AI search visibility tool: how to check if AI is recommending your business

An AI search visibility tool checks whether your business appears in the answers AI platforms generate for your customers’ questions. It works by sending prompts to ChatGPT, Claude, Perplexity, Gemini, and Grok, multiple times each, and cataloguing who gets mentioned. The good ones add a layer of intelligence on top: prompts grouped by audience, validated against what AI engines actually search for, scored with enough reliability to mean something.

88% of businesses are invisible in AI search (Omni Eclipse , 1,700 businesses tested across 32 industries). Read that again. Nearly nine out of ten businesses do not appear when their customers ask AI a relevant question. There is overlap between ranking well on Google and showing up in AI search, but that overlap is conditional. You only get the benefit if you understand how AI search works, and how the tools built to measure it work too.

I spend a lot of my time at Fifty Five and Five building AI tools that marketers can actually use, and one of the things I keep coming back to is this: you cannot improve what you cannot measure. We built our own ai search visibility tool, the GEO conversation opener , specifically so we could baseline a client’s AI visibility before we started any GEO work. This blog teaches you what I know about how these tools work, by walking you through how ours works. The thinking is the same thinking behind every AI search visibility tool on the market. Once you see it, you can pick a tool and use it properly, or you can do a version of this yourself.

The AI visibility gap: why ranking on Google doesn’t mean AI recommends you

Ranking on Google does not mean AI recommends you. 77% of businesses sitting on page 1 of Google are still invisible in ChatGPT (Omni Eclipse ). Your Google position is one signal among many that AI engines use to decide which sources to trust. It is not the whole picture, and treating it as one is why so many businesses are caught out.

Here’s the thing: historically, if I was asked to improve organic visibility, I would focus on keyword volume and keyword strength for Google ranking. High-volume keyword, low competition, build the right page, chase the ranking. That playbook still works. It is not dead.

But it is no longer the whole job. AI search uses Google ranking as one of many factors, and what is far more interesting is query fan-out. When someone asks an AI a question, the AI does not just search for that question. It breaks the question into smaller sub-queries and searches for each one separately. Some of those sub-queries target keywords with low traditional search volume. A volume-first keyword audit would skip straight past them. AI uses them to contextualise what your customer is really asking, which means they carry value even when the volume looks unexciting.

The sweet spot, the best case, is when you find query fan-out terms that also have meaningful search volume. I call these synergy keywords. They rank on Google (so they still earn traditional traffic), and they appear in the sub-queries AI engines fire (so they show up in AI answers too). Every page in this AI search optimisation cluster was built around synergy keywords because they work in both channels.

Why Google AI Overviews matter differently from traditional Google ranking

Google AI Overviews appear in a different place than traditional search results, and getting into them is a different problem from ranking. They now show up in 25.11% of Google searches, up from 13.14% in March 2025 (Conductor ). AI Overviews pull from the same Google index as the standard results, but they apply extra filters on top: expertise signals, answer-first structure, whether the content is easy to extract in a short passage, and whether the author is verifiable. A page can rank at position 3 in the standard results and still be skipped in the AI Overview because it failed one of those filters. Traditional ranking answers “is this page relevant to the query?” AI Overview inclusion answers “is this page relevant AND extractable AND trustworthy enough to be summarised for the user?” Those are different thresholds. It is why you can hit page 1 on Google and still be missing when the same question returns an AI answer.

And the stakes are harder. 93% of AI sessions end without a click through to a website (Semrush ). There is no second-page traffic, no branded search recovery. Either AI cites you, or you’re invisible.

To really understand query fan-out, to the point of learning that you can actually test and scrape the exact terms AI search uses, read the query fan-out page in this cluster. It covers the extraction method with real Gemini and OpenAI API data. For the rest of this post, all you need to hold onto is the concept: AI engines break every question into multiple smaller searches, and ai search visibility depends on whether your content matches what those searches are looking for.

How AI search monitoring tools work (it’s simpler than you think)

AI search monitoring tools work by sending prompts to AI platforms, capturing the responses, and cataloguing which businesses get mentioned. That is the core mechanism. Everything else is layered on top.

When these tools first started landing on the scene, Chris (our founder) and I had a suspicion that they were doing something very simple behind the branding. Search a bunch of times. Log who appears, who doesn’t. Present the whole thing as a visibility score. Now I’ve tried plenty of them, and we’ve built our own, I can confirm: that is what they are doing. It is clever and it is necessary. It is not magic.

Every AI search visibility tool on the market, Semrush, Peec AI, Otterly, Profound, Brandlight, and HubSpot’s AEO Grader, the lot of them, is running some version of this pipeline:

1. Define what you want to check (a business, a product, a topic)
2. Generate or import prompts that represent how customers would ask AI about it
3. Send those prompts to multiple AI platforms, multiple times each
4. Extract the responses and catalogue which businesses appear, how often, and in what context
5. Turn the raw results into a visibility score, share of voice, or citation count

The expensive ones bolt on dashboards, historical tracking, competitive benchmarking, and integrations with your CRM. The branded terminology varies. The core mechanism does not.

