How to Choose an AI Search Agency

Knowing how to choose an AI search agency has become a practical problem for marketing teams, because many SEO providers now offer some version of GEO, AEO or AI visibility work and the proposals are hard to compare. The terminology changes from one pitch to the next and the evidence offered is often a screenshot of a single answer from ChatGPT. The tests below help separate a provider that can measure and improve your visibility from one that has added a new label to an old service. They apply to a full programme of AI search services for B2B organisations or a smaller piece of work.

Each section covers one test, starting with what to decide internally and finishing with the contract. Working through them in order tends to shorten the shortlist quickly, because the weaker proposals struggle to explain how results are measured once you ask the same questions of every agency.

Decide What AI Search Visibility Has to Achieve Before You Shortlist

Before contacting any agency, decide which questions your buyers ask AI tools and what you want to happen when they do. Visibility in AI answers is only useful if it reaches the people who buy from you, so the starting point is a list of the questions that come up during their research, not a list of platforms.

That list becomes the brief. It lets every agency on the shortlist propose against the same prompts, the same competitors and the same measures of success, which makes their answers comparable.

• The questions prospects ask when they research a supplier, a product or a problem you solve
• The competitors you expect to be compared with in those answers
• The pages on your website that should be cited when those questions come up
• How enquiries are currently tracked, so any change can be connected to commercial results
• Who inside the organisation approves content changes and technical work

Agencies that receive a clear brief can be judged on how they respond to it. An agency that ignores the brief and leads with its own scoring tool is showing you how the reporting will feel once the contract is signed.

Understand What an AI Search Agency Does

An AI search agency works on how often your organisation is mentioned, described accurately and cited as a source in answers from AI platforms. The work overlaps heavily with SEO, because a page has to be found, crawled and understood before it can be cited as a source, but the measurement and much of the content planning are different.

The labels vary between providers. GEO usually refers to making content easy for generative tools to select and cite, AEO to formatting pages as direct answers to questions and AI SEO or AIO to the whole discipline. Our guide to generative engine optimisation explains the first of these in more detail. None of the labels matter as much as what the agency will do each month.

Paid placement inside AI assistants is a separate discipline with its own skills and contracts. If advertising is part of the plan, the questions in choosing a ChatGPT ads agency apply alongside these, while what to expect from an AI SEO agency covers the organic work in a B2B setting.

Test Every Claim of AI Visibility Results

Treat every claim of AI visibility results as something to reproduce yourself before you rely on it. A screenshot of a client named in a ChatGPT answer proves that the answer appeared once for one person on one day. It says nothing about whether it still appears or whether the agency caused it.

A screenshot captures one response at one moment and nothing in it shows whether the same prompt gives the same answer to anyone else. A case study that holds up will name the prompts, the platforms, the dates and the starting position, so you can run the same prompts and see whether the result is still there.

Branded prompts deserve particular care. An answer that names a company when the prompt already contains its name shows very little, so look for results on the unbranded questions a buyer asks before they know who you are.

The agency should also be able to explain why a result happened. A credible answer points to specific changes, such as a page rewritten to answer a question directly or a crawler that had been blocked, rather than to a proprietary method it cannot describe.

Ask How Visibility Is Measured and Reported

Good AI search reporting measures the same fixed set of prompts on the same platforms at regular intervals, then connects what it finds to traffic and enquiries. Ask each agency to show a real report with the client details removed, because the format tells you more than any description of it.

The report should separate a brand mention from a citation, since being named in an answer and being linked as a source are different outcomes. It should also show which competitors appear when you do not, how answers describe your organisation and whether any of those descriptions are wrong. Proprietary visibility scores are fine as a summary, provided the agency can explain which prompts, platforms and dates sit behind the number.

Google’s own reporting covers part of the picture. Search Console’s guidance on how clicks, impressions and position are counted explains that a click on a link in an AI Overview counts as a click and that every link in the overview shares a single position. Priority Pixels measures AI Share of Voice as a companion metric to traditional rankings, because AI SEO sits alongside organic SEO, not as a replacement for it.

Reporting frequency matters less than consistency. A monthly report that runs the same prompts in the same way is more useful than a weekly one that changes its method whenever the numbers look poor.

