GEO Agency: What to Ask Before You Sign

What you are actually buying

A generative engine optimization engagement contains four kinds of work that differ enormously in cost, speed and how much anybody controls them, and a proposal that does not separate them is describing a category rather than a service. Getting that separation on paper before you sign is most of what distinguishes a good engagement from an expensive one.

The four are access and technical work, page rewriting, building genuine depth on a subject, and being referenced in places you do not own. The first is days and is nearly free. The second is weeks and is mostly approval time rather than writing time. The third is quarters and does not compress. The fourth has no timeline at all because it is a consequence rather than a task.

Most disappointment comes from a proposal that blends all four into one program with one schedule and one monthly fee. The fast work gets done, gets reported, and looks like progress. The slow work is where the result actually comes from, and it is still six months away when the contract is up for renewal.

The vocabulary makes this harder than it should be. GEO, AEO, LLM SEO and AI visibility all describe roughly the same activity, and which label an agency uses tells you about their marketing rather than their method. Our page on AEO against SEO sorts the terms out, and the same four tiers sit underneath whichever one is on the invoice.

The four questions that sort the market

These can all be asked in a first conversation and they separate a service from a vocabulary faster than any case study.

Which crawlers will you check, by name, and where. A good answer names specific clients and mentions checking both the site and the edge in front of it, because those are two separate places a request can be refused and only one is usually inspected. A vague answer about crawlability means this has not been thought about.

What baseline will you take before starting, and how. The right answer involves a fixed set of questions in buyer language, asked several times each, across more than one assistant, recorded before any work begins. An agency that does not take a baseline cannot demonstrate anything afterwards, and neither can you.

What will you report monthly, and what does a good month look like. If the answer is traffic, the engagement will end badly, because this channel does not reliably produce attributable visits. If the answer is presence frequency against a fixed question set, they understand the medium.

What are you not going to do. The most revealing question of the four. An agency that names limits, particularly around being referenced elsewhere, is describing real work. One that has no limits is selling a category.

The promises nobody can keep

Four claims appear in proposals and none of them can be delivered by anybody, regardless of skill or budget.

A guaranteed position. There is no position. A name is either in an answer or it is not, and the same question asked twice can differ. Anybody offering a ranking commitment here has borrowed a promise from a discipline where it made sense.

Submission or inclusion. There is no directory, no submission process and no relationship that places you in answers. Anybody offering it is selling something other than what they describe.

Attribution of revenue. Somebody who reads your name and then searches for you arrives as direct traffic with no trace of the assistant. Any promise to close that loop is describing something the plumbing does not support.

Fast results from the slow half. Depth and outside reference run on quarters. A proposal showing them landing in eight weeks has either misunderstood the work or is planning to report the fast half and call it the whole thing.

What to do before briefing anyone

Two hours of your own work changes the conversation completely and saves you paying to be told what you could have found out.

Read your own access rules, on the site and at the edge, and list which named clients are permitted. If something is refused, you have found the most common cause of total absence and it costs nothing to correct.

Read the first two sentences of your ten most commercially important pages. If they do not answer the question in the page title, you have found the highest-return editorial work available, and you now know it rather than being told it.

Take a rough baseline yourself. Six questions, asked five times each, recorded. Our walkthrough on tracking mentions by hand covers the method, and it takes an hour.

Arrive with those three and the conversation changes from a search into a scoping. You are asking about depth and outside reference, which are the parts you genuinely cannot do from inside, rather than paying somebody to discover a firewall rule.

How the money should be shaped

The shape of the fee tells you what an agency thinks it is selling, and there are three patterns with different failure modes.

A fixed monthly retainer with no defined deliverables suits the slow work and creates no pressure to show anything, which is honest about the timescale and dangerous if nobody is watching. It works when there is a named person on your side reading the monthly reading and asking questions.

Project pricing for the technical and editorial half is the cleanest arrangement available and it is what most businesses should buy first. The work is definable, it is fast, and you can tell whether it was done.

Performance pricing tied to visibility is the one to be careful with, because the metric can be constructed. If a bonus depends on a percentage, ask who chooses the questions and who can change them. A question set the agency controls is a metric the agency controls.

The arrangement that works most often is a defined project for the fast half, then a smaller retainer for depth once the fast half has proved the agency does what it says.

