If you deliver an AI search service, you need client work that connects observed visibility with useful content and publication research. An AI search retainer needs work beyond measurement. Reporting that a client appears in four of ten answers does not describe what you did for the client.
Recurring work includes researching cited sources, assessing placements and improving content.
The deliverable is the cited sources, not the score
For any question a client wants to appear in, the answers cite a set of pages. Those pages are demonstrably being used to answer that client's buyers.
Use the list to plan the month's research. For each page, check whether the client appears, whether the description is current, and whether the publisher accepts submissions or corrections.
A roundup listing competitors and not your client is a placement with a known audience. An out-of-date description is a correction request. Both are specific enough to put on an invoice.
The diagnosis that decides the quarter
Nearly every client finding falls into one of four categories, and they need different work.
| Finding | Work |
|---|---|
| Absent while competitors are named | Placement work from the cited sources. The largest job. |
| Named but described wrongly or out of date | Find the source page and get it corrected. Fastest win. |
| Present without persona context, absent with it | Content work on audience fit. Commonly missed. |
| Cited but not named | The client is a source, not a candidate. Fix what the cited pages say they sell. |
The third row only appears if you track personas, which is the argument for setting them up properly at onboarding.
What to promise and what not to
Nobody controls the assistants. There is usually no position to hold, answers vary between runs, and the source selection mechanism is not observable from outside.
So do not promise placement in answers. Promise the work: the corrections sent, the placements pursued, the pages revised, and a measurement of presence across a fixed prompt set over time.
Set that expectation in the first review rather than the fourth. A client shown a rising line every month for six months will not handle a flat one.
Running it across clients
Each client is a campaign holding their business description, brands, competitors and personas. Tracked items are account-level, so the collection allowance pools across the roster and one prompt can serve two clients in the same market.
Staff access is not charged per seat.
The engine limitation to check before selling
Prompts go to ChatGPT, Gemini and Google AI Mode. Claude and Perplexity are not available and cannot be added.
If a client's reporting requires either, say so before the contract rather than after. Several competing products cover both, and a service commitment you cannot meet is worse than a narrower one you can.
What clients will still buy elsewhere
Site auditing, content writing, outreach sending and email verification. This produces the research and the decisions; the execution tools remain separate.
When planning the next client period, specify the sources to investigate and the work each finding supports.
See the work sequence, reporting it, or roster capacity and cost.