Running AEO for Ten Clients at Once

Agency AEO advice tends to stop at positioning: how to sell the service, how to price it, why clients need it. That is the easy half. The hard half is what happens in month three, when ten clients are live and the work that was pleasant for one brand has become a queue.

The useful question is which parts of this discipline generalise across a client base and which do not — because the ones that do not are where an agency’s margin quietly goes.

What generalises well

The technical audit. Can crawlers reach the site, does content render server-side, is robots.txt accidentally excluding the search crawlers — the convention is specified in RFC 9309, and misreading it is a recurring finding — is Organization markup present and correct. This is a checklist, it is the same checklist for every client, and it is the single most common cause of a client being invisible. It is also the fastest visible win, which matters commercially in month one. Run it first, every time.

Reporting structure. Presence, position quality, share against named competitors, trend. The metrics are identical across clients; only the numbers differ. Build the report once.

The measurement discipline. How many samples before a change is real, what counts as a citation, why week-over-week deltas are noise. These are properties of the systems, not of the client, and getting them wrong is a house-wide error rather than a per-client one. How many runs before you trust an AI visibility number is the arithmetic; it applies unchanged to every account.

Explaining the limits. The conversation about what AEO cannot do — no guaranteed placements, no editing a model, no revenue attribution — is the same conversation every time. Write it down once and use it in every kickoff. It prevents the most expensive category of client relationship failure, which is a client who expected something the mechanism cannot deliver.

What does not generalise

The prompt set. This is the one that consumes the hours and it resists templating almost completely. A prompt set has to reflect how that client’s buyers actually phrase questions, which means it comes from their sales calls, their support tickets, their Search Console queries, and their competitive position. You can template the shape — fit questions, comparison questions, pricing questions, objection questions — and the contents are bespoke.

Agencies routinely try to solve this with a category template (“best {category} for {segment}”) and it produces prompt sets that are technically valid and commercially useless, because they track the head terms rather than the questions that precede a purchase. Choosing tracked prompts covers the selection.

Diagnosis. Two clients can present with identical symptoms — absent from AI answers despite ranking on Google — and have entirely different causes. One has an entity collision with a similarly-named company; the other has all its substance behind a JavaScript render; the third has clear content and no third-party presence at all. The symptom is the same and the remedies share nothing. This is why the ranked-but-not-cited analysis is worth running per client rather than assuming: it eliminates most causes at once. When Google ranks you and AI doesn’t works through the quadrants.

Anything requiring the client to act. Third-party presence, original data, review-site standing, press. The agency can identify and can sometimes execute, but the substance usually lives with the client, and client responsiveness varies by an order of magnitude across a book of business. This is where timelines actually slip.

The three operational failures

Reporting cadence outrunning the mechanism. Monthly retainers imply monthly reporting, and AEO does not produce monthly-legible movement. Grounded answers respond in days to weeks, third-party consensus over months, training-derived knowledge across model generations. A monthly report showing a three-point change is reporting sampling noise, and doing that for twelve months teaches a client that the service does nothing.

The fix is structural rather than cosmetic: report activity and leading indicators monthly, and outcome quarterly, and say so at kickoff. How long does AEO take is the version to send them.

Scan budget concentrated in the wrong accounts. Every scanned prompt costs money across every engine. Ten clients with fifty prompts each is five hundred prompts per cycle, and the natural drift is that the loudest client gets the deepest coverage rather than the one whose contract value justifies it. This wants a deliberate allocation, revisited, not an accumulation.

Nobody reads the answers. The single highest-value activity in this discipline is reading what engines actually said, and it is the first thing to be dropped when a queue forms, because it is unstructured work with no completion state. A dashboard can be checked in ninety seconds; twenty answers take half an hour. Teams under load check the dashboard, and the dashboard cannot tell you why.

If you protect one hour per client per month, protect that one.

The model-update problem is a portfolio problem

A worthwhile agency-specific advantage, and one in-house teams do not have.

When a model updates, it can change what an engine says about an entire category overnight — with no action by anyone, and no notice. For a single brand this is indistinguishable from “something we did went wrong,” and teams burn weeks looking for a cause that does not exist.

Across ten clients you can tell the difference immediately. A shift affecting most of your book on the same date is systemic. A shift affecting one client is theirs. That is genuinely useful diagnostic capability, it comes free with the portfolio, and it requires only that you look across accounts rather than within them. Model updates are the new algorithm updates covers living with them.

The related discipline: record which model produced which result. A March number and a September number from different model versions are two different instruments, and comparing them is not a trend.

What to charge for

The uncomfortable commercial note. The technical audit is fast, templated and produces the most visible early result — which makes it the easiest thing to sell and the worst thing to build a retainer on, because it finishes. The durable work is the slow compounding kind: third-party presence, original data, entity clarity, sustained monitoring. That is harder to sell and it is the actual service.

An agency that sells the audit as the engagement is selling a project. One that sells the audit as month one of a programme, with the limits stated honestly up front, is selling something that can survive a year. Selling AEO without overclaiming covers the scoping conversation.

The counter-argument

The fair objection: this describes a level of per-client bespoke work that does not scale, and agencies survive on leverage. If prompt sets and diagnosis genuinely cannot be templated, the economics only work at premium pricing.

Largely right, and it is worth accepting rather than arguing around. AEO is closer to consulting than to a productised service, at least for now. The leverage that does exist is real but narrower than agencies would like: the technical audit, the reporting layer, the measurement standards, the client-education material, and the cross-portfolio view of model changes. Those genuinely amortise.

What does not amortise is the thinking. An agency pricing this as a productised deliverable will either lose money or ship template prompt sets that do not reflect any client’s actual buyers — and the second failure is worse, because it is invisible for two quarters.

Where to start

1. Standardise the audit and the report. One checklist, one template, every client.
2. Build prompt sets from client sales calls, not from a category template.
3. Allocate scan budget explicitly rather than letting it drift to whoever asks loudest.
4. Look across the portfolio monthly for shifts hitting multiple clients on the same date.
5. Set the cadence expectation at kickoff: activity monthly, outcome quarterly.

AEO for agencies covers the service shape, and AEO team workflow has the internal-process version that maps onto a pod structure.