Methodology

We hold a fixed list of questions for each sector, written the way people actually ask them. On a weekly schedule we put every question to four AI assistants and store the complete answer. We then check each stored answer for every business we track in that sector.

Because one answer is checked against every business at once, tracking more businesses does not change what is asked or what any engine returns.

The exact model matters: two versions of the same assistant can answer differently, so a figure is only reproducible if the model is stated.

Each question therefore produces 7 answers per run. We ask each assistant twice where we can, once using only what the model already knows, and once letting it search the web, because the two often disagree, and collapsing them into a single number would hide that.

Perplexity always searches the web; it has no offline mode to test. It contributes one answer per question rather than two, and we do not invent a second.

We record two different things and never merge them:

• Mention rate, how often the business's name appears in the text of an answer. Matching is done by exact name and a short list of confirmed alternative names, ignoring case, punctuation and accents. Partial words never match: an answer about “Rock Hoteliers” is not a mention of “Rock Hotel”.
• Citation rate, how often an assistant linked to a page on the business's own website as a source.

A business can be named without being cited, cited without being named, or both. Reporting one as though it were the other would misstate what actually happened, so we publish them separately.

In order of precedence:

1. If the first appearance falls inside a numbered list, position is that list number.
2. Otherwise, position is the rank of the first appearance among all businesses named in that answer.
3. A business cited but never named takes the rank of its citation in the assistant's source list.

The exact character position of every match is stored, so this rule can be re-examined against past answers without re-running anything.

Questions run weekly. Published figures cover a trailing 28-day window rather than a single run, because AI answers can shift noticeably from a model update alone and a single-run figure would swing on noise.

Every individual run is kept permanently and never edited. Only the published figure is smoothed.

Sponsored placements. AI assistants have begun carrying paid placements in consumer products. Those placements are not exposed through the interfaces we use, so we cannot detect them. Every appearance we record is treated as organic, not because we checked and found no advertising, but because we have no way to check. We will say so here if that changes.

Answers vary between people. Assistants personalise, test variants, and change over time. Our figures describe what was returned to our requests, not what every person everywhere sees.

Failed runs are excluded, not counted as absences. If an assistant is unavailable, that run is recorded and then left out of the totals. Counting it would make every business look worse whenever a provider had an outage.

Sector pages show how many businesses are tracked, the shape of the distribution, and the single most-mentioned business named as a highlight. We do not publish the position of any other individual business.

A business can obtain its own figures. We do not publish one business's standing to another.

Every question is published verbatim on its sector page. There are currently 210 active questions.

• Cruise & Port , 60 questions
• Education & Training , 5 questions
• Restaurants & Hospitality , 30 questions
• Insurance , 15 questions
• Legal , 5 questions
• Real Estate & Property , 5 questions
• E-commerce , 5 questions
• Fintech & Payments , 5 questions
• Fund Administration , 5 questions
• Yacht & Marina Services , 5 questions
• Automotive , 5 questions
• Banking , 5 questions
• Gaming & Betting , 5 questions
• Retail , 5 questions
• Accountancy & Tax Advisory , 5 questions
• Construction & Trades , 5 questions
• Web Design & Digital Marketing , 5 questions
• Health & Wellness , 5 questions
• Cross-Border Shopping , 5 questions
• Yacht & Marina , 5 questions
• Duty-Free Retail , 5 questions
• Hotels & Accommodation , 5 questions
• Relocation & Residency , 5 questions
• Weddings , 5 questions

Changing a question's wording creates a new question rather than editing the old one, so past results always belong to the exact words that produced them.

Is Visibility a rating agency?

No. We do not rate, score or grade any business. We run a published set of questions against AI assistants and report which businesses appeared in the answers. There is no composite score, no weighting and no tier.

What does 'mentioned' mean?

A business is recorded as mentioned when its name appears in the text of an answer, or when the engine cites a page on that business's own website. These are tracked and reported separately as mention rate and citation rate, because they are not the same fact.

How is position calculated?

If a business's first appearance falls inside a numbered list, its position is that list number. Otherwise it is the rank of its first appearance among all businesses named in that answer. A business that is cited but never named takes the rank of its citation in the engine's source list.

How often do the numbers update?

Questions are run weekly. Published figures are rolled over a trailing four-week window so a single unusual run does not move a number on its own.

Can a business pay to appear higher?

No. Nothing a business buys from Visibility changes what an AI assistant answered, and nothing changes a published figure.