AI visibility analysis

Does ChatGPT name your brand when someone asks about your industry?

We measure whether ChatGPT, Gemini, Claude, Grok and Perplexity mention your company, cite your website as a source and recommend you. Then we carry out the work needed to change the picture.

Request an analysis See how we measure

Rather start with the technical side? Check for free whether AI crawlers may read your website, no sign-up.

At a glance

GEO (Generative Engine Optimization) is the optimisation work that makes an AI system name, cite and recommend a brand inside its answers. An AI visibility analysis measures those three things separately, because each one fails for a different reason: mention (does the model write the brand name?), citation (does it name your domain as a source?) and recommendation (does it actively propose the brand?).

Freshlab Iberia measures across ChatGPT, Gemini, Claude, Grok and Perplexity through the official APIs, with web search enabled, using queries written natively in English, Spanish, German and Danish. The measures that follow from the report are then carried out in house, which is what separates the service from a subscription monitoring dashboard.

This page was last reviewed on 25 July 2026. Methodology and delivery: Manfred Henning, sole director of Freshlab Iberia S.L.U., more than 30 years in IT, business processes and pharmaceutical contract manufacturing.

The blind spot

More and more buyers ask a model, not a search engine

When the answer arrives already written, with two or three suppliers named, being on that shortlist stops being marketing and becomes distribution. Most SMEs are not on it, and have no way of finding out.

Invisible with no warning signal

In a classic result list you can look up your position. In a generated answer there is nothing to look up: if your brand is missing, it is simply missing, and no analytics dashboard will flag it.

A website blocked by accident

The single most common finding: robots.txt blocks GPTBot, ClaudeBot, PerplexityBot or Google-Extended. Often an agency added those lines years ago with the best intentions. The effect today is that the models never read your site.

A competitor holds your slot

Someone does show up in the answers for your industry. Usually it is not the best operator, it is the one present in the sources the models read: trade portals, directories, local press, review platforms.

Wrong facts about your company

Models also invent things: the wrong head office, services you do not offer, a mix up with a similarly named business. We check the claims made about your brand against your real data and document the false ones.

Three figures, with date and source

  • 900 million weekly users was the ChatGPT figure reported in February 2026, per OpenAI's official statement, which has not been updated since. That is the order of magnitude of the audience now asking a model instead of a search engine.
  • More than 2.5 million websites were fully blocking AI training in August 2025, according to network data published by Cloudflare. Blocking training is a legitimate decision. The problem starts when the same rule also blocks retrieval at answer time, because then the model can no longer read the site at all.
  • Gartner predicted in February 2024 that search volume would fall by 25 percent by 2026. As of mid 2026 that drop has not happened at that scale: search has not collapsed, it has shifted. We quote it because we would rather give you an inconvenient figure than a headline number.

How we measure

Three levels, three causes, three different remedies

Reducing brand visibility in ChatGPT and other models to a single score sounds tidy and helps nobody. We always separate mention, citation and recommendation, because each level fails for a different reason and is fixed with different measures.

Level 1

Mention

Does the model write your brand name into the answer?

Typical cause when it fails
The models do not know your company as an entity. There are not enough consistent signals about who you are and what you do.
Measure
Entity groundwork: one consistent self description across the whole site, Schema.org markup, a solid company page, aligned directory listings, presence in third party sources.
Level 2

Citation

Does the model name your domain as the source of the answer?

Typical cause when it fails
Your website is not reachable for AI crawlers, or it is not quotable: the content depends on JavaScript, it does not answer concrete questions, or it offers no data of its own.
Measure
Crawler access, rendering without JavaScript, answer formatting (the question as the heading, the answer in the first two sentences, figures with a date and a source) and publishing your own data.
Level 3

Recommendation

Does it put you forward positively, or just list you in passing?

Typical cause when it fails
Evidence is missing: reviews, documented projects, press mentions, references that can be verified. The model knows you but has no reason to recommend you.
Measure
A review strategy, published case studies, presence in trade and local press, and handling of the negative mentions we find in the answers.

A real example of why the split matters: a company can show a healthy mention rate and zero citations. The models name it because a trade portal lists it, but they never read its website. The reverse case, plenty of citations and few mentions, calls for exactly the opposite measures. Behind a single score, both cases would look like the same problem.