How AI models decide which businesses to recommend

AI models decide which businesses to recommend through a combination of training data, real-time retrieval, and source credibility signals. Different platforms use different mechanisms. ChatGPT with browsing on searches the live web; without browsing, it pulls from training data only. Perplexity always searches the web and cites sources with numbered references. Gemini uses Google Search grounding and exposes the sub-queries it fires. Each platform evaluates sources for expertise, cross-referenceable claims, content freshness, and clean structure. A business mentioned consistently across trusted third-party sources, with content structured for easy extraction, gets recommended. A business that only shows up on its own domain, making vague claims, gets skipped. The prompts you test with matter because they determine which retrieval paths the tool exposes. Test a narrow set of prompts, and you see a narrow slice of your real visibility.

Two things make the simple mechanism worth paying for. First: who has time to manually run 50 prompts across 5 platforms, 10 times each? That’s 2,500 queries per audit. Nobody does that on a Tuesday morning. Second: AI answers are wildly inconsistent. Only 30% of brands stay visible in back-to-back AI responses (AirOps ). A single check is a coin flip. Run the same prompt five times and the results move. Run it 50 times across 5 platforms and the actual pattern starts to surface.

Let’s be honest: there is a lot of smart thinking that should go into how you generate prompts and collect the data. Who is the audience? What language do they actually use? Which prompts align with real search volume and query fan-out signals? That thinking is where the good tools earn their money, and it’s where a lot of the others fall short.

But at its core, the job is simple. Search AI the way your customers would, see if you feature in the answer, tally the result. Good hey?!

How our AI search visibility checker works, step by step

Our AI search visibility checker, the GEO conversation opener, works in five steps: brief, personas, prompt intelligence, collect, and analyse. I am not going to walk you through the code. What matters is the thinking behind each step, because the thinking is the part you actually need, whether you use our tool or somebody else’s. We use this internally for every new client to baseline their AI visibility before we start any work.

Step 1: Brief. Before you run a single prompt, define what you are checking. Business name. Website. Core offerings. Main competitors. And a short list of verified facts about the business (headcount, founding year, flagship clients, certifications). Skip this and the tool cannot recognise a mention that does not include the exact business name, and it cannot flag a hallucination when AI makes something up. The brief is the ground truth everything else checks against.

Step 2: Personas. Different audiences prompt AI differently. A CFO evaluating “enterprise AI consulting” does not sound anything like a developer looking for “AI API integration,” even when they are asking about the same company. Test one persona’s prompts and you see one persona’s slice of your visibility. The tool treats each persona as a separate test group so you find out where each audience is seeing you, and where each one is not. You can be the darling of the developer community and completely invisible to the buyer who actually signs the contract. Single-persona testing would never show you that.

Step 3: Prompt intelligence. The tool generates candidate prompts from the brief and personas, then validates them against query fan-out data and keyword research. Only prompts that clear both filters make it through to testing. A prompt that reads nicely but has no search volume and does not appear in query fan-out results is not going to tell you anything useful. A prompt that matches traditional search behaviour AND what AI engines fire as sub-queries is a far better mirror of what is actually happening. (I know what you’re thinking: this sounds like extra work. It is. It is also where most competing tools stop short, because it is slower.)

Step 4: Collect. The validated prompts go to five AI platforms (ChatGPT, Claude, Perplexity, Gemini, and Grok), multiple times each. The multiple runs are non-negotiable. We call it multi-run consistency scoring: instead of asking once and reporting a single answer, the tool runs every prompt several times on every platform and reports how consistent the result was. A business that appears in 9 out of 10 ChatGPT runs is in a very different position from one that appears in 3 out of 10. A single-run check would not know the difference.

Step 5: Analyse. The raw collection data gets passed through an analysis layer that produces visibility scores per persona, share of voice against named competitors, accuracy checks against the verified facts from the brief, competitive displacement (who’s appearing instead of you, and what they’re doing differently), and a prioritised set of recommendations. The goal is not a single vanity score. The goal is something you can use to shape strategy and prove ROI.

The labs page for the GEO conversation opener has more detail if you want to poke around the prototype. It is the same tool we use to run a GEO audit before we start client work.

What an AI citation checker tells you about your business

An AI citation checker tells you where your business appears across AI platforms, how often, how consistently, and how that stacks up against your competitors. A decent one does not stop at a single number. It gives you a structured output you can actually do something with.

The headline metric is visibility per persona. That is the percentage of prompts where your business appeared in the AI’s answer, broken out by audience. You might be visible in 80% of your decision-maker prompts but only 15% of your end-user prompts. That gap is strategic intel. It tells you which audience your content is speaking to, and which one is squinting at their screen wondering where you went.

