The agency playbook for AI search

Every agency we know is having the same client conversation this year. It starts with “rankings look fine but leads are down,” detours through some uncomfortable silence, and ends with somebody asking what the agency’s “AI strategy” is. If your answer is a slide about content quality, this guide is for you. It’s the operating model we run: how to scope, deliver, measure, and report GEO work without overpromising on engines you don’t control.

The honest pitch

Here’s the framing that works, because it’s true: AI engines are already answering questions about your client’s category, right now, today, citing somebody. The only question is whether anyone is watching who. You’re not selling magic influence over ChatGPT. You’re selling measurement of a channel the client can’t see, plus the content and entity work that demonstrably moves it. Measurement first. The moment you show a client the citation matrix for their own buyer questions, the rest of the engagement sells itself.

We’ve found the demo matters more than the proposal. Pull up a live citation check in the room. No deck survives contact with a competitor being named four times in a row.

Scoping the work

A GEO engagement has four recurring parts. Budget them separately:

• The query portfolio. 30–60 real buyer questions per client, tagged by intent and priority. Revisit quarterly. This is strategy work and it deserves senior hours; a junior picking keywords from a tool will fill the portfolio with questions nobody asks an assistant.
• The measurement loop. Weekly automated checks across the five engines, logged so you can compute trend lines. Costs cents per query through SERP APIs. Never skip a week; gaps in the data become gaps in the story you tell the client.
• The build queue. Pages, rewrites, schema, freshness passes, and off-site corroboration, prioritized by the competitor gap analysis. This is where most of the hours go.
• The narrative. Monthly reporting that a CMO can forward to a CEO without translating. More on this below, and in the reporting guide.

Setting expectations you can survive

The engines are non-deterministic. Ask the same question twice and the citations can differ. If you let a client believe citations work like rankings, stable, inspectable, attributable, you will spend every QBR explaining variance instead of progress. Set three expectations in the kickoff:

1. We report citation rate over repeated checks, not single results. A query is won when it cites you most weeks, not once.
2. Engines move at different speeds. Perplexity can pick up a new page in days; AI Overviews follow Google’s index and take longer. We’ll see leading indicators on fast engines first.
3. Some queries won’t be winnable at sane cost, because the engines have settled on Wikipedia or a major publication. We’ll tell you which ones and stop spending there.

Delivery rhythm

Weekly: checks run, log updated, anything anomalous flagged internally. Nobody emails the client about one bad week. Monthly: the trend report, the wins and losses called out by name, and the next month’s build queue with expected impact. Quarterly: portfolio review, the “stop list” of unwinnable queries, and the citation-share chart against the top three competitors, which is the single most renewal-correlated artifact we produce.

The team shape

You don’t need new headcount, you need reassigned attention. The SEO lead owns the query portfolio and the gap analysis. Content takes the build queue, with a new brief format: answer first, numbers, sources, one question per page. Whoever does digital PR adds corroboration targets, the directories, review sites, and publications engines keep citing in your client’s category. The tooling does the checking; humans decide what to do about it. That last sentence is roughly our product philosophy, and it’s why Crescendo ships the matrix, the gaps, and the report draft, but never pretends the strategy writes itself.