AI visibility testing

Test what AI systems preserve, omit or invent about your travel brand.

A traveller needs a step-free entrance. An answer that overlooks two entrance steps can change their decision about a visit.

AI Visibility Testing records how a system answers a real question about your place, route or organization. Compare the answer with identified sources, examine what the reader can do next, and decide what needs attention.

Start with a question your audience actually asks

Choose a question from visitor enquiries, guide conversations, search research or practical planning. Begin where an inaccurate or incomplete answer could affect someone’s choice.

A destination might examine a question about visiting without a car. An operator might check how a journey’s difficulty is described. A cultural organization might review access information or reservation conditions.

Be clear about what you are testing. An open question about suitable places can show whether your organization appears among the options in that observation. A question naming your organization examines how it is represented. Record those findings separately.

Choose the systems and languages relevant to your audience. Start with a scope you can review carefully.

A timber entrance has two stone steps, while a separate optical representation of the same entrance depicts a level approach.
Fictional comparison: the representation omits the entrance steps. A plausible-looking answer or image must still be compared with its identified source; this is not a recorded model test.
AI-generated illustration · Tamaga

Keep the question, answer and testing conditions

Retain the exact question and complete answer, including any citations and destination links.

Alongside them, record:

  • The date, system or product, and model details where available.
  • The language, search or browsing mode, and relevant conversation context.
  • Location or account conditions that could affect the response, where known.
  • The source pages and versions used for comparison.

Mark unavailable details as unknown. If you supply a page directly, record that: the observation shows how the system uses supplied material, rather than whether it would find that page independently.

Note separately whether the organization was mentioned, whether a source was linked, and whether the answer preserved the information needed for the decision.

Compare the answer with what the source establishes

Consider this fictional cultural-site example. The response below is invented to explain the comparison; it is not a captured model result.

Compare the answer with what the source establishes
Material Wording
Published entrance information “The main entrance has two steps.”
Illustrative AI answer “The entrance is step-free.”
Finding The answer introduces an unsupported access claim. The source does not establish whether another entrance offers step-free access.

A careful answer would retain the two steps and identify what remains unresolved. It could direct the visitor to detailed access information or an enquiry about an alternative entrance.

Compare consequential statements individually. Check dates, places, services and conditions before calling two statements contradictory. A short answer can omit detail without becoming misleading; the question is whether the omission changes what the reader can reasonably conclude.

Use Source Confidence when the supporting material itself needs assessment.

Make the correction that the evidence supports

First inspect the maintained source. Is the relevant information accurate, explicit and easy to find? Does a heading, summary, translation or other public version weaken its conditions?

If the source needs correction, update it and review the versions you control. If the source already answers the question adequately, retain it and document the discrepancy in the observed response.

Where the structured representation disagrees with the visible page, review it using the Schema.org approach. A problem in one AI answer does not, by itself, establish a structured-data problem.

Keep the proposed action specific: clarify an entrance description, correct an outdated period, repair a destination link or investigate an unsupported statement.

Check the reader’s next step

When the answer provides a link, follow it. Does it reach the relevant explanation? Can the reader find the condition, assess suitability and take an appropriate next step?

For the fictional cultural site, that continuation might be detailed access information and a way to ask about an alternative entrance. Other questions may lead directly to a published timetable, a suitable journey or a reservation page.

Review your own channel versions too. A newsletter introduction or video caption should retain the condition that matters and lead to the maintained explanation.

Repeat the observation and retain the difference

After a correction, repeat the documented question under comparable conditions. Start a fresh conversation where appropriate, and record any differences in the system, mode, context or source version.

Keep both answers. A later accurate response is useful evidence of that observation; it does not establish what caused the change or guarantee future answers.

Where referral visits or relevant enquiries can be identified, review them separately. A mention, a citation and a useful visitor action each tell you something different.

The resulting record can stay short: the question, observed answer, supporting material, finding, action and next check.

For a separate fixed source, the fictional late-arrival citation experiment tests whether an answer retains advance notice after 23:00 without inventing an approval requirement. It reports no measured visibility result.

Bring one question and the answer you received

Share the question, the response and the public pages that should support it. We can examine what was preserved, what changed and which correction or further check would be useful.