How AI visibility tools track your brand across platforms

AI visibility tools track your brand by parsing the raw responses from each platform and identifying where your business is mentioned (both linked and unlinked). When the tool captures a response from ChatGPT, Claude, Perplexity, Gemini, or Grok, it reads the text and the citations to spot any mention of your business. A direct link to your domain is one signal. An unlinked mention by name is another, and for brand visibility, just as important. Each platform surfaces brands differently: Perplexity cites with numbered references, ChatGPT tends to mention inline, Gemini uses grounded sources pulled from Google Search. Consistent tracking across all five means handling each format on its own terms so nothing is missed. Multiple runs per prompt matter because brand visibility is unstable: a mention in run 1 can disappear in run 2. Only by aggregating across runs can the tool give you a real read on how often your brand actually surfaces, rather than a one-off snapshot.

Share of voice compares your visibility to named competitors. If you asked the tool to track three competitors and ran 100 prompts, share of voice tells you what percentage of those 100 responses mentioned you versus each of them. Low share of voice in your highest-priority persona is one of the clearest flags the output produces. If a competitor shows up in 80% of responses and you’re at 10%, that’s not an abstract score. That is your customer being told about your competitor eight times more often than about you.

Why AI platforms give different answers to the same prompt

AI platforms give different answers to the same prompt because each model has different training data, different real-time retrieval behaviour, and different internal ranking systems. ChatGPT with browsing and Perplexity both search the live web, but they use different sources and weight them differently. Gemini grounds in Google Search results and exposes the sub-queries it fires. Claude and Grok draw on different knowledge bases again. Even within a single platform, answers to the same prompt vary between runs because AI models are probabilistic, not deterministic. Only 30% of brands stay visible in back-to-back AI responses (AirOps ). That inconsistency is why single-run checks are unreliable, and why multi-run consistency scoring exists. Running the same prompt ten times on one platform gives you a stable picture. Running it ten times on all five tells you how visibility varies across the AI ecosystem, not just one corner of it.

The accuracy check is the output that surprises most clients. When we run an AI citation checker for a new client and hand back the news that AI is misstating their headcount, attributing a competitor’s product to them, or describing services they stopped offering two years ago, that is often the first anyone’s confirmed it. Accuracy matters because it is actionable. Schema fixes. Third-party listing updates. Authoritative content that sets the record straight. You cannot fix a hallucination you do not know about.

Competitive displacement is the last output worth flagging. It shows you who is appearing instead of you, and what those competitors are doing differently. Often it is a content pattern: competitor X has long-form guides covering specific sub-topics AI keeps searching for, competitor Y has strong Reddit and LinkedIn presence feeding their mentions. That is raw input for your content strategy. Close the gap. Stop guessing at it.

How to track AI referral traffic using tools you already have

Running the visibility check regularly gives you one half of the picture: whether AI is citing you more over time. The other half is whether AI is sending actual visitors, and you can track that using tools you already pay for. Google Analytics shows AI referral traffic as a distinct source channel; ChatGPT, Perplexity, Gemini, and Claude show up under referral traffic whenever their answers contain links to your site. Google Search Console shows which queries are driving impressions and clicks from AI Overviews specifically, separate from the standard organic report. AI referral traffic is currently 1.08% of all web traffic and growing roughly 1% month on month, with ChatGPT driving 87.4% of it (Superlines ). Most businesses are not yet tracking this as its own channel. Setting it up is a 30-minute job, and it will matter more every month. Catch the upward slope now and you can tie visibility work back to real traffic, not just a tool’s score.

For the full playbook on how to action these insights, the rest of the cluster goes deep: AI citations covers offsite engagement and brand mentions, and AI content marketing covers the onsite content structure that AI actually rewards.

So, how do you find out if your business appears in AI search?

You find out by testing the prompts your audience actually uses, across multiple AI platforms, multiple times each, and reading the results properly. That is what every AI search visibility tool on the market is doing. Pick one, or build a version of it yourself. Both are valid.

Now for the uncomfortable bit. Nobody knows exactly how these AI platforms work behind the scenes. The weightings, the retrieval models, the ranking signals: all proprietary, all moving, largely opaque to outsiders. Anyone who tells you they have fully mapped the inside of ChatGPT is guessing, or lying.

What I can tell you is what we are seeing when we do this work. Persona-driven prompts validated against query fan-out data produce better visibility data than generic prompts. Multi-run consistency scoring across five platforms is more reliable than a single check on one. Treating an AI visibility audit as the starting point, pairing it with accuracy fixes, content changes, and offsite engagement, produces measurable change in visibility over months, not years. I’m not speculating. I’ve built the ai search visibility tool we use, watched it work across multiple clients, and iterated on it based on what the data was showing us.

88% of businesses are invisible in AI search today. The ones that will not be invisible in twelve months are the ones baselining their visibility now, understanding the mechanism behind the tools, and acting on what the results show.

If you want to see our ai search visibility tool in action, the GEO conversation opener is on our labs page. If you would rather we run the audit for you and hand you the report, get in touch .

Frequently asked questions