How GEO agencies charge, and what each model does to the work

It is a fair thing for a buyer to want. Pay for results, not effort — the structure aligns everyone, and in classic SEO it is at least arguable, because rankings and clicks are reported by a source neither party controls.

AI search removes every one of those conditions. There is no rank to peg a bonus to: an assistant returns two or three names, not a numbered list, and being named is closer to binary than to positional. Most engines report no click data to you at all, so the traffic half of the attribution is missing rather than merely noisy. And the same prompt produces different answers between runs, which means any single measurement can be contested honestly by either side.

That leaves share of voice as the only candidate metric, and share of voice requires a defined universe: which prompts, which engines, which competitors. Whoever defines that universe controls the number. Handing that definition to the party being paid on it is not an alignment of incentives; it is an invitation, and the failure mode is not fraud but a slow, reasonable-seeming drift in what gets counted.

There is also a scope problem. Being named in an answer depends heavily on things an agency does not control — whether the business has anything citable, whether a competitor launches, whether an engine changes retrieval behaviour in a week. Paying on an outcome with that much exogenous variance means one party is gambling, and the price of the gamble ends up in the base fee anyway.

Most retainer disappointment traces to the definition of the deliverable rather than the price. Three definitions are common and only one of them is buying anything.

Hours. A number of hours per month, sometimes with a rate card. This buys capacity and makes the supplier indifferent to outcome, because a month of productive work and a month of meetings both bill identically. It is the honest structure for genuine staff augmentation and the wrong one for a discipline you are hiring expertise in.

Output volume. Four articles, ten pages, a set number of deliverables. This is the most common structure in search retainers and the most reliably wasteful: it guarantees production regardless of whether production is the constraint. A client whose actual problem is crawler access receives four articles a month about it.

A measured position against a frozen panel. The retainer buys movement on a defined set of prompts, reported per engine, with the baseline recorded before work starts. This is harder to write, harder to sell, and the only one of the three where you and the supplier are looking at the same scoreboard. Ask for it explicitly, because it is not the default.

More often than agencies suggest. A meaningful share of businesses asking why an assistant never names them have a bounded problem: a robots directive blocking AI crawlers, pages that render empty for a bot, a business identity scattered across records that disagree. That is two to six weeks of defined work with a clear end.

Selling a twelve-month retainer against that is not fraud, but it is answering a question nobody asked. The correct sequence is an audit, then a project against what it found, then a decision about ongoing work made with evidence rather than at the point of maximum uncertainty.

The genuine argument for a retainer is that measurement here takes weeks to read and corroboration compounds over months, so a project that ends at delivery ends before anyone can tell whether it worked. That argument is sound and it applies after the bounded problems are fixed, not instead of fixing them.

A reasonable compromise, and what we tend to recommend: buy the audit, buy the project, and make the retainer decision at the end of it with a baseline panel already running.

Find the deliverable definition. If it is expressed in hours or output counts, you are buying activity, whatever the covering letter says.

Find the measurement. If the only numbers named are Search Console metrics, the proposal is for SEO with AI language applied. Look specifically for a prompt panel, a baseline date, and named engines.

Find the exit. What happens to the panel, the reporting history and the documentation if you stop. If the answer is that they live in the supplier's platform, factor the switching cost into the price.

Find the boundary between delivery and coordination. Any agency describing itself as full service across paid, PR, video and CRO is delivering some of it and subcontracting the rest, which is fine and should be stated. Ours is on the service pages.

Find what the first month actually contains. Diagnosis and a baseline, or production starting immediately. Production in week one means nobody checked whether the engines can reach you.