Measure whether the AI cites you. And fix what stops it.
More and more customers ask ChatGPT, Claude, Gemini or Perplexity instead of opening a results page. The APEspider GEO Tracker measures how often you show up in those answers, who shows up instead of you and which sources the AI uses to answer about your business. And because it lives in the same account as the crawler and your Search Console data, what you find out can be fixed without switching tools.
Measuring without being able to fix anything is a nice dashboard nobody opens twice.
Almost every AI visibility tool stops at the diagnosis: it tells you nobody cites you and that's where its job ends. The problem is that knowing changes nothing, because the lever is somewhere else — in your site and in your content.
In APEspider the loop closes inside the same account:
- You measure. The tracker fires your customers' questions at the four engines and tells you in what share of answers you show up.
- You see why. The report points out which pages take the citation instead of you and which sources feed that answer.
- You fix. The technical audit and your Search Console data, in the same app, tell you what the page that should be answering that question is missing: structured data, content that's visible without running JavaScript, a page that exists and is indexed.
- You write. APEscribe is included and turns that work into articles you only have to review and publish.
- You measure again. Repeating the report with the same setup is one click, and comparing it with the previous run tells you whether the work was worth anything.
Your own search data, not metrics some third party made up.
Half the SEO industry lives off numbers nobody can audit: domain authority, composite scores, estimated volumes presented as if they were facts. APEspider works on the only thing that is indisputably yours — your clicks, your impressions, your CTR and your real positions from Google Search Console — and the GEO Tracker starts there too.
That solves the hardest problem in all of GEO, which isn't asking a model things: it's knowing what to ask. You don't know what your customer types into a chat, but your search data hints at it. The tracker takes your Search Console and rank tracker keywords as seeds and builds conversational questions from them, each with its intent classified — how to, best, comparison, brand. You edit them, drop the ones that don't represent you and add your own.
Next to each keyword you'll see an AI volume column. It's labeled as what it is: a statistical estimate built from Google's related questions, not records of real conversations. It's there to prioritize which ones go into the report, never as hard data.
You decide what gets asked, to whom, from where and how many times.
A report is a matrix: prompts × engines × locations × repeats. Nothing is imposed and every multiplier is in plain sight before you run it.
- Your brand and its aliases. Mention detection looks for the name and the variants you give it — the brand, the domain, your own name — inside the text of every answer.
- Whichever engines you want. ChatGPT, Claude, Gemini and Perplexity. Each with its credit cost written next to it, because they don't cost the same.
- However many repeats you decide. How many times each question is repeated on each engine. More repeats, tighter margin of error: the form tells you what reliability you're buying before you spend a credit.
- The locations you work. Country and city, depending on what each engine really accepts.
- And the matrix, if you need to. If a prompt makes no sense on one engine or in one location, you drop that cell and it's neither run nor charged.
If you'd rather start with the minimum picture, the quick check fires a single question —something like "recommend me tools for X"— once per engine, with the cost capped and visible. Depth comes later, once you know it's worth it.
Five different things, and not one of them is a score.
- Whether the AI names you. A mention of your brand inside the text of the answer, with or without a link.
- Whether it cites you as a source. When the model attributes part of its answer to a URL of yours, with the link resolved down to its real domain.
- Which domains feed the answer. The ranking of sites the AI cites in your niche, including —above all— the ones that show up when you don't. That list is GEO's equivalent of link building: those are the sites where you're missing.
- Who shows up instead of you. You don't add the competitors yourself: they come out of the answers themselves. And you can ask for a head-to-head against a specific one.
- Why it didn't cite you. The model searching and not finding you is not the same as the model not searching and answering from memory. The report separates the two, because they're two different problems with different fixes.
In AI there is no position 3.
Ask twenty times which is the best tool in your sector and you'll get twenty answers that don't agree with each other. Models aren't deterministic: a single query isn't data, it's an anecdote. Any tool that hands you "your position" in ChatGPT is selling you a certainty that doesn't exist.
So the tracker repeats each question as many times as you decide and states the result as what it is: an appearance rate with its margin of error.
You show up in 12 of 20 answers · margin ±21 points
That sentence survives a meeting with whoever pays you, and it survives a comparison with next month. "Your GEO score is 68" survives neither, because nobody —not even whoever calculates it— can tell you what that 68 means exactly.
No composite scores here, no made-up volumes. Every number you see comes from real answers that are stored and that you can open and read exactly as they arrived.
The AI doesn't answer the same in Spain as in Mexico.
