Customers on Windows and Microsoft 365. It often gives an explicit "recommendation" call-out |
Claude | New chat. Note whether web search is on | Yes, when web search is on | Comparing what a model "knows" with and without live search |
Signing in isn't cheating, but history is. If you've spent weeks talking to ChatGPT about your own business, it already knows what you want to hear.
Keep Testing Conditions as Consistent as Practical
AI answers are personalised by account history, location and login state, so an inconsistent testing setup will quietly bias your results. Where practical, use a fresh or logged-out session for each platform (or incognito if using your main platform of choice where you have worked on your business in depth), avoid feeding the AI your business name in earlier messages in the same conversation, and note the location setting you tested from (city, suburb, or generic "New Zealand").
If the platform has a web search setting (ChatGPT and Claude both do), note whether it was on. The same prompt with and without live search can produce two completely different answers.
You don't need laboratory-grade consistency, but you do need to know what conditions produced each result, so a change later can be attributed to something real rather than a different login state.
Repeat Important Prompts to Separate One-Off Mentions From Consistent Visibility
AI search outputs are non-deterministic, the same prompt run twice can produce two different answers, with different businesses mentioned or a different order of recommendations.
Run your core prompts two to three times per platform, and more for the ones that matter most, before drawing any conclusion from them. A business that appears in three consecutive runs is different from one that appeared once and vanished. Rank in a single response is close to random, what's meaningful is the percentage of runs where you show up at all.
How to Test Whether Your NZ Business Shows Up in AI Search
Most guides to AI visibility testing are written for the US market, and it shows. For a New Zealand business, a few things change.
Test where your customers actually are. "Best physio in New Zealand" and "physio in Mt Albert" can surface completely different businesses. Put the city or suburb in the prompt the way a customer would, and use the same wording every time you re-test. If you serve several areas, test each one separately rather than assuming Auckland results carry over to Hamilton.
Check your Google Business Profile before anything else. On the Google platforms, your profile often appears before any written answer. When we tested "Auckland NZ AI search visibility experts", Google AI Overviews led with a Business Profile panel (star rating and review count) before the written answer even started. If your profile is thin, out of date or missing services, that's the first fix. It's also usually the fastest.
Look for NZ directories in the sources. New Zealand directories like NoCowboys, Yellow and CompanyHub turn up as cited sources in local answers. If a competitor keeps being cited from one you're not listed on, that's a gap you can close in an afternoon.
Write prompts the way New Zealanders write them. NZ spelling, local terms and local place names. If your customers say "optimisation" and "tradie", your prompts should too.
Expect broad prompts to lean offshore. Ask a general question with no location and you'll often get a list of large US businesses. That's a reflection of what exists online, not necessarily a verdict on you. The prompts that matter most for a local business are the ones with a place name in them.
Step 3: Record What Each AI Answer Says About Your Business
"I asked ChatGPT my business name and it knew who I was" proves very little on its own. A useful manual check goes further: does the AI understand who you are, associate you with the things you want to be known for, and recommend you when a potential customer asks the questions that matter? The "why" behind that answer comes from three specific things you need to capture for every prompt, not just a yes or no but exactly what kind of appearance you got, whether the description was accurate, and which sources and competitors the AI leaned on to build the
Separate Mentions, Listings, Recommendations, Citations and Position
These are different signals, and treating them as interchangeable will flatten the results into something meaningless.
A mention is your business named anywhere in the answer.
A listing is your business appearing as one option among several.
A recommendation is the AI actively suggesting you as a good fit.
A citation is your website or profile linked as a source the answer relies on.
Position matters on top of all four, being named first is a different result to being buried third, or only surfacing after the customer asks a follow-up question.
These combinations tell you different things, and they're not equally bad. A business that's merely listed behind three competitors has a real visibility problem, it isn't the AI's first choice. A business that's recommended first but never cited as a source is actually in a decent spot: that's brand awareness doing its job, and it can send someone straight into a branded search and a direct visit. What's worth watching with this one isn't lost customers, it's lost control, without a citation, you can't see what's actually driving that recommendation, and the AI isn't pulling from your live site, so there's nothing keeping the answer current if your pricing, services or details change.
If your business is mentioned often but rarely recommended, What Is Recommendation Reinforcement in AI Search? explains why that gap opens and what closes it.
Check Accuracy, Relevance, Sentiment and How Your Business Is Described
Once you know your business appeared, check what it actually said. Is the description accurate? Does it have the correct address, suburbs served, phone number, opening hours, services, pricing, qualifications, booking details? AI systems can repeat outdated pricing or old service names long after your website has moved on, so compare every claim against your current site and listings rather than assuming it's caught up.
We've had this happen directly: after restructuring services and pricing, the old pricing kept surfacing in AI answers long after the website itself had been updated. The actual fix wasn't on the website at all, it was our Google Business Profile, where the services list still hadn't caught up. Once that got updated, the AI stopped quoting a specific figure altogether and defaulted to a rough estimate based on what similar services generally cost, a reasonable fallback but a reminder that if you don't control the numbers everywhere they live, the AI will happily average one from whatever it can find.
