Short answer, as of September 2026: package GEO as fixed-scope tiers built around a defined prompt set and a fixed reporting cadence, never as open-ended hours — and price the explaining, because that is the line item that actually consumes the margin. The delivery work is repeatable. The conversations about why a number moved are not, and they are what turns a profitable retainer into a loss.
What to charge is a different question, and the market context for it is in how much GEO costs. This is about how to structure the offer so the number you charge survives contact with month three.
Why hourly fails here specifically
Most agency services have a natural stopping point. A page is written. An audit is delivered. A campaign runs for a month.
GEO has two properties that defeat hourly billing. First, the causal chain between work and outcome is long and noisy, so "why did this change" is a genuinely hard question that can absorb unlimited time. Second, the surface is unfamiliar to clients, which means every report generates questions that are really requests for education. Neither is billable in practice, and both scale with client anxiety rather than with scope.
The fix is not to refuse the explaining. It is to make it a defined, bounded deliverable — a scheduled call of a known length — rather than an unlimited entitlement that arrives by email at nine on a Tuesday.
Build the package around the prompt set
The prompt set is the natural unit of scope, because it determines almost everything else: the sampling cost, the reporting surface, the content backlog and what "improvement" even means.
Fixing it in the contract fixes the rest. "We track 30 prompts across 3 engines, sampled 3 times monthly" is a scope a client can understand and you can cost. "We improve your AI visibility" is not a scope at all.
It also gives you a clean change-order mechanism. A client who wants six more prompts, another location or a fourth engine is asking for more sampling volume, which has a real cost you can quote. Without a fixed set, that request arrives as an assumption rather than a purchase. The method for choosing the prompts in the first place is in how to build a prompt set worth tracking.
Three tiers that hold up
| Foundation | Managed | Programme | |
|---|---|---|---|
| Best for | Testing the channel | The standard retainer | Multi-location or competitive categories |
| Prompt set | 15-20, 2 engines | 30-40, 3 engines | 60+, 3-4 engines, per location |
| Sampling | Monthly | Monthly, weekly on watchlist | Weekly |
| Technical work | One-off access and render audit | Audit plus quarterly re-check | Continuous monitoring |
| Entity work | Audit and fix list | Fixes implemented | Implemented and monitored |
| Content | None; recommendations only | Fixed number of pieces or retrofits | Larger fixed allocation plus roadmap |
| Reporting | Written summary | Summary plus one scheduled call | Summary, call, quarterly review |
| Included Q&A | Email, next-report response | One call per cycle | Named contact, defined hours |
Three things about that shape matter more than the specific numbers.
The entry tier deliberately excludes content. Content is where the cost is unbounded and where results depend most on client cooperation. Selling an audit-and-measure tier first lets a sceptical client buy in cheaply, and gives you a baseline before you are accountable for moving it.
Everything countable is counted. Number of prompts, number of engines, number of pieces, number of calls. If a line cannot be counted, it will be argued about.
Reporting Q&A is in the package explicitly. This is the one most agencies leave out, and it is the one that bleeds.
Front-load the work that is genuinely one-off
A substantial part of GEO value is delivered once: crawler access, render checks, entity and profile consistency, structured data, the first baseline. It is real work, it produces real improvement, and it does not recur.
Bill it as an onboarding or setup fee rather than burying it in month one of a retainer. Two reasons. It protects your margin in the month when effort is highest. And it stops the client concluding, in month four, that they are paying a recurring fee for a one-off fix — which is the objection that kills renewals.
The sequence for that first month is set out in the 30-day agency playbook; packaging it as a distinct paid phase is what makes it survivable commercially.
The margin leaks, named
Unbounded reporting questions. Covered above: schedule it, bound it, include it.
Re-baselining on request. A client who asks you to add prompts mid-quarter is asking you to break comparability. Either refuse until the next cycle or charge for the re-baseline, and explain why — the reasoning is in benchmarks and share of voice.
