Sell a defined research and improvement process, with clear limits on guarantees.
Set the service boundary
Agree on the client’s product, markets, audiences, question groups, and business objective. State which surfaces you will observe, how often, and which deliverables are included. A vague promise to ‘make AI recommend you’ cannot be turned into an accountable work plan.
Keep technical fixes, content production, research, and outreach separately scoped. Each has different owners and dependencies. Identify what requires client access, product verification, legal review, or an independent publisher’s cooperation.
Maintain a client-approved baseline
Build the prompt set from customer evidence and review it with the client. Record versions and collection conditions. Separate branded validation from unbranded discovery so the baseline does not flatter the business by design.
Provide definitions for visibility, share of voice, position, and citation measures. Include sample sizes and collection failures. Client comparisons should not combine different prompt panels into a league table without explaining those differences.
Make the report an action document
Show a small number of material findings with representative answers, cited pages, and a proposed owner. Distinguish factual corrections, information gaps, technical problems, and product limitations. A source leaderboard alone leaves the client wondering what to do next.
Keep a work register: evidence, hypothesis, action, owner, change date, verification, and status. Use it to explain what your team completed and what the result supports. Avoid claiming credit for every favorable chart movement after the engagement began.
Connect to commercial evidence carefully
Track identifiable referrals and qualified outcomes under the client’s existing analytics and CRM rules. Keep estimated influence separate from observed acquisition. Document the attribution window and cost boundary when discussing performance.
End each review with a concrete decision: continue the test, change the information, investigate access, or gather better customer evidence. That makes the service useful even when the answer engines are variable and a visibility chart is inconclusive.
