How do you evaluate an AI search or GEO agency?

Until roughly 2024, hiring an SEO agency was a relatively well-understood transaction: technical audit, content roadmap, link building, rank tracking. The arrival of AI answer engines. ChatGPT search, Perplexity, Gemini, Google's AI Overviews. Has fragmented the optimization target. A brand can rank #1 organically and still be absent from the AI answer that 40% of high-intent users now see first. Conversely, a brand can be cited in an LLM answer and capture intent without ever showing in a traditional SERP.

This has spawned a new sub-category of agency variously labeled 'GEO' (Generative Engine Optimization), 'AEO' (Answer Engine Optimization), or 'AI search.' Some are credible technical SEO firms that have extended their practice. Others are 2024 vintage startups with strong marketing and thin operating depth. The buying decision is now about whether the agency can navigate both surfaces. And whether they can prove it with citation-share measurement, not just rankings.

Every credible AI search engagement starts with a technical SEO baseline because the same crawlability, indexability, structured data, and entity clarity that drive Google rankings also drive LLM retrieval and citation. Agencies that skip the technical layer and lead with 'GEO content' are usually selling a thin product wrapped in new vocabulary.

Ask any AI search agency to walk through their technical audit framework. The strong ones cover Core Web Vitals, render-blocking, JavaScript SEO, internal linking topology, schema coverage and validity, robots and sitemap hygiene, and crawl budget. They also cover the newer surface: llms.txt, AI bot user-agent allowances, structured data that maps to entity types LLMs surface, and content architecture that supports extraction.

Traditional rank tracking measures position on Google for a defined keyword set. AI citation measurement is harder: LLM answers are stochastic, queries are conversational and long-tail, and there's no canonical SERP to scrape. Strong agencies have built or licensed citation-tracking infrastructure. They monitor brand mentions in ChatGPT, Perplexity, Gemini, and Claude across a defined query set, measure share-of-voice against competitors, and tie citation gains to content and structural changes.

Weak agencies wave at this problem with screenshots. The diligence question is straightforward: ask to see a recent citation-share report for a real client, the methodology behind it, and the query set construction. If the answer is anecdotal, the agency is not actually operating in the AI search layer.

LLMs preferentially cite content that is structurally clear, entity-rich, and from sources they have already learned to trust. This means a credible AI search engagement is also a content program. Original research, structured explainers, definitive resources, and authoritative pages on the entities (people, products, companies, concepts) that anchor your category.

The agencies worth hiring publish content production cadence, editor depth, and quality controls. They build entity hubs and internal linking structures that signal authority. They measure both traffic outcomes and pipeline outcomes (assisted conversions, brand search lift, AI-attributed sessions where measurable). The agencies to avoid promise quick wins, focus on keyword stuffing dressed up as GEO, or treat content as an afterthought to a technical audit.