The claim: “We will get your company into AI answers”
Being mentioned in one answer can be useful evidence. It does not, by itself, establish that relevant buyers regularly discover the business through AI search. The question supplied to the system is part of the test, not an incidental detail to crop out of the screenshot. Google: AI features and websites.
A prompt such as “Explain why Example Supplier is a good choice” already supplies the brand and a favorable frame. A question such as “Which suppliers can handle this specification in this region?” tests a different discovery problem. A report should not describe those observations as interchangeable.
The reality: mention, citation and referral are different outcomes
A mention is the appearance of a name. A citation is an identified source or link under the measurement definition being used. A referral is a visit attributed to an AI service in the available analytics. The presence of one does not guarantee the other two.
For Google AI features, Google documents the continuing relevance of normal SEO practices. That supports work on clear, accessible, useful information. It does not establish that a provider can control the contents of an answer, and it should not be generalized into a guarantee covering every AI platform.
Write the test before observing the answer
Create a fixed set of buyer questions covering the offered services, decision criteria and service area. Separate discovery questions from branded research and comparison questions. Preserve the exact wording and identify exclusions before running the test.
For each observation, retain the platform, date, visible model or mode where disclosed, location context and whether web search was used or available. If a setting is unknown, record it as unknown. Save the answer and actual citation destinations. Counting an answer that merely resembles an old answer is not source verification.
An original prompt-selection example
Consider a hypothetical 30-question review. Ten questions explicitly name the business; twenty ask unbranded supplier-selection questions. The brand appears in all ten branded answers and in two unbranded answers.
The combined mention rate is 12/30, or 40%. The unbranded discovery rate in this defined sample is 2/20, or 10%. Both calculations are arithmetically correct. Reporting only the larger number would omit the distinction most relevant to new-customer discovery. Neither percentage estimates the share of all AI users who see the business.
Repeat the same set on a documented schedule and preserve answers that do not mention the company. Changing to easier prompts after a weak result makes the trend difficult to interpret. Repeated observations reveal variability; they do not make a provider's influence causal by themselves.
What evidence would strengthen the claim?
Ask for persistent visibility across relevant unbranded questions, accurate descriptions of the offer and source links that resolve to useful pages. Review attributable visits where the platform and analytics expose them. Compare those visits with meaningful inquiry activity without assuming every dark or unattributed visit came from AI.
SharkBlade's AI Visibility category should be interpreted within its configured evidence scope. A score can summarize that scope; it is not a market-wide share-of-voice census or a promise of inclusion. Missing platform coverage should remain a limitation in the report.
Questions before purchasing
- Are the prompts branded, unbranded or a disclosed mixture?
- Who chose them, and can I see the entire set?
- What counts as a mention and what counts as a citation?
- Can I inspect answers with no favorable result?
- How are changing answers and unavailable observations handled?
- What would count as a useful business outcome beyond the chart?
ZeroDark's test is whether an unfamiliar, relevant buyer can find accurate, useful information about the business. An AI screenshot earns its place in the evidence; it should not be allowed to stand in for the entire buying journey.
Supporting documentation
Google: AI features and websitesNIST: AI Risk Management Framework