AEO for Agencies: How to Track AI Visibility Across Many Client Brands Without Drowning

Ten clients, five engines and twenty questions each is a thousand AI answers a week. Nobody reads a thousand answers. An agency that tracks AI visibility well is not one that collects more; it is one that has decided in advance what a monthly deliverable looks like and set the tracking up to produce it. Here is the structure that works.

One workspace per brand, one metric set for all of them

Each client is its own workspace: its own questions, its own competitors, its own schedule and its own history. Nothing is shared between clients except the definitions — what "mentioned" means, how position is scored, how the confidence range is computed. That is what makes brands comparable to each other and to themselves over time, and it is what lets an account manager say "this client moved and that one did not" without qualifying every sentence.

Resist per-client customisation of the scoring. A client who asks to count branded questions towards the headline number is asking for a better-looking report, and the next month's report will be worse for it.

Twenty questions, chosen once, changed rarely

The question set is the measurement instrument. Choose the twenty questions a buyer types before they know the client's name, agree them with the client in writing, and then leave them alone for at least a quarter. Every change to the set resets the trend. If the client wants to add questions, add them as a second group and report it separately.

Mark branded and near-brand questions and exclude them from the headline. A brand whose name is its category ("Pet Insurance Co") will be surfaced by the engines for spelling reasons on any question containing the category, and the number will flatter everyone.

Per engine, never blended

Engines disagree far more than clients expect. One will name the client consistently and another never. The blended score hides this, and worse, it hides which engine's sources to work on. Report the mention rate and citation rate per engine, and let the client see the row of chips.

Competitors: named, not discovered monthly

Share of voice is only meaningful against a fixed set. Name three to five competitors per client at the start — the ones the engines actually name, which the first check will tell you — and hold the set. Adding a competitor changes every share-of-voice figure retroactively, so do it deliberately and note it in the report.

The deliverable

The monthly report has four sections and fits on two pages:

1. Where we stand. Mention rate and citation rate per engine, non-branded questions only, with the confidence range and the change from last month. A change inside the range is reported as "no significant change", in those words.
2. Who is winning and where. Share of voice against the named competitors, and the two or three questions where the gap is largest.
3. What the engines are reading. The ten most-retrieved sources for the client's questions, which of them name a competitor and not the client, and which the agency worked on this month.
4. What we are doing next. Three actions, each tied to a source or a question, with an expected timeline of weeks.

Everything else — the full answers, the per-question drill-down, the crawler logs — is available in the tool for the client who wants it and is not in the report.

Alerts that do not cry wolf

AI answers vary between runs. An alert on every score change trains the account team to ignore alerts. Gate them on the confidence range: alert only when the new score's range no longer overlaps the old one. That is the difference between a tool the team checks and a tool the team mutes.

What AEOTrack does and does not do for agencies

Business gives ten websites, ten seats and Slack, Zapier and webhook alerts, and every workspace shares the same scoring, branded/near-brand detection and confidence ranges. What it does not have yet is white-label PDF and client-facing shareable links — the report above is exported per workspace and assembled by you. If that is the blocker, say so; it is the most-requested agency feature and it is on the list.