How to Choose a GEO Agency

Suddenly, everyone is a GEO agency. Scroll through any agency’s services page and you’ll find “AI Search” or “Generative Engine Optimisation” sitting somewhere near the top, often added in the last few months. Some of that is genuine expertise, but a lot of it is SEO relabelled, with no real change in what’s actually being delivered.

The trouble is, there’s no shared definition yet, no industry standard, no accreditation body checking anyone’s work. So how do you actually tell the difference between an agency that understands this properly and one that’s just moved fast on positioning?

When we’re talking to clients, these are the sorts of questions we’d talk through. GEO is moving fast enough that anyone offering total certainty should probably be the first thing that gives you pause.

Ask what they’re actually optimising for

“AI” isn’t one channel. Optimising for “AI” without specifying which platform is a bit like optimising for “social media” without distinguishing between LinkedIn and TikTok. ChatGPT, Claude, Gemini, Perplexity and Google’s AI Overviews are distinct environments, with different retrieval mechanisms, different training data cutoffs, and different rules about what they will and won’t crawl.

A credible agency should be able to name the specific platforms they’re working across and explain, at least at a high level, how the approach differs between them. If the answer stays at the level of “we optimise for AI,” that’s worth pressing on.

Ask how they measure success, and be wary of a single tidy score

This is probably the single most important question on this list, because it’s the easiest one to get away with answering badly.

Research from Rand Fishkin and Patrick O’Donnell found less than a 1-in-100 chance that two runs of the same AI prompt return the same brands, and less than a 1-in-1,000 chance they return them in the same order. With that in mind, a single sampled AI response tells you almost nothing on its own – and that’s not a minor caveat, it’s the central challenge that anyone trying to measure AI visibility right now should be focused on.

Any agency reporting a single “AI visibility score” without acknowledging that volatility is either not aware of the problem or not being upfront about it. What you want to hear instead is an agency that tracks multiple signals over time, treats individual responses as directional samples rather than absolutes, and can explain the limitations of their own reporting as readily as its strengths. We do this ourselves through Ebb, our AI visibility tool, and we’d expect any agency worth hiring to have an equivalent answer.

Ask for evidence of real citation wins, not just theory

“We understand GEO / AI search / AEO” (or whatever you want to call it!) is easy to say. Proof looks different with named clients, specific platforms, specific outcomes. A citation rate that moved. An AI Overview slot that was won and can be shown. A share-of-voice gain against a named competitor.

When we rebuilt PureGym’s content strategy around AI Overviews, the result was concrete and their blog now appears in over 7,000 Google AI Overviews, roughly 10 times more than their nearest competitor. When we worked on Remitly’s Immigration Index, the campaign page went on to hold the number one AI Overview slot worldwide for high-intent terms like “immigration index.” That’s the kind of specific, checkable evidence worth asking for, from us or from anyone else. If an agency can only offer general claims about capability rather than named, verifiable results, that’s a gap worth noting.

Ask for a definition of fan-out queries, and the difference between ranking and being selected

This is a good litmus test question, because it filters out agencies that have relabelled their SEO offering without genuinely engaging with how AI platforms work.

When an LLM answers a prompt, it doesn’t just search the original phrase. It breaks it into several sub-queries, fan-out queries and searches those instead. A good GEO agency should be able to explain this mechanism to you, unprompted and describe how they identify and target these queries rather than just the obvious keyword.

It’s also worth asking about the difference between ranking and being selected. A page can rank well for a fan-out query and still never get cited, because the title and snippet didn’t clearly signal relevance to what the AI was trying to answer. If an agency’s explanation stops at “we help you rank,” rather than addressing what happens after that, they may not have gone deep enough into how citation actually works.

Ask how GEO fits with their existing SEO and content work

We are not starting from scratch and neither should any credible GEO strategy. Technical SEO fundamentals, structured data, clean semantic HTML, entity mapping, crawlability, remain essential for AI discoverability. The pages that rank well in organic search are overwhelmingly the pages that get cited in AI responses too, because many AI systems still ground their answers in live web searches that hit the same Google and Bing indexes SEO has always targeted.

What GEO adds is additional, not a replacement, so things like prompt-aware content structure, self-contained sections that work in isolation, and a focus on topical comprehensiveness over keyword targeting. If an agency is pitching GEO as a totally separate, standalone service, disconnected from your existing organic performance, that’s worth questioning. The best foundation for AI visibility is strong SEO, strong content and strong digital PR. AI Search and GEO sit as a layer on top, not a replacement underneath.

Ask about their own visibility, not just their clients’

A slightly cheeky question, but a genuinely revealing one – ask the agency to show you their own AI search visibility. If they’re advising you on how to be found and recommended by AI platforms, they should be able to demonstrate they’ve done the same work on themselves.

We’ve been through this exercise ourselves. When we audited our own AI search presence, we found real gaps, weak visibility on cold discovery prompts, missing citations on the exact terms we wanted to own. We’ve spent the past few months closing those gaps and we’re still not finished. If an agency claiming GEO expertise can’t show you an equivalent process, or worse, hasn’t looked at their own visibility at all, that tells you something.

The honest answer is that some of this is still unsettled

GEO is moving faster than any confident framework can fully capture. The only honest response is to keep testing, resist publishing what’s convenient rather than what’s grounded in evidence, and treat confident-sounding claims that aren’t backed by data with real scepticism, regardless of how impressive the vendor dashboard looks.

That’s the standard we’re trying to hold ourselves to. It’s a reasonable one to expect from anyone you’re considering working with.

If you want to talk through what any of this means for your brand specifically, get in touch. You can also read more about our approach to AI here.