Ask which answer, for which audience.
“AI visibility” is too broad to be a useful brief. Ask which customer questions matter, which platforms the client means and which products or services should be associated with those questions.
Start with a short, agreed prompt set. The aim is to create a repeatable observation method, not a score that pretends to represent every possible answer.
Keep the foundations in view.
Google’s guidance says that AI Overviews and AI Mode do not require special optimisation beyond established search fundamentals. That statement applies to Google’s features; it is not a universal promise about every AI platform.
For an agency, the practical response is to review whether priority pages are accessible, useful and clear about the business and its expertise. Check whether factual claims have support and whether the content actually addresses the question being asked.
Sell a defined review, not a guaranteed mention.
A useful scope can include a review of answer-led content, inconsistent business information and technical access, followed by a set of recommended changes. Content production and implementation need their own allowances.
Avoid a promise such as “we will get you cited by every major model.” You cannot control the answer a platform generates, which sources it chooses or how those behaviours change. A delivery commitment should describe the work you can complete.
Write down how you measured.
For each observation, record the prompt, platform, date and visible response. Distinguish a brand mention from a source link. Keep the method consistent enough that a later review has useful context.
Answers can vary. A small prompt set is a sample, and a movement in that sample does not establish why a platform changed its response. Pair the observations with the changes made on the site and any relevant business data.
Give the client language they can repeat.
A straightforward explanation is: “We will improve the clarity and usefulness of the pages that answer your customers’ questions, check the relevant technical foundations, and monitor an agreed set of prompts. We cannot guarantee inclusion.”
That sets an understandable expectation while leaving room for a useful service. The work is concrete even when the platform outcome remains uncertain.