How to Hire a GEO Agency That Actually Earns Citations
A buyer-side playbook for hiring a GEO agency, covering the technical deliverables, AI citation KPIs, and pricing models worth paying for in 2026.

The costly mistake is signing a GEO retainer that still measures itself in keyword rankings and organic sessions, then paying it for a full year before anyone notices it never moved a single AI citation. At the Ahrefs survey average of $3,209 per month, that is nearly $38,000 against the wrong scoreboard.
A procurement document prevents it. Pin citation-level KPIs to the SOW, bind the deliverables that produce them, and pick a fee structure where the agency carries the right slice of the risk.
The three artifacts that follow belong in every generative engine optimization contract before any signature. Headline pricing is the last thing to negotiate, not the first.
Citation KPIs That Belong in the SOW
Three metrics have to appear in writing before any scope conversation begins.
- Share of model voice. For a defined prompt set, the share of answers that cite your domain versus competitors. Define the prompt set in the contract itself, with volume, engine coverage, and refresh cadence.
- Cited URLs. The specific pages the model surfaced. A rising share of model voice that only pulls one page is a different problem than coverage spread across the site, and the SOW should treat them as separate reporting lines.
- Answer presence by engine. ChatGPT, Perplexity, Gemini, Google AI Mode, and Claude return different results for the same query, so citation lift has to be reported per engine, not pooled into a single number.
The per-engine split matters because generative engines cite differently. ChatGPT and Perplexity both lean on product and service pages for a large share of their citations, while Gemini and Claude pull more from editorial and reference content. An agency that only reports aggregate AI referral traffic hides which surface is actually working, and hiding that per-engine split is where the fee loses its link to results.
Deliverables Worth Paying For
A GEO agency's monthly scope should produce measurable artifacts, not activity logs.
- Entity and prompt audit. A list of the questions that drive demand in your category, mapped to the entities your site is (and is not) associated with inside each model.
- Answer-engine landing surfaces. Pages built or rewritten to carry the direct answer, with the supporting evidence, source attribution, and internal links that make a citation easier to produce.
- Off-site citation building. Mentions and links on the pages the models already trust: Reddit threads, regulatory sites, trade publications, legitimate review surfaces. Watch the quality floor here; many of the shortcuts an AEO agency might hide show up in this line.
- Prompt and engine reporting. A weekly or biweekly dashboard that reads the KPIs above, exported in a format you can keep after the contract ends.
Schema is often sold inside this scope. Include it only if the agency will describe its real role honestly: an Ahrefs study of 1,885 pages that added JSON-LD markup between August 2025 and March 2026 found no meaningful AI citation lift, with Google AI Overviews citations down 4.6% and ChatGPT and AI Mode changes statistically indistinguishable from zero. Treat schema as crawl hygiene, not a growth lever, and price it accordingly.
Pricing Models and Who Carries the Risk
Three structures appear in GEO retainers today, each distributing risk differently.
- Fixed retainer. Predictable, but the fee runs whether or not citations move. It remains the default: 78.2% of providers in the Ahrefs sample charge a monthly retainer at an average hourly rate of $111. Pair it with a termination clause tied to a KPI floor, not a notice period alone.
- Performance. Fee pegged to citation lift: share of model voice gains, newly cited URLs, or pipeline traced to AI referral traffic. The buyer holds the advantage, but the metric needs forensic clarity so the agency cannot inflate it with low-volume prompt sets or cherry-picked engines.
- Hybrid. A base retainer covering research, content production, and reporting, with a bonus or step-up pegged to citation KPIs. This is the structure most procurement and legal teams can defend, and the one that survives engine volatility without renegotiation.
The line chart below shows the ambient click compression every GEO retainer is negotiating against. The slope is why the KPI language has to name citations, not click-through from traditional SERPs.
Where the task list looks closer to legacy SEO than to generative work, step back and compare it against how SEO and GEO duties actually differ before agreeing to any model. Agencies that priced traditional retainers for a decade often repackage the same scope with a GEO label and the same hourly rates.
Instrumentation the Agency Must Own
Before the ink dries, verify three operational details.
- Prompt set size and refresh. 50 prompts tested monthly is directional; 500 tested weekly across five engines is diagnostic. The contract should state the number and the cadence.
- Engine coverage. A vendor that tracks only ChatGPT is behind the surface area you are paying them to influence. Require AI Mode, Perplexity, Claude, and Gemini in the baseline report.
- Site-side observability. INP, crawl stats, and log-level fetch records belong in the agency's workflow, not bolted on later. Google defines a good INP as 200 ms or less at the 75th percentile; AI crawlers fail silently when pages stall above that and the citation never happens.
Our Claude citation tracking guide covers the per-engine telemetry worth owning in-house even when the agency delivers the dashboard. Ownership of the data is bargaining power at renewal, so make exports and schema part of the SOW rather than a favor to ask later.
Picking the Model That Fits the Buy
Choose a fixed retainer when GEO is new on the roadmap, the category is well-indexed, and the internal team is ready to hold the agency to a KPI floor. Choose performance when citations themselves are the product being purchased, usually in a mature category with crisp attribution and a prompt set broad enough to resist gaming. Choose hybrid when the category is moving faster than any twelve-month contract can keep up with, since the base covers the work while the bonus captures the upside.
Beyond the model, walk from any agency that cannot state, in writing, the prompts it will test, the engines it will measure, and the pages it expects to get cited. For a fuller picture of what to ask on the way in, which AI search services earn their fee is the companion piece. Those three answers, and the KPI definitions above them, are what the retainer is actually buying.