Whatever the shape, keep the question set and the readings on your side of the relationship. An engagement that ends with the agency holding the only record of what happened leaves you unable to brief the next one, and it removes the main thing you were buying, which was the ability to tell whether it worked.

When not to hire anybody

Three situations argue against an agency and they are common enough to be worth naming.

If you have never checked your own access rules, do that first. Paying somebody to find a setting you can read before lunch is the most expensive way to learn a cheap fact, and the finding will arrive with an invoice attached.

If your constraint is being referenced in places you do not own, an agency cannot deliver it as a unit of work. What you need is to do things worth writing about and be easy to reach when somebody wants to, and no retainer produces that.

If nobody internally will own the monthly reading, do not start. The work is unfalsifiable without it, and an engagement with no measurement becomes a matter of opinion at renewal, which the agency will win because they have more practice at that conversation than you do.

What a good first ninety days looks like

A well-run engagement has a fairly predictable shape, and knowing it lets you tell early whether you are in one.

The first two weeks should be almost entirely diagnostic. Access rules read and reported with the specific clients named, a baseline taken across a fixed question set, and an honest statement of which of the four tiers is actually binding for you. If month one produces a content calendar instead, the diagnosis was skipped and the plan is generic.

Weeks three to eight should be the fast half. Access corrected if needed, and the openings of your commercially important pages rewritten so they answer their own question immediately. This is where an early result comes from and it should be visible in the readings on assistants that cite their sources.

Weeks nine to twelve should start the slow half, which means committing to one subject and beginning the connected set of pages that covers it properly. One subject, not four. If this arrives as a broad calendar spanning many topics, the depth argument was heard and not understood.

By the end of ninety days you should have a baseline, a second and third reading against it, a list of corrected access rules, a set of rewritten openings and a subject chosen with reasons. What you should not expect is movement on the broad category question, and an agency promising that in the first quarter has set itself up to disappoint you.

Five things in a proposal worth pausing on

None of these are proof of anything on their own and all five are worth a direct question before signing.

No baseline in the scope. Without a reading taken before the work, nothing afterwards is demonstrable. If it is missing, ask for it to be added and watch how that request is received, because the answer tells you a great deal.

A deliverable count as the main unit. Twenty pieces a quarter is an output measure, and in this discipline output and outcome are only loosely related. Depth on one subject beats breadth across many, and a proposal priced per piece has an incentive pointing the wrong way.

No named crawlers anywhere in the document. The technical section should name specific clients and say where each was checked. Generic language about crawlability means this was written rather than done.

Traffic as the reported metric. This channel does not reliably produce attributable visits. An agency reporting traffic will either be judged unfairly or will be reporting something else and calling it this.

No limits stated. Real work has edges. An agency that cannot tell you what it will not do, particularly around being referenced in places you do not own, is selling a category rather than a service. Our page on how LLMs choose sources covers why that last tier is genuinely outside anybody's control.

The honest trade-off

The strongest argument for hiring is that the baseline and the depth judgment are genuinely hard from inside. A defensible baseline is repetitive work that only means something done consistently, and consistency is the first thing an internal team drops. Judging whether your own subject coverage is thin is close to impossible when you remember writing every page.

The strongest argument against is that the fast half is self-serviceable and it is where most of the early result comes from. Reading your access rules and relocating your own conclusions to the tops of your pages needs no supplier, and doing it yourself first means anything you buy afterwards is aimed at a real gap.

The version that works is buying the parts you cannot see from where you sit, after doing the parts you can. That sequence also changes the commercial relationship in your favor, because an agency briefed by somebody who has already read their own access rules and taken their own baseline is an agency being asked precise questions rather than one being handed a blank page.

Bottom line

Ask for the hours split across the four tiers. Access, rewriting, depth and outside reference differ by an order of magnitude in cost and speed, and a proposal that blends them will report the fast half and quietly defer the rest.

Insist on a baseline before any work starts, taken against a fixed question set you can see. It is the cheapest item in the engagement and the only one that cannot be produced later, which is exactly why it goes missing.

And do your own two hours first. Read your access rules, read your ten opening sentences, take a rough reading. You will brief better, pay less, and be able to tell whether what comes back is worth what it cost. That last part is the one that matters at renewal, when the conversation is about whether to continue and the only thing that settles it is a record you kept yourself.