In-house or agency for GEO: an honest comparison

On the hire: the salary is the visible part and rarely the majority of the cost. Add employer contributions and benefits, the tooling the person will reasonably ask for, the management time of whoever they report to, recruitment cost, and a ramp of one to three months during which output is real but low. Add the opportunity cost of the search itself, which in a market where AI-search skills carry a premium can run months.

The less countable hidden cost is ceiling. A single hire brings exactly the knowledge they have, and the organisation's capability becomes that person's last job plus whatever they read. Without a peer to argue with, wrong approaches persist for quarters because nothing contradicts them.

On the agency: the retainer is the visible part. Add your own team's time in calls and reviews, which is routinely underestimated. Add the cost of the work not done in months where attention went elsewhere, which you will not see itemised. Add switching cost if the panel and documentation live in their systems rather than yours.

The uncountable one here is distance. An agency without access to your sales calls is building a prompt panel from assumptions, and will measure the wrong questions with complete rigour. That failure is invisible in reporting, because the reporting is accurate about the wrong thing.

Hearing the business. The single highest-value input to AI search work is the vocabulary buyers actually use, and it lives in sales call recordings, support tickets and the questions that come up on every first call. Someone sitting near those absorbs it continuously. An agency has to be given it deliberately, and most clients do not.

Small changes at speed. A person inside the company with access to the CMS makes a fix in an hour that becomes a two-week ticket through an agency, a scoping call and a deployment window.

Institutional trust. Getting engineering to prioritise a rendering fix, or getting legal to approve claim language, is a relationship problem more than a technical one, and relationships are held by people who are present.

Caring on a Tuesday in November. An employee whose performance review depends on this will chase the unglamorous thing. An agency has other clients, and the honest version of that sentence is that your quietest month is somebody else's busiest.

Pattern recognition across accounts. Most of the diagnostic value in this work is having seen the failure before — knowing that a particular symptom usually means a rendering problem rather than a content problem saves a quarter. That instinct comes from volume, and a first in-house hire has one or two previous employers to reason from.

The unglamorous technical layer. Log analysis, entity reconciliation across dozens of third-party records, redirect archaeology. This work is tedious, intermittent and needs doing properly a few times a year, which is a poor fit for a permanent role and a good fit for a supplier who does it constantly.

Being fireable. This is underrated. Ending an agency relationship that is not working takes a notice period. Ending an employment relationship that is not working takes months, a process, and a cost, and the knowledge that it is hard makes organisations tolerate underperformance far longer.

Not being a single point of failure. When the in-house specialist resigns, the capability leaves the same day, usually along with the only understanding of why things were set up as they were.

After trying a pure version of either, most businesses over a certain size converge on roughly the same split, and it follows the strengths above fairly precisely.

In-house owns the inputs and the relationships: the prompt panel's contents drawn from real sales conversations, the content that requires knowing the business, the internal politics of getting a fix prioritised, and ownership of whether any of it is working.

The supplier owns the specialist and intermittent layer: technical diagnosis, entity architecture, measurement infrastructure, and the pattern knowledge that only comes from seeing many accounts. This is the work that is genuinely hard to justify as a permanent role because it is needed intensely and then not for months.

The arrangement that fails is the inverse, which is also the most common: an agency writing content from assumptions while an in-house generalist is nominally responsible for technical work they have not been trained in. Both parties are doing the half the other is better at, and everyone is busy.

Is your bottleneck knowledge or capacity?

If nobody in the business knows what should be done, hiring one person to find out is a large bet on one hiring decision, made at the moment you are least equipped to assess candidates in the discipline. Buying the knowledge first, then hiring against a plan you understand, is the lower-variance sequence.

If you already know what needs doing and there is simply nobody to do it, hire. An agency is an expensive way to buy hands, and you will spend a meaningful share of the retainer explaining context an employee would already hold.

One caveat we would give against our own interest: if you can hire someone genuinely senior in this discipline, do it. A strong in-house lead beats any agency arrangement, because they get everything the agency gets plus the access an agency never will. The reason the hybrid is common is that such people are rare and expensive, not that the model is theoretically superior.