AI visibility examples: five businesses, prompts and fixes

AI visibility becomes useful when it is tied to a question a customer would genuinely ask. The examples below show the prompt, the evidence to inspect and a measurable next step. They are patterns, not promises: the same edit can produce different results across models, markets and dates.

Example 1: a local service business

A plumbing company in the Netherlands tests "emergency plumber in Utrecht open on Sunday." It appears for general Utrecht questions but not when Sunday availability is required. The website has no explicit service hours, while two directories show conflicting information.

• Evidence to inspect: service-area pages, stated opening hours, local business profiles and recent reviews.
• Useful action: publish accurate emergency hours and service boundaries, then reconcile profiles the business controls.
• Measurement: rerun the same location-and-time prompt and verify both the mention and the stated availability.

Example 2: a B2B SaaS product

A SaaS vendor is mentioned for broad category searches but disappears when a buyer asks for a specific CRM integration and EU data residency. The product supports both, but the facts are buried in a PDF and an old partner page lists the integration as beta.

• Evidence to inspect: integration documentation, security pages, partner listings and dated release notes.
• Useful action: create crawlable, versioned pages that state current compatibility and boundaries in plain language.
• Measurement: track recommendation rate for prompts that include each hard requirement, plus factual accuracy in the response.

Example 3: an ecommerce brand

An online shop is cited for "waterproof commuter backpacks," but an assistant repeats an expired price. The product page, shopping feed and an affiliate review disagree. More mentions would amplify the wrong fact, so accuracy is the first priority.

• Evidence to inspect: visible product data, Product schema, feeds, availability and the pages the answer cites.
• Useful action: correct controlled sources and use consistent current price and stock information.
• Measurement: record correct-price rate and product eligibility, not merely whether the domain was cited.

Example 4: a professional-services firm

A consultancy publishes strong articles but is rarely recommended for regulated projects. Its website makes broad "industry-leading" claims without naming qualifications, jurisdictions or representative experience. Independent evidence is thin.

• Evidence to inspect: named credentials, professional registers, speaking or editorial coverage, and clearly scoped case studies.
• Useful action: replace vague claims with verifiable facts and earn legitimate third-party coverage. Do not manufacture reviews or citations.
• Measurement: factual description quality and qualified mention rate for the jurisdictions the firm can actually serve.

Example 5: an international company

A supplier is visible in its home country but omitted from global prompts because answer engines assume it serves only that market. The site uses one language, hides country availability in checkout, and has no country-level support information.

• Evidence to inspect: shipping or service countries, languages, currencies, support hours and local legal entities.
• Useful action: make country eligibility explicit on relevant pages and keep regional facts consistent.
• Measurement: use a prompt matrix by country and language. Do not merge results into one global score that hides market gaps.

How to turn an example into a reliable experiment

1. Save the exact prompt, model, search mode, country context and date.
2. Record the complete answer, citations, competitors and factual errors.
3. Change the smallest set of sources needed to correct the observed gap.
4. Allow time for sources to be crawled or retrieved, then rerun the same test.
5. Compare multiple runs. Generated answers vary, so a single favorable response is not proof of a trend.