Marbella SEO Agency Insights

AI Search Visibility Insights & Research

Original research into how businesses appear in AI-generated recommendations, search results and cited sources. Marbella SEO Agency analyses visibility across leading AI platforms to identify which businesses are recommended, how often they appear and which sources the assistants cite.

Flagship research

Marbella AI Visibility Index

The Marbella AI Visibility Index examines how businesses appear in AI-generated recommendations and cited sources when people ask relevant commercial and informational questions.

Each study compares visibility across industries, AI providers and a defined set of search prompts, published word for word. Measurements come from controlled research samples. They are not an exhaustive measurement of every AI interaction, and we report them that way.

What each study compares

  • Industries – five Marbella sectors, each with its own permanent research page.
  • AI providers – OpenAI, Google Gemini and Anthropic Claude, with exact model versions recorded.
  • Prompt sets – defined, published questions, repeated to account for variation between answers.

AI visibility by industry

AI visibility by industry

Each industry has one permanent research page. When a new study is verified, it updates that page rather than creating a new report.

What we measure

What we measure

Recommendation frequency, ranking position, mentions and citations describe different things, and we keep them separate. A study shows only the metrics it has actually calculated.

01AI Recommendation Frequency

The number of valid AI responses in which a business is explicitly recommended. It is a count of responses, not a position.

02AI Visibility Rate

The percentage of valid, relevant AI responses that recommend a particular business.

03AI Brand Mentions

How often a business is mentioned at all, including mentions that are not explicit recommendations. Counted separately from recommendations.

04AI Source Citations

Verified citations or source references associated with a business or its website, where the provider exposes suitable citation information.

05AI Visibility Trends

Changes in measured visibility between comparable research datasets: same methodology, prompt set and comparable model configuration.

Research methodology

Research methodology

This is our general research framework. Each published study also includes its own methodology section on its industry page, stating the exact prompts, models, dates, sample and any deviation from this framework. Where the two differ, the study-specific methodology describes what was actually done.

Research question selection

Prompts are written to reflect how customers actually ask for a business in each sector: by need, location, budget and occasion. Wording is neutral and never names or hints at a particular company.

Commercial and recommendation intent

Each prompt set is weighted towards questions where a person expects named suggestions, such as “which agency” or “where should we stay”, with informational questions included where they commonly lead to recommendations.

AI providers and model configurations

Studies use OpenAI, Google Gemini and Anthropic Claude. For each we record the exact model version and relevant configuration, such as whether web access or grounding was enabled.

Repeated measurements

AI output varies between runs, so each prompt is sent more than once to each provider. The number of repetitions is published with every study.

Languages tested

Each study states its languages. Prompts are published in their original language and wording; a prompt tested in English says nothing about Swedish, Finnish or Spanish answers unless those were also tested.

Business identification and name normalisation

Different spellings and brand variants are mapped to one verified business identity using its website and location. Ambiguous names are flagged and left uncounted rather than guessed.

Recommendation extraction

A business is counted as recommended only when the response explicitly suggests it. Mentions in passing, lists of things to avoid and generic references are recorded separately as mentions.

Ranking position analysis

Where a response presents an ordered list, the position of each business is recorded. If order cannot be established reliably, position-based metrics are not reported.

Source citation verification

Cited URLs are checked and attributed to a business only when the source clearly relates to it. Citation metrics are reported only for providers and configurations that expose sources.

Scoring methodology

Rankings are ordered by a stated metric, usually recommendation frequency across valid responses, with the tie-breaking rule disclosed. No undisclosed weighting is applied.

Sample size

Every study states the number of prompts, repetitions, providers, collected responses, valid responses and excluded responses, so readers can judge the base of each figure.

API-based testing versus consumer applications

Our data is collected through provider APIs under controlled settings. Consumer apps can add personalisation, location, memory and different default models, so results describe the tested configuration, not every user’s experience.

Research limitations

Results are a controlled sample from a defined period. They do not measure service quality, predict future answers or cover every question a customer might ask.

Reproducibility and editorial review

Prompts, model versions and dates are recorded so a study can be repeated. Every dataset is reviewed by an editor before publication, and commercial relationships have no influence on inclusion or order.

Latest research & findings

Latest research & findings

The most recent approved study for each industry. Research is published when verified results are available, not on a fixed schedule.

How Visible Is Your Business in AI Search?

Understand how your business appears in AI-generated recommendations, how competitors are represented and where opportunities for improvement may exist.