Conflicting name, address and phone details across your website, Google Business Profile, directories and social profiles cause a similar problem in a different way. In practice, almost every business we've audited has some conflicting NAP information sitting somewhere online, and it's often the quiet reason an AI answer gets a detail wrong even when the source content itself is correct. Is it relevant and associated with the right services, audience and location, not a neighbouring suburb or the wrong specialism?
What's the sentiment and framing - are you positioned as a leader, a budget option, or omitted from a comparison you should reasonably be part of?
Record Competitors, Citations and the Sources Supporting the Answer
This is often the most useful part of the exercise, and it's where it pays to track AI mentions properly rather than skimming for your own name.
For every answer, note which competitors appeared alongside or instead of you, and which sources the AI is actually drawing from, is it your own website, your Google Business Profile, industry directories, review platforms, local news, professional associations, social profiles, or a competitor's page. Perplexity makes this especially visible through its citations panel. If a competitor is repeatedly cited from a directory or review site you're not listed on, that's telling you exactly where an authority gap sits, it's rarely just about your own content.
The mix of sources also shifts by industry, so don't assume your category behaves like the last one you looked at. In financial services, for instance, third-party sites tend to show up as cited sources more heavily than some other categories. LinkedIn and YouTube also get pulled in far more often than you'd expect but only when the business actually has an active presence there, so a dormant profile on either is worth fixing before you spend more time on your website.
Step 4: Compare Your AI Visibility Against Competitors
Your own numbers only mean something in context. Select three to five genuine competitors and run the identical prompt set for each of them, under the same conditions you used for your own business. This is what turns a list of individual answers into a comparison you can actually act on.
Compare Mention Frequency, Recommendation Frequency and Share of Voice
For each competitor, record three things:
How often they appear at all.
How often they're actively recommended rather than just listed.
Where a prompt names several businesses at once, what share of those mentions is theirs versus yours.
That last figure is your share of voice and easiest to see as a worked example: say you ran 20 prompts, and between you and two competitors, businesses were actually named (mentioned) a combined 40 times across those answers.
Business | Times Named | Share of Voice |
You | 6 | 15% |
Competitor A | 18 | 45% |
Competitor B | 16 | 40% |
Total mentions | 40 | 100% |
Share of voice is just the amount of times a brand was named (mentioned) divided by total mentions, multiplied by 100. It gives you a clearer picture of who's actually dominating the category than simply counting how many prompts each of you appeared in. In this example, Competitor A is winning three times over even though none of you were ever completely shut out.
A competitor who shows up in every category prompt but never gets recommended outright has a different kind of visibility to one who's named first every time, the first is strong on discoverability with a conversion gap, the second is winning outright.
Identify Prompts, Services and Customer Needs Where Competitors Appear but You Do Not
Go through your prompt set and flag every case where a competitor appeared and you didn't, these gaps are your action list.
If competitors dominate "best provider" prompts but you hold your own on educational ones, that's not really a content problem, your content is clearly doing its job, since the AI already understands what you do and surfaces you for it. What's missing is trust: third-party reviews, citations from directories or review platforms, the proof points that tip an AI toward recommending one business over an equally well-understood competitor. The fix there is building authority, not writing more.
If it's the reverse, you get recommended once someone's actively comparing providers, but you're invisible on the earlier problem and educational prompts then AI trusts you enough to suggest you, it just never encounters you early enough for that trust to form in the first place. The fix there is a genuine content gap: you need the educational and problem-stage content that earns you a place in the conversation before a customer has already started comparing named businesses.
Compare the Citation and Source Patterns Behind Competitor Visibility
Look at what's actually backing a competitor's visibility, are they cited from a specific directory, a review platform, a piece of local press, or an active Reddit or LinkedIn thread you're not part of? If a competitor is consistently pulled from sources you don't appear on, getting listed on those same source domains is often a faster fix than rewriting your own website content.
Step 5: Turn Individual AI Answers Into a Measurable Baseline
A spreadsheet of raw answers is a start, but it isn't yet a baseline you can track. The next step is turning what you've recorded into a small set of numbers you can compare over time and against competitors on equal terms.
Measure Mention, Citation and Recommendation Rates by Stage and Platform
Three simple calculations do most of the work. They're the same three we use in What Is Recommendation Reinforcement in AI Search?, so the numbers mean the same thing wherever you see them.
Metric | How to calculate it | What it tells you |
Mention rate | Answers where you were named ÷ total answers logged | Whether AI has encountered you at all |
Citation rate | Answers where you were cited ÷ answers where you were named | Whether AI is using your own content as a source |
Recommendation rate | Answers where you were recommended ÷ answers where you were named | Whether AI trusts you enough to put you forward |
A high mention rate with a low recommendation rate is not a visibility win. It's a business AI has heard of, not one it trusts yet.