And you don't fix that by writing "in Spain" inside the question. Search location is an engine parameter: it changes where it searches before answering, and with it the sources it cites. APEspider sends it as a parameter — country and city on the engines that accept it, country only on the ones that go no further.
One of the four engines accepts no search location of any kind. In that case the form tells you so on that engine's own row and offers you the alternative: adding the location to the question text, labeled as per-prompt location and not presented as equivalent, because it isn't. The rule is the same throughout: each engine offers only the controls it actually runs, instead of accepting any setting and failing quietly.
What happens between hitting run and seeing the report.
Asking a model things is the easy part. What turns those answers into data you can work with is everything else:
- Validating before spending. Each engine accepts different parameters and its model catalogue changes without warning. The setup is checked against what the engine really supports before running, not after charging you.
- Running on its own queue. A report is dozens or hundreds of queries that take minutes to come back. They run on their own queue, without stealing room from your crawls or your rank tracker.
- Normalizing four different formats. Every engine returns the text, the sources and the metadata its own way. It all ends up in the same structure so the figures can be added up across engines.
- Resolving citations down to the real domain. Some engines return sources as intermediate links that say nothing. They're resolved at the moment they're stored, because otherwise the cited-domains ranking would be a list of redirects.
- Deduplicating and attributing. The same source cited three times in one answer is one source, not three. And you have to know which of those domains is you.
- Charging for what runs. The price is set per query and recorded in each run, so a report from three months ago still knows what it cost even if the rate has changed since.
- Keeping the raw answer. Complete, with its sources and its metadata. Re-reading a report you already paid for costs no credits, and you can always drill from any figure down to the exact text the model returned.
None of this shows up on screen, and it's exactly what separates a defensible number from a ChatGPT screenshot.
The questions I get about this
"Does ChatGPT recommend me when someone asks about tools in my sector?"
That's exactly what the report measures. You fire the question as a customer would write it, repeat it N times on the engines you pick and get the share of answers you show up in, with its margin. It's not a yes or a no: it's "12 out of 20".
"What sources does the AI use to answer about my sector?"
Every answer arrives with its attributed sources, resolved to their real domain. The report aggregates them, so you end up with the ranking of domains the models cite in your niche — including the ones that show up when you don't. That list is your to-do list.
"Which competitors does the AI compare me with?"
You don't add the competitors yourself: they come out of the answers themselves. Whoever takes the mention when the question was meant for you shows up in the report, and you can ask for a head-to-head against a specific one.
"I don't know what people ask ChatGPT about what I do. Where do I start?"
With your own data. APEspider starts from your Search Console and rank tracker keywords and builds conversational questions from there, with their intent classified. You arrive with a domain and leave with a list of realistic prompts you can edit.
"Why repeat the same question several times?"
Because models aren't deterministic: the same question produces different answers. A single query tells you nothing, it just gives you an anecdote. Repeating and counting turns the anecdote into a rate with a margin of error, which is the only thing you can compare month to month.
"Is it any use if I work Spanish-speaking markets outside Spain?"
Yes, and that's one of the reasons it exists. Search location is sent as a parameter, so you can fire the same prompt from Spain and from Mexico and see that the answers and the cited sources don't match.
"How is this different from the rank tracker?"
The rank tracker measures positions on the results page. The GEO Tracker measures appearances inside AI answers, where there are no positions, only probabilities. They complement each other: what makes a model cite you is usually the same thing that makes Google rank you, and both live in the same app.
"I already know the AI doesn't cite me. Now what?"
The report tells you which questions you don't show up in and which sources take the citation instead of you. From there, in the same account: the technical audit says what the page that should be answering that question is missing, APEscribe helps you write it, and you repeat the report to see if anything moved. That loop is the product.
"How much does it cost to run a report?"
You pay in credits and every engine has its own, written next to the engine before you run anything. The total depends on how many prompts, engines, locations and repeats you pick, and it recalculates in the form as you set it up. Re-reading a report you already paid for costs no credits.
"Is it finished?"
It's in beta and working: you set it up, you run it and it gives you the report. What's coming after that is scheduled tracking — repeating on its own, with its history and its alerts — without you having to fire it every time. It's in the Founder License at no extra cost.
It's included in the Founder License.
The GEO Tracker isn't sold separately: it's part of APEspider PRO along with the technical audit with its own crawler, your Search Console data, the SERP rank tracker by market, the Word and Excel reports and APEscribe. A subscription of €290 a year — €24 a month — and everything published during your year included at no extra cost.