Data and interpretation
Does it measure answers in buyer-facing AI apps or responses from developer model APIs?
Buyer-facing apps let people ask questions and read answers. Developer model APIs let software request responses from a model service.
Check the measured experience and collection method before comparing results.
Request the saved answer, question, date, market, and AI app behind a reported result.
Check how the implementation distinguishes the customer's business from another business with a similar name.
Keep unresolved identity and failed analysis separate from a confirmed absence.
Check what the platform interprets and what the agent must interpret.
Analysis can identify the business, examine cited sources, and show which other businesses appear and how answers compare them.
Check which analysis and history the implementation actually provides.
Supported work and permissions
Ask the agent to complete the chosen task through the documented interfaces.
Record which data it can read, which actions it can perform, and where it needs another connected tool.
Record read access and write access separately.
Check that the connection reaches the intended account and project.
Check the allowed actions and any requests that cost money or change data.
If access expires or a tool fails, the result must identify the unfinished work instead of claiming completion.
Outcomes and limits
Separate a business mention from a citation. A source link does not necessarily mean that the answer presents the business as an option.
Separate a mention from a recommendation. An answer can name a business while criticizing it or recommending something else.
Compare equivalent questions, AI apps, markets, and counting rules over a stated period.
Treat unavailable observations as unavailable, rather than zero.
Keep visits, accounts, and payments separate from answer visibility. Record attribution limits instead of inventing a customer result.
A useful evaluation states what the agent completed, what failed, and what remains unknown.
The evaluation describes tested work and observed results. It cannot promise future recommendations.