Frequently asked questions
Last updated: 8 August 2026 · Ελληνική έκδοση
What customers ask before and after signing up, answered properly. Anything missing? Write to info@maivisibility.com.
What GEO is
What is Generative Engine Optimization (GEO)?
GEO is the work of appearing — and more importantly being recommended — inside the answers AI assistants give. The difference from SEO is structural: a search engine returns ten links and the user chooses, while an AI returns one answer with one or two recommendations. If your brand is not in that answer you do not exist for that buyer; you are not "further down the list", you are absent. GEO means understanding which sources the AI reads for your category, being present in them, and giving your own site the structure that lets a model cite you safely.
Why does this matter now rather than in two years?
Because the shift has already happened. Buyers ask "which POS system for a coffee shop", "which invoicing software connects to myDATA", and get answers containing names. What matters is that AIs form their view from sources that already exist — directories, articles, comparison pages. Whoever gets into those sources early accumulates recommendation for a long time. The cost of waiting is not losing a rank position; it is that someone else becomes the default answer.
How is this different from an SEO tool like Semrush or Ahrefs?
Classic tools measure positions on results pages: keyword, rank, search volume, backlinks. mAI Visibility does not measure positions — it runs your buyers' questions inside the AIs themselves and reads the answer: whether you are mentioned, in what order, whether you are recommended or merely listed, which competitors appear beside you, and which sites the model cited to get there. It is a different unit of measurement, not a better version of the same one. The two coexist.
Does GEO replace SEO?
No. Most sources AIs trust are pages that already rank well organically, so good SEO helps GEO. But it is not sufficient: a site can rank first on Google and never be cited by ChatGPT because it lacks the structure that lets a model extract claims safely — clean answers to questions, structured data, unambiguous organisation identity.
How measurement works
How do you measure whether a brand appears in ChatGPT?
We do not infer it — we ask. You build a library of your buyers' real questions. The platform runs each one against the AIs with live web search, stores the whole answer, then analyses it: it finds your brand and its aliases in the text, records where it appears, whether the model recommends it or merely mentions it, which attributes it assigns, which competitors appear, and which sources were cited. Every run is stored so the same question can be compared over time.
Do you use live web search or static model answers?
Live. ChatGPT with web search, Google AI with Google Search grounding, Claude with its web search tool, Perplexity with search built in. This matters: an answer without search reflects the model's memory from training — the world as it was months ago — while your buyer gets an answer built from today's sources. We measure what the customer sees, not what the model remembers.
How many prompts are needed for a reliable measurement?
In practice 15–40 questions cover one category in one market. Below 10 the picture is noisy: a single question can flip between runs, so your mention rate jumps without anything having changed. Above 40 you usually add variations of the same intent and pay without learning more. A good split: one third category questions, one third constrained ("X that integrates with Y"), one third comparative and branded.
How often should GEO monitoring run?
Weekly is the default and right for most: AI answers change when their sources change, which happens on a scale of weeks. Daily only makes sense during active intervention or in a very competitive category. The cycle is configurable per AI, so you can run ChatGPT weekly and the rest monthly to keep cost down.
How quickly do results from an action show up?
Technical changes on your own site show in the technical audit immediately, but usually take 2–6 weeks to reach AI answers — the time needed for the page to be re-crawled and re-read. Listings on external sources are slower and stronger. This is why the platform measures the same question before and after: to separate real movement from noise.
What data does a company need to provide to start?
Very little: VAT number, the site URL and the brand name. From the Greek VAT number we pull official company details from the ΓΕΜΗ business registry, and by reading your site we draft the brand description for you to edit. Then you add aliases, competitors if you know them, and the first questions.
AI providers
Which AI systems do you measure?
Four: ChatGPT (GPT models with live web search), Google AI (Gemini with Google Search grounding), Anthropic Claude (with its web search tool) and Perplexity (sonar models). We additionally track Google AI Overviews — the answer box Google places above the organic results, which follows its own citation rules.
Why do results differ so much between AIs?
Because they are different systems with different sources. Google AI leans on Google's index, Perplexity on its own search pipeline, ChatGPT and Claude on theirs. It is entirely normal for a brand to be the top recommendation in one and absent from another. That is what the Prompt × Provider matrix is for: to show the disagreement and tell you which ecosystem you have a problem in.
How do you handle Greek and English names for the same brand?
With aliases. You declare every form your brand is referred to by — Greek, English, with or without legal suffix, abbreviations, former names — and the analysis treats them as the same brand. Without correct aliases your visibility is systematically underestimated.
Results and metrics
What is the difference between mention, recommendation and top recommendation?
