Lesson 6 of 8

What to ask an agency

How do I tell a good agency from a plausible one?

Ask questions whose answers are checkable. The most useful ones are not technical: what will you do in the first ninety days, what will you show me at the end of it, what would tell us this is not working, and who specifically does the work. A good agency answers these easily, because they have thought about them. The questions are diagnostic precisely because they are ordinary.

Most guidance here assumes you are being conned. Usually you are not. The more common situation is that both sides are sincere and the expectations were never made explicit, which produces the same disappointment as bad faith and is much easier to prevent.

So the point of these questions is not to catch anyone. It is to get the implicit agreement into the open early, while changing course is still cheap.

What will you do in the first ninety days, specifically. You are listening for named deliverables on a timeline, not a description of a methodology. A good answer sounds like a plan. A vaguer answer is often honest, and the follow-up is simply: what would you need to know to make it specific, and how quickly can you find that out.

What will you show me at the end of it, and what would tell us this is not working. The second half matters more than the first. Every engagement should have a condition that would make a reasonable person stop, agreed while everyone is optimistic. An agency that has one is telling you it has been wrong before and noticed.

Who does the work, and who do I talk to when something is wrong. Named people. This is not a trick question. Work gets delegated, which is normal and fine. You are establishing who to call, not catching anyone out.

What do you need from us. Search work usually needs access, approvals and subject knowledge from the client, and engagements far more often stall on that than on anything the agency did. An agency that names its dependencies up front is describing how the work actually goes.

The subsection on claims that cannot be true. Rare, and worth recognising. A guarantee of a specific position on a specific term: nobody controls the ranking, so nobody can promise the position. A promise of results in days: the mechanism is slower than that, and if something does move in days it is usually paid traffic, which is fine but is a different product. A promise of a specific number of AI citations: lesson 3 measured why that is not a stable thing to promise. None of these mean bad faith on its own, but each one is worth pausing on and asking how they would achieve it.

Five questions, and what you are listening for

The questions are ordinary on purpose. What makes them diagnostic is the shape of the answer, not the question.

Ask thisYou are listening for
What will you do in the first ninety days, specificallyNamed deliverables on a timeline, not a description of a methodology.
What will you show me at the end of itA specific artefact, and a date you will see it.
What would tell us, in three months, that this is not workingAny falsifiable condition at all. If nothing could indicate failure, no evidence can indicate success.
Who does the work, and who do I talk to when something is wrongNames. Delegation is normal; you are establishing who to call.
What do you need from usAccess, approvals, subject knowledge. Engagements stall here more often than anywhere else.
A vaguer answer is usually honesty rather than evasion. The useful follow-up is what would you need to know to make it specific.

Takeaway: Agree up front what failure would look like.

All eight lessons · The one-page field guide