Answer Engine Optimisation
How to Choose an AI Search Agency
An AI search agency works on a narrow, newly valuable problem: when a buyer asks ChatGPT, Claude, Gemini or Google's AI Overviews to recommend a provider, which companies get named. This page explains what those agencies actually do, how the work differs from SEO, what it costs, and the questions worth asking before you hire one.
What an AI search agency actually does
The category goes by several names — AI search, answer engine optimisation (AEO), generative engine optimisation (GEO) — and the good agencies all converge on similar work, because the levers are the same. An assistant's answer is assembled from what it retrieved and what it absorbed during training. You cannot edit the model. You can change what it finds.
- Prompt-set benchmarking. Define the questions your buyers actually type, then measure which brands each assistant names today. Without this you have no baseline and no way to prove movement.
- Source auditing. Assistants cite their sources. Pull the citations behind your category's answers and you have a target list — usually review sites, industry directories, comparison posts and a handful of publications.
- Entity and factual hygiene. Consistent, machine-readable facts about the company: structured data, an accurate profile everywhere it appears, no contradictions between your site and third-party listings.
- Citable content production. Original data, benchmarks and comparisons — the formats assistants quote — rather than the brand-voice marketing copy that never gets cited.
- Placement work. Getting into the specific listicles, directories and roundups that keep showing up in the citations.
- Monitoring. Re-running the prompt set on a schedule, because answers drift as models update.
How it differs from SEO
The two overlap, and any agency claiming AI search has nothing to do with SEO is selling novelty. But the differences are real enough that the tactics diverge.
| Traditional SEO | AI search | |
|---|---|---|
| What you win | A position in a list of links | A mention inside the answer |
| Runner-up value | Position 10 still earns clicks | Not being named earns nothing |
| Main levers | Your own pages, links, technical health | Third-party sources the model retrieves and cites |
| Measurement | Rank tracking, impressions, clicks | Prompt testing across assistants; share of mentions |
| Feedback speed | Days to weeks | Weeks to months, and answers drift on model updates |
| Attribution | Referrer data in analytics | Mostly invisible — you have to ask buyers how they found you |
That last row is why most companies discover the channel by accident. A customer mentions offhand that an assistant recommended you, and it turns out a channel has been running unmanaged for months. Google's own documentation on AI features in Search is a reasonable starting point for how AI Overviews select and cite sources, and Google's helpful content guidance still describes the substance assistants tend to quote.
How to evaluate an AI search agency
This is a young category with an unusually wide quality spread. A handful of questions separate the operators from the repackagers.
- "Show me a before-and-after prompt test." Real practitioners have screenshots of assistant answers from before and after their work. If the case study is a traffic chart, they are selling you SEO with a new label.
- "Which assistants do you measure, and how often?" ChatGPT, Claude, Gemini, Perplexity and AI Overviews behave differently. One-assistant reporting is a partial picture.
- "What are the citations in my category today?" A serious agency will have looked before the first call and can name the sources shaping answers about you.
- "What do you produce, and what do you only advise on?" Strategy decks are cheap. Earning placements and producing citable data is the expensive, effective part.
- "What happens when a model update wipes out the gain?" It will happen. The answer you want describes monitoring and re-work, not a guarantee.
One red flag worth naming plainly: nobody can guarantee a placement inside a specific model's output. Anyone who does either misunderstands how the systems work or is counting on you not to check.
What it costs
Pricing has not settled, but the shape is becoming clear. Advisory-only engagements — audit, prompt benchmark, a roadmap you execute — tend to run in the low thousands per month. Full-service work that includes producing citable assets and chasing placements runs higher, often into low five figures. Project-based audits are common as a first step and a sensible way to find out whether the channel matters for your category before committing.
Expect the first meaningful read in one to two quarters. Sources have to be published, indexed and retrieved before an assistant's answer changes, and that chain is slower than a ranking change.
How the category is searched in other markets
Demand for this work is not English-first, which surprises people. Buyers in Germany, the Netherlands, Italy and Spain search for it in their own languages — KI-Suche Agentur and AI Search Beratung in German, bureau voor generatief zoeken and AI vindbaarheid in Dutch, agenzia AI search in Italian, agencia AI search in Spanish. The German queries in particular tend to be long and specific, often describing the exact situation: a company noticed a customer arrived via an assistant recommendation and wants to turn that into a repeatable channel.
If you are an agency doing this work, that is a useful signal about where the least-contested demand sits.
Frequently asked questions
What is an AI search agency?
An AI search agency works to make a brand appear inside the answers that AI assistants give. When someone asks ChatGPT, Claude, Gemini or Google's AI Overviews to recommend a provider, the assistant names a handful of companies. An AI search agency's job is to make sure yours is one of them. The discipline is often called answer engine optimisation (AEO) or generative engine optimisation (GEO).
How is AI search optimisation different from SEO?
SEO competes for a position in a list of links a person then chooses from. AI search competes to be the answer itself, where there is no list and usually no click. Ranking tenth on Google still earns some traffic; being the eleventh brand an assistant might mention earns nothing. The work also draws on different sources: assistants lean heavily on third-party write-ups, comparisons, directories and structured data rather than on a brand's own marketing pages.
Can you actually influence what an AI assistant says?
Partly, and not by trickery. Assistants ground their answers in what they can retrieve and what they absorbed in training, which means the levers are the sources: independent coverage, up-to-date directory and profile entries, comparison pages, original data worth citing, and consistent, machine-readable facts about your business. Any agency promising a guaranteed placement inside a specific model's output is overselling.
How is it measured?
By prompt testing rather than rank tracking. A credible agency defines a set of buyer prompts, runs them across several assistants on a schedule, and records whether you were named, in what position, how you were described, and which sources were cited. That last column is the useful one, because it tells you exactly which pages to go and influence.
What does it cost?
Most engagements land between a few thousand and low five figures per month, depending on whether the agency is producing citable content and earning placements or only auditing and advising. Treat a quote under about $2,000 a month with suspicion unless the scope is genuinely narrow, and expect a first meaningful read on results in roughly one to two quarters.
Is it worth doing yet?
It depends on whether your buyers are already asking assistants for recommendations. The honest test is to ask your last ten customers how they found you. If any of them say an AI assistant suggested you, the channel already exists for your category and is currently unmanaged. If none do, keep an eye on it rather than budgeting for it.
Find a provider
Our AI search agency listings collect the providers doing this work specifically, and the wider AI agency directory lets you filter by service and location, and Get Matched puts a shortlist in front of you if you would rather describe the problem and let us do the filtering. If you are an agency doing this work, you can list your agency. For related reading, how to get AI clients covers the demand side, browse by service shows the adjacent categories, and the glossary defines the terms.