The process

From measurement to result in four steps

  1. 1. Technical check of your website

    We crawl your site in line with your robots.txt and review what decides whether the models can read it: AI crawler access, structured data and its required fields, readability without JavaScript, consistency of name, address and description, language versions with hreflang, answer formatting and how current the content is.

  2. 2. Measurement across five platforms

    We build a query set from your website, your industry and your region, written natively in each language. It covers every type of intent, from the informational question through to a straight supplier comparison. In most queries your brand is not named, because otherwise we would be measuring self confirmation instead of visibility. Every query is repeated so that we can work with frequencies.

  3. 3. Report with an action plan

    You receive the three rates by language, platform and query type, the comparison with your competitors, a ranking of the sources the models actually cite in your industry, the striking answers quoted verbatim, and an action plan prioritised by effect and effort. The scoring formula is in the report and can be checked.

  4. 4. Execution

    This is where tools stop and we start. We carry out the measures: technical fixes, content in answer format, structured markup, aligned listings, press work and training for your team. You decide what your own people take on and what we handle.

Coverage

Five platforms, always with web search enabled

We use each provider's official API only, with no scraping and no circumvention of access restrictions. The default models are the economy tiers, because they come closest to what a free user sees.

ChatGPTOpenAI
GeminiGoogle
ClaudeAnthropic
GrokxAI
PerplexityPerplexity AI

For every answer we record whether the system really searched the web and which domains it cited. That source list is the most useful part of the report: it shows precisely where your brand needs to be present.

Why us

Tools measure. We measure and then fix it

Subscription monitoring dashboards exist, and some of them are good. Measurement is not the problem. The problem is that afterwards somebody has to do the work.

Aspect Subscription monitoring dashboard Freshlab Iberia
What you receive A dashboard of figures you have to interpret yourself Diagnosis, interpretation and a prioritised action plan
What happens next Execution stays on your desk We carry out the measures in house
Who operates it Needs a marketing department Built for SMEs without a marketing department
Languages Translated queries, or English only Native phrasing per language, ES, DE, EN, DA
Technical basis of the website Out of scope in most cases Technical check included: crawlers, schema, rendering
Who you deal with Chat support, often outside the EU Spanish company, EU contract, one named contact

Packages

Four formats, depending on how far you want to go

Scope, languages and the number of queries drive the price, which is why we always quote individually after a first conversation.

Snapshot

The initial diagnosis. Knowing where you stand before you invest in anything.

  • Reduced query set in one language
  • Technical check of the website
  • The three rates: mention, citation, recommendation
  • Short report and a walkthrough call

Individual quotation

Deep Dive

The full analysis with competitor comparison and a 90 day roadmap.

  • Broad query set, several languages, all five platforms
  • Comparison with your competitors
  • Ranking of the sources the models cite
  • Check of false claims made about your brand
  • Prioritised 90 day action plan

Individual quotation

Deep Dive with execution

The analysis plus the implementation of the measures by us.

  • Everything included in Deep Dive
  • Technical fixes on your website
  • Content in answer format and structured markup
  • Aligned listings and work on external sources
  • Control measurement once the work is done

Individual quotation

Monitoring

Continuous tracking. AI visibility moves, and so do your competitors.

  • Monthly measurement with the same versioned query set
  • Development against the previous measurements
  • Alert when a new competitor enters the answers
  • Short monthly report with recommendations

Individual quotation

Honesty first

Who this service is not for

We would rather say this before the first invoice than after it. If you recognise yourself in any of these four points, let us talk it through, and we will probably point you somewhere else.

If you only want a dashboard

If you have your own marketing team and only need numbers to interpret yourself, there are monitoring subscriptions that cost less than this service. Our value sits in the interpretation and in the execution.

If you expect a guarantee

If what you want is a promise of appearing in the answers, no serious provider can give you that, and neither can we. We can commit to the diagnosis, the plan and the work, not to what a model says.

If there is nothing to measure yet

If your website is still a one page brochure with no content of its own, measuring adds nothing, the result is a foregone conclusion. Something citable has to exist first, and that is a different piece of work.