Doing the client's implementation for free. Entity fixes often need access you do not have, to systems you do not control. When the client cannot act, agencies tend to absorb the chasing. Define whose job each fix is, in writing, at the start. The ownership question is worked through in who owns AI search.
Custom reports per client. The fastest way to lose a scaled service line is to let every client have a bespoke format. One template, white-labelled, is the whole point; what clients actually read is covered in white-label SEO reporting.
Tool cost per client, unmanaged. If your tooling bills per brand, per campaign or per check, a growing roster quietly erodes the margin you priced at five clients. Check how your stack scales before you sell the tenth retainer — DigiRank's Agency plan covers up to 15 client tenants and a monthly check allowance rather than billing per client, which is the shape that makes per-client margin predictable.
Say the timeline out loud, in the proposal
The single most effective margin protection is setting the expectation before the contract is signed, because an expectation set afterwards reads as an excuse.
Put it in writing: technical and entity fixes land in weeks; citation change is measured over months; the first review point is at 90 days. A client who agreed to that in the proposal does not send a panicked email in week five. A client who did not will, and answering it is unbilled. The realistic curve is in how long GEO takes, and the framing for the first conversation is in when a client first asks about AI search.
Three things never to put in the contract
A citation guarantee. You do not control the engines, their source selection or their release cadence. Guarantee the work and the measurement, never the outcome.
A specific number of citations by a date. Same reason, with an added trap: it incentivises optimising for prompts you can win rather than prompts that matter commercially.
Unlimited anything. Unlimited prompts, unlimited revisions, unlimited support. Each converts a known cost into an unknown one, and GEO has enough unknowns already.
What to sell instead of outcomes
Sell the capability and the visibility into it: a defined prompt set tracked on a stated cadence, dated answer history the client owns, a prioritised fix list, an agreed volume of content or retrofits, and a report somebody will actually read. That is a concrete, deliverable, defensible package.
It also happens to be honest. The client is buying systematic attention to a surface they currently cannot see, which is a real thing to sell — and considerably easier to renew than a promise about somebody else's algorithm. The features overview shows what the underlying tracking and reporting covers if you are assembling the delivery side of this.
Frequently asked questions
How should an agency package GEO services? As fixed-scope tiers built around a defined prompt set and a stated sampling cadence, with an onboarding fee for the one-off technical and entity work. Hourly billing fails because the explaining and the causal analysis expand without limit.
Why does hourly billing not work for AI search work? Two reasons. The link between work and outcome is long and noisy, so explaining a change can absorb unlimited time. And the surface is unfamiliar, so every report generates education requests that scale with client anxiety rather than with scope.
Should the entry tier include content production? Usually not. Content is where cost is least bounded and results depend most on client cooperation. An audit-and-measure entry tier lets a sceptical client buy in cheaply and gives you a baseline before you are accountable for moving it.
Should the setup work be billed separately from the retainer? Yes. Crawler access, render checks, entity fixes, structured data and the first baseline are genuinely one-off. Billing them as onboarding protects margin in the heaviest month and prevents the month-four objection that a recurring fee is buying a one-off fix.
What should never go into a GEO contract? Citation guarantees, a specific number of citations by a date, and anything described as unlimited. You do not control the engines, and outcome guarantees also push you towards optimising for winnable prompts rather than commercially important ones.
How do I handle a client who wants to add prompts mid-quarter? Treat it as a change order. Adding prompts breaks comparability with your baseline, so either hold the change until the next cycle or charge for the re-baseline and explain why the existing trend line cannot simply absorb it.
What is the biggest hidden cost in a GEO retainer? Unbounded reporting questions. Clients ask why a number moved, and the honest answer often takes real investigation. Put a scheduled call of a known length into the package so the explaining is a bounded deliverable rather than an open entitlement.
How do I stop tool costs eating the margin as I add clients? Check how your stack scales before selling the tenth retainer. Tooling that bills per brand, per campaign or per check erodes margin you priced at five clients, whereas a plan covering a set number of client tenants and a monthly check allowance keeps per-client cost predictable.