Generative engine optimization agencies: how to choose one
The questions that separate genuine GEO practice from vocabulary, plus the contract terms worth negotiating before you sign.
GEO agencies are a young category with thin credentials and unverifiable case studies. This article gives a practical way to evaluate one.
What competent agencies do differently
Three observable differences separate genuine practitioners from traditional SEO agencies that renamed a service line.
They measure generated answers systematically. There is a method: a prompt set, repeated runs, stored responses, per-engine reporting, and competitor tracking on identical prompts. If this capability is absent, the offering is SEO with new vocabulary.
They treat retrievability as its own discipline. AI crawler access, server-side rendering, and edge bot protection are not part of a standard SEO audit. An agency that does not raise them unprompted has not encountered the failure mode.
They work off-site. A large share of what models believe about a brand comes from third-party sources, so agencies confined to your domain will move one half of the problem.
Questions that reveal competence
- How many times do you run each prompt, and how do you handle variance between runs? A vendor who has not encountered variance has not measured seriously.
- Which engines do you cover, and do you use APIs or the consumer interfaces? Both are defensible; not knowing the difference is not.
- Show me a report where the numbers went down. Agencies that only present improvements are selecting their cases.
- What would you check before writing any content for us? The right answer starts with retrievability and existing page structure.
- What did you believe a year ago that you no longer believe? This field moves fast enough that no revised opinions means no attention paid.
- Who writes, and what is the fact-checking process? Unreviewed AI output published at volume is a genuine risk to your domain.
- What happens to the prompt set and data if we leave?
Contract terms worth negotiating
- Ownership and export of prompt sets, response history, briefs, and content — the durable assets of the engagement.
- A documented, frozen baseline within the first month.
- Per-engine reporting including the competitor set, not a single composite.
- A review point at three months, which is long enough to see retrieval-side movement and early enough to change course.
- Clarity on whether content production is included and at what volume.
Warning signs
- Guaranteed AI answer placement, guaranteed rankings, or a guaranteed score.
- Proprietary metrics with undisclosed methodology used as the primary success measure.
- Large content volumes committed before any measurement exists.
- Case studies quoting percentage improvements with no absolute numbers and no prompt set.
- Reluctance to show the raw responses behind reported metrics.
Agency, freelancer, or in-house
Freelance specialists are often the best value for the diagnosis and measurement setup, which is a bounded project. Agencies earn their premium on sustained off-site work and content production capacity. In-house ownership is the right end state for measurement, because the prompt set is a strategic asset that should not live with a supplier.
A sensible sequence for most companies is a bounded diagnostic engagement first, then a decision about whether the ongoing work is better placed inside or outside.
Common questions about GEO agencies
- Are GEO agencies worth it? For the initial diagnosis, usually yes — experience across many accounts shortens the learning curve considerably. For ongoing measurement, less so, because that is tooling and process you can own. For off-site work, it depends entirely on their outreach capacity.
- How do I verify a case study? Ask for the prompt set, the absolute numbers rather than percentages, and the engines measured. Percentage improvements with no denominator are unverifiable and common.
- What size agency should I look for? Size matters less than whether they have a documented measurement method. A two-person specialist with a rigorous process beats a large agency that added a service line last quarter.
- Should the agency own the prompt set? No. It is a strategic asset that should live with you, and portability of prompts, response history, and briefs belongs in the contract.
- What is a reasonable first engagement? A bounded diagnostic of four to six weeks — question research, baseline, retrievability audit, prioritised gap list. You keep the deliverable regardless of whether the relationship continues.
- How do I know when to bring it in-house? Once the measurement is running steadily and the backlog is understood, the ongoing work is mostly content and outreach. That is the natural handover point.
- Should the GEO agency also handle my technical SEO? At minimum they need access to whoever does, because retrievability fixes — rendering, crawler rules, edge bot protection — sit with engineering rather than marketing. Engagements stall most often at exactly this handoff, so agree the route to a deployed change before starting.
- What reporting cadence is reasonable? Monthly, with the underlying measurement running weekly. Weekly reporting on a noisy metric generates meetings about variance rather than decisions.
Tip: before any pitch, run your five most valuable buying questions through two engines and keep the answers. Agencies that engage with what those answers actually show are the shortlist; those that pivot straight to a content proposal are not.
Still stuck? We typically reply within 1 business day.
Contact support