Break all three down by intent stage (problem, educational, comparison, decision) and by platform, because an average across everything can hide the fact that you're strong in ChatGPT and effectively absent in Perplexity, or visible for educational questions but gone by the time a prompt turns into a buying decision.
Separate Visibility Gaps From Accuracy, Association and Recommendation Gaps
Not every problem is the same problem, and treating them all as "low AI visibility" will send you fixing the wrong thing.
If you don't appear at all for non-branded prompts, that's a discoverability gap.
If you appear but the description is wrong, that's an accuracy gap.
If the description is correct but you're not linked to the service, location or specialism a customer actually searched for, that's an association gap.
If you show up for educational and category questions but disappear the moment a prompt turns toward choosing a provider, that's a recommendation gap.
Each of these has a different fix.
Use the Findings to Identify What Needs Further Investigation
Treat the baseline as a diagnostic, not a final verdict. If competitors dominate "best provider" prompts, dig into what authority signals they have that you don't. If the AI consistently misunderstands what you do, the issue is probably entity clarity or messaging rather than volume of content. If accurate information exists on your site but isn't being cited, look at content structure, topical depth and third-party authority before assuming you need to write more.
This is the gap we run into most often when we run these audits for clients: not missing facts, but content that's accurate and still indistinguishable from a dozen competitor pages covering the same ground the same way. AI systems seem to favour whichever source says something no one else is saying - a specific method, a genuine opinion, a first-hand result over content that's simply correct and well-optimised. If your page reads like it could have been written by anyone else in your industry, that's usually the real problem, not that the AI has somehow missed you.
Step 6: Repeat the Audit to Measure Change Over Time
A single audit is a snapshot, the real value comes from running it again as you take action on the changes/additions that were highlighted that you need to implement. AI answers shift as models update and new content gets indexed, so a one-off check tells you where you stood on the day you ran it and nothing about whether you're moving in the right direction.
Keep the Core Prompt Set and Testing Conditions Consistent
To make repeat audits comparable, keep your core prompt list, platforms and testing conditions the same each time, maintain the same wording, same locations, same logged-out approach where practical. If you change the prompts every time, you won't be able to tell whether a shift in results reflects a real change in visibility or just a different question. A monthly or quarterly re-check, using the same query set, is usually enough to catch meaningful movement without turning the audit into a constant chore.
Compare Changes in Mentions, Recommendations, Citations, Accuracy and Competitor Visibility
Each time you repeat the audit, compare the new mention rate, citation rate, recommendation rate and competitor share of voice against your last baseline. Note whether previously inaccurate descriptions have corrected themselves, whether new competitors have entered the picture, and whether gaps you identified last time have closed. These internal trend measures are far more useful as a comparison against your own past results and against competitors tested under the same conditions than as any kind of universal ranking.
The chiropractic clinic mentioned earlier is a good example of what this looks like in practice. On the repeat check after their entity cleanup, they weren't just described correctly, they started appearing more often as a recommended local provider in their area, not just listed alongside competitors. That kind of shift is exactly what a single audit would never catch, because it only becomes visible once you compare one baseline against the next.
Add New Prompts When Services, Locations or Customer Priorities Change
Keep the core set stable, but don't let it go stale. When you launch a new service, expand into a new suburb or region, or notice customer priorities shifting, add fresh prompts that reflect that change alongside your existing ones. That keeps the audit honest about what your business actually offers today, while still preserving enough of the original prompt set to track genuine change over time.
None of this needs a subscription or a specialist tool, just the discipline to sit down every few weeks and actually look. The businesses that get ahead here aren't the ones with perfect AI visibility from day one, they're the ones who stopped guessing early and started treating it as something to check, measure and act on like any other part of the business. Go back to the question this article opened with: would your name come up? You now have a way to find out, and a way to check again once you've done something about the answer.
Common Questions
How can I check if ChatGPT recommends my business?
Open ChatGPT logged out, or in a temporary chat, with web search on. Ask the questions a customer would ask before they've heard of you, such as "who are the best [your service] providers in [your city]?" Don't type your business name. Run each question two or three times, because answers vary. Then look at whether you were named at all, whether you were actively recommended ("start with", "the best option is") or just listed, and whether your website was cited as a source.
How do I run a manual AI visibility audit?
Build a set of 15 to 20 prompts that match the customer journey, from problem questions through to "who should I choose". Run them across ChatGPT, Perplexity, Google AI Overviews, Google AI Mode, Gemini, Copilot and Claude under the same conditions each time. For every answer, record whether you were mentioned, recommended and cited, and which competitors and sources appeared. Then calculate your mention rate, citation rate and recommendation rate, and repeat the audit every 4 to 8 weeks.
How do I test whether my NZ business shows up in AI search?
Include the city or suburb in your prompts the way a customer would, for example "physio in Mt Albert" rather than "best physio". Test the same location every time. Check your Google Business Profile, because Google's AI answers often lead with it. Look at which New Zealand directories, such as NoCowboys, Yellow and CompanyHub, appear as sources, and make sure you're listed there with consistent details.
Comments