Three levels, measured separately because they mean very different things commercially. Mention: the brand appears somewhere in the answer, possibly in a list. Recommendation: the model actively proposes it. Top recommendation: it is the first or main recommendation. There is also a fourth state, exclusion: the model knows you and explicitly rules you out for that need — usually a sign of wrong or missing information in the sources, and fixable.
How is the AI visibility score calculated?
It combines, on a 0–100 scale, mention rate, recommendation rate, top-recommendation rate, average position within the answer, and the authority of the sources citing you. Every score carries an explanation and history — no black boxes. The full breakdown is on the methodology page.
What is share of voice?
Your share of all brand mentions across the same questions. If 20 questions produce 100 brand mentions in total and 12 are yours, your share of voice is 12%. It is the most honest competitive metric, because it measures the space you occupy relative to everyone else rather than your absolute performance.
What counts as a citation and why does it matter?
A citation is a source the AI referenced to support its answer — an article, a directory, a comparison page. It matters because it is the mechanism by which the answer changes: you do not persuade the model, you change the sources it reads. The platform records every source, how often it appears and in which AI, and manages it as a listing pipeline with a ready-to-send outreach email.
AI answers change every time. How do you handle that?
It is a real problem and we do not hide it. Models are probabilistic: the same question can produce different answers on consecutive runs. We handle it three ways: measuring many questions so noise cancels out, keeping history so trend separates from variance, and for every question that moves, producing an explanation — what changed in the answer, who entered, which source stopped being cited. A single number from a single run is not a conclusion.
What do I get beyond measurements?
For every question you lose, a concrete plan of 3–4 actions on channels you control: which page to create, which source to get listed on, what to fix on the site. Plus content grounded only in facts you have approved, automatic llms.txt and schema generation, a technical site audit including Google PageSpeed, and weekly reports. When you mark an action done, we re-measure the same question before and after.
How do I know whether a rate is reliable or just noise?
Every rate is shown with its interval and the number of runs behind it. The interval is a Wilson score interval at 95%, not the textbook normal approximation — that one breaks down at exactly the small samples real use produces and can report an upper bound above 100%. In practice: at 3 runs, 67% and 33% are statistically indistinguishable, while around 8–10 runs per context the interval narrows enough to carry a decision. The product shows everywhere how far from that you are.
What is a neutral measurement and how does it differ from a persona measurement?
Neutral is the primary benchmark: clean context, no persona, a fixed system prompt and temperature 0, identical settings on every run. A persona measurement places the buyer's own words before the question, so the assistant answers as it would for that kind of customer. The two are never merged into one number. Neutral is the absence of a persona, not a persona named “neutral” — if it were an editable record, a wording change would silently break comparability with every earlier measurement.
What is an audience profile and why do I need one?
It is the real characteristics of your customers: business size, turnover, operating model, working environment, technology literacy, ranked needs, pain points, purchase criteria and budget. You fill it in through an eight-question wizard where the AI drafts and you decide. You need it because personas generated from the industry alone come out the same for every company in that industry, so they predict nothing. The profile also states explicitly who the product is not for, without which the generator invents segments that would never buy.
What is the Persona Relevance Score?
An assistant that recommends a multi-thousand-euro ERP to a market stallholder can rank you first and help nobody. A scoreboard counting only order of appearance records that as a win. So every persona measurement is also scored on whether the recommendation actually suits the buyer who asked, across eight axes: budget, business size, mobility, features, technology, working environment, complexity and operating model. It sits beside the mention rate, never folded into it: the interesting case is high visibility with low relevance.
How many personas and how many runs do I need?
Few personas with many runs, not the reverse. The common mistake is 12 personas with 2 runs each — it yields nothing, because none of the twelve rates has enough sample behind it. Three or four personas matching real customers, with roughly ten runs each, give intervals narrow enough to act on. The rest can be disabled for now and enabled when their turn comes.
If I edit a persona, what happens to the earlier measurements?
Every change to the wording mints a new version, and each measurement stays pinned to the version it was taken under. Earlier measurements are not altered and remain explainable. Without this, a wording fix would retroactively redefine every earlier measurement: the numbers would still add up and would have stopped meaning anything. With it, a drop in visibility can be told apart from a change in the question.
How is product fit against an audience computed?
From two stored profiles rather than asked of a model each time: the audience states what it requires and the product states what it offers, on the same eight axes. The score is weighted by your own purchase criteria. An axis on which the audience expressed no requirement counts in neither direction, and exceeding a requirement is a full match. Every claim in a product profile keeps its source and whether it has been verified — a score built on unverified marketing copy is a score for the marketing copy, and you can see which one you have.
I serve more than one audience. How does that become a single number?
Each audience is measured separately and combined using the weight you assign, so the headline describes your real market rather than an average over unrelated segments. With no weights set, tier defaults apply: primary ×3, secondary ×2, experimental ×1. The weighted figure is renormalised over the audiences that actually have measurements — an unmeasured audience is missing from the calculation, it does not count as zero.