If it has to work within two weeks

Technical fixes show up quickly, but entity foundation and reputation take months to appear in the answers. If your window is a fortnight, this is not the right route.

Frequently asked questions

AI visibility and GEO: the questions we hear most

What is GEO (Generative Engine Optimization)?

GEO is optimisation work aimed at generative AI systems. Instead of competing for a position in a list of ten blue links, the goal is that a model such as ChatGPT, Gemini, Claude, Grok or Perplexity names your brand in its answer, cites your website as a source and recommends you when someone asks for advice. It is a separate discipline from classic SEO, because the output is not a ranking but a text written fresh for every query.

How does GEO differ from classic SEO?

Classic SEO works against a stable result list that rank tracking tools can measure. In generative systems there is no list: there is a written answer that changes between queries, languages and platforms. Measurement therefore relies on repeated query sets that are evaluated statistically. The technical foundations overlap (a crawlable website, correct structured data, content readable without JavaScript), but content and reputation factors carry different weight.

What exactly do you measure: mention, citation and recommendation?

Three separate levels with separate causes and separate remedies. Mention: the model writes your brand name into the answer. Citation: the model names your domain as the source of the information. Recommendation: the model puts you forward positively rather than listing you in passing. A company can have a solid mention rate and zero citations, because the models know it through an industry portal but never read its website. That diagnosis calls for entirely different measures than the opposite case.

Why do your results differ from what I see in the ChatGPT web interface?

We measure through each provider's official API with web search enabled. Consumer interfaces add further layers: personalisation based on your history, conversation memory, A/B tests and different model versions. Generative models are also not deterministic, so the same query can return different answers. That is why every query is repeated and the results are reported as frequencies rather than individual screenshots.

Do you guarantee that my brand will appear in the answers?

No, and nobody working seriously in this field can. What we measure are mention frequencies under defined and documented conditions. Model answers vary, providers update their systems without notice, and API answers differ from the consumer web interfaces. What we do commit to is a reproducible diagnosis, a concrete action plan and the execution of those actions.

Which languages do you work in?

Spanish, German, English, Danish and other European languages on request. Queries are written natively in each language rather than translated: a German buyer phrases a question differently from a Spanish one, and translated queries would distort the result. This matters because many companies are visible in one language and invisible in another without knowing it.

What does an AI visibility analysis cost?

We work with four packages: Snapshot (initial diagnosis), Deep Dive (full analysis with competitor comparison and a 90 day plan), Deep Dive with execution (includes implementing the measures) and a monthly monitoring retainer. The price depends on scope, the number of languages and the volume of queries, so we prepare an individual quotation after a first conversation.

Is the analysis GDPR compliant?

Yes. The queries we send to the models contain no personal data, only market and brand questions. We do not scrape the consumer web interfaces and we do not circumvent access restrictions: we use the official APIs only. The crawl of your website respects your robots.txt. Freshlab Iberia S.L.U. is a Spanish company, so your contracting party sits inside the EU. More detail in our Privacy Policy.

Methodology note and limitation of liability

What we measure are mention frequencies under defined and documented conditions: a specific version of the query set, specific models, a specific point in time and web search enabled. These are not guaranteed placements, and they cannot be.

The answers of generative models vary between runs, even for an identical query. Providers update models and search mechanisms without prior notice. Answers obtained through an API differ from those shown in the consumer web interfaces, which add personalisation and further layers. Every result is therefore expressed as a frequency across repetitions, never as a single screenshot.

We use the providers' official APIs only. We do not scrape the web interfaces and we do not circumvent access restrictions, and the crawl of your site respects your robots.txt. Every statement in the report can be traced back to the specific answer that supports it, and those answers are stored.

First step

Request your AI visibility analysis

Tell us your website and your industry. We reply with a proposed scope and an individual quotation, with no obligation.

Your data is processed on our server in Spain (EU) in order to handle your request. Legal basis: Art. 6(1)(b) and (f) GDPR. Spam protection via Cloudflare Turnstile. More information.

Prefer to talk it through first?

Twenty minutes on a call is usually enough to tell whether your case warrants a full analysis or whether a Snapshot will do.

Talk to us See also Local AI