Can I see what a batch run will cost before starting it?
Yes. Before any batch you get an estimate drawn from your own recent charges, per provider — not from a price table, because the real charge depends on the model, the answer lengths and your account's multiplier. It shows an average and a ceiling the cost is unlikely to exceed, alongside your available balance. From the same dialog you choose providers and which questions to run, and the cost recalculates live.
Can I use my own OpenAI or Anthropic keys?
Yes, provided the platform administrator enables it for your account and you have an active subscription. Keys are stored encrypted, never returned and never logged; you see only the last four characters. When your own key is used, that usage is not charged against the platform AI wallet.
Agencies
Can an agency use it for multiple clients?
Yes. An agency account manages many client organisations, each with its own workspace, brands, markets and reports, with strict data isolation between them. There are roles and permissions per team and client, a full audit log, and PDF reports with a public share link you can send to the client.
Do clients created by an agency get the free month?
No. Organisations created by an agency are provisioned bare: no trial subscription and no starter credit. The reason is simple — free resources exist so a business can try the product once, not to be multiplied through new accounts. The agency funds its clients.
Cost and subscription
What does it cost?
Subscription pricing has not been announced yet and will be published soon. What is true today: the first month is free with no card and includes a starter AI credit, and usage beyond that is prepaid with per-action cost reporting. You will not find prices on this page because we do not want to publish numbers that may change.
What is the AI wallet and what consumes credit?
A prepaid balance charged for every action that consumes AI: running a question against an AI, analysing the answer, generating content, the consultant-grade site analysis. What does not consume credit: the technical site audit, Google PageSpeed measurement, the business-registry lookup, and the in-app support assistant. Every charge is logged with date, action, model and amount on the Billing page.
What happens when the balance runs out?
AI features stop and a top-up message appears. There is no automatic charge and no meaningful negative balance beyond the call already in flight. The technical audit and existing reports stay accessible.
Can I cancel with no commitment?
Yes. There is no minimum term. Unused credit is retained but usable only with an active subscription, and is not redeemable for cash. Details are in the Terms of Use.
What do I have to do to activate the free month?
Nothing. As soon as you finish signup with SMS verification and register your tax ID, the free month and the €5 of starting AI credit activate automatically — we confirm it in the app and by email. There is no activation task and no card required.
Data, security and GDPR
Are prompts and AI answers stored, and for how long?
Yes — it is the substance of the product. Without the stored answer there is no comparison with last week and no explanation of what changed. We store the text of each answer, the sources cited, the analysis and the timestamp, for as long as your account exists. Deleting the account deletes them.
Is my data sent to third-party AI providers?
Yes, necessarily: to measure what ChatGPT says about your brand we must send the question to ChatGPT. What is sent: the questions, the brand name and description, and content you have approved for generation. What is not sent: your end customers' personal data. The providers used are OpenAI, Google, Anthropic and Perplexity.
Is my data used to train models?
Not by us. We use the providers' APIs rather than their consumer apps; data passed through the API is not used for training under their terms. We do not train models on customer data.
Is mAI Visibility GDPR-compliant?
The service is built for the European market and operated by a Greek company based in Athens. Each organisation's data is strictly isolated, access is controlled by roles and permissions, there is a full audit log and optional MFA. What is collected, on what legal basis, where it is hosted and how you exercise your rights are set out in the Privacy Policy. A DPA is available for agencies on request.
Who is behind the platform?
RETAIL MANAGEMENT SOLUTIONS ΜΙΚΕ, Grammou 73, 15124 Maroussi, Athens, Greece. VAT 801291892, business registry (ΓΕΜΗ) 153674701000. mAI Visibility (internally also mAI GEO) is one of the company's mAI products. Contact: info@maivisibility.com.
Comparisons and choosing
How does a platform differ from a GEO agency service?
Two different things, often combined. An agency service gives you people: strategy, execution, content, one accountable owner. A platform gives you continuous measurement: the same questions running every week against the same AIs, history, alerts when something changes, and before/after proof for each action. A one-off GEO audit is a photograph; monitoring is a film. If you already have an agency, the platform gives it the numbers it currently lacks; if you do not, it tells you what to ask for.
When is it not the right choice?
If nobody will implement the recommendations. The platform measures, recommends and drafts content — but someone has to approve and publish it. Also if your audience does not consult AI for that decision: in purely local, impulse categories the effect is smaller than in research-and-compare markets.
Can I try something before creating an account?
Yes — the free AI-readiness check scores any site with no signup, in about 20 seconds. It is the technical check only: it does not include measurements inside the AIs, which is the main product. One check per site.