Check Which Platforms Are Tracked

A capable agency tracks the platforms your buyers use and tells you exactly which ones are in the report. ChatGPT, Perplexity, Gemini, Copilot and Google’s AI Overviews and AI Mode each select sources in their own way, so a result on one platform says little about the others.

Priority Pixels monitors ChatGPT, Google Gemini, Microsoft Copilot and Perplexity as standard, with monitoring run on a monthly cycle. The LLM performance tracking and analytics service page sets out what that tracking covers.

Whichever agency you choose, ask it to justify the platform list against your audience. A provider that tracks one platform because its tool only supports that one is choosing the measurement to suit itself rather than you.

Look for Technical Foundations Before Tactics

An agency worth hiring checks the technical foundations first, because AI platforms cannot cite pages they cannot reach or read. Google’s guidance on AI features and your website says a page must be indexed and eligible to be shown with a snippet to appear as a supporting link. The same guidance states that there are no additional technical requirements.

Other platforms run their own crawlers. OpenAI’s crawler documentation explains that websites opted out of its search crawler will not be shown in ChatGPT search answers, while its training crawler is controlled separately. Perplexity’s crawler documentation recommends allowing PerplexityBot in robots.txt and notes that a web application firewall may need to allow its requests explicitly.

A good proposal therefore starts with robots.txt, firewall and CDN rules, indexing, page speed and content that loads as text rather than only through scripts. Structured data should match the visible content and structured data errors are worth fixing before anyone talks about new markup. Much of this is ordinary technical SEO, which is why an agency that skips it is building on weak ground.

Ask the agency what it would check in the first month and what it expects to find. A specific answer about crawler access, indexing and templates suggests experience, while a promise of special AI markup suggests a product looking for a buyer.

Judge the Content Approach

The right content approach improves the pages buyers already rely on, rather than producing large volumes of new pages written only for machines. Pages that answer specific questions clearly and accurately serve readers first, so the work usually starts with rewriting key service and product pages before anything new is commissioned.

Ask to see a before and after example of a page the agency has rewritten. The useful changes are easy to spot, such as a direct answer at the start of each section, facts that can be checked, named authors where expertise matters and descriptions of your services that match what you sell.

Be cautious of plans built around publishing hundreds of thin pages for every prompt variation. That approach adds pages that compete with each other and dilutes the clear answers your key pages give. It also leaves your team maintaining content that nobody reads.

Accuracy deserves its own place in the plan. If AI tools describe your organisation wrongly, the agency should trace where the wrong information comes from, correct it on your own website and identify the external sources that repeat it.

Spot the Red Flags Early

The clearest red flags are promises of guaranteed citations, vague methods and heavy use of new jargon without evidence behind it. Google’s guidance on hiring an SEO warns that no one can guarantee a number one ranking on Google and suggests checking whether advice on AEO and GEO is aligned with Google’s official guidance on generative AI features.

A related warning sign is a proposal that cannot name the people doing the work. AI search work touches content, development and analytics, so ask who carries out each part and how much of it is done by people rather than automated tools.

Watch as well for a pitch that dismisses SEO entirely. The technical points above show how much AI visibility still depends on ordinary search work, so an agency that treats AI search as a replacement for it is setting expectations it cannot meet.

Agree Contract Terms That Protect Your Accounts and Data

The contract should keep your accounts, data and content under your control and make the agency’s work measurable against the brief. Settle these points before work starts, because they are much harder to agree once the relationship is under strain.

Access is the first point. Search Console’s guidance on owners, users and permissions explains that owners have full control while full and restricted users have view rights, so your organisation should hold ownership and give the agency a user role. Analytics should work the same way, with the agency added through user access management rather than owning the property.

The remaining terms decide whether the arrangement can be judged fairly and ended cleanly. Each one is easy to agree at the start and difficult to add once the work is under way.

• The agreed prompt set, platforms and reporting dates, written into the statement of work
• Ownership of all content, reports and prompt data produced during the contract
• A record of every change made to the website, with the reason for each one
• A notice period that allows a handover without losing reporting history
• Clear separation between organic AI search work and any paid advertising

The right agency will be comfortable with every one of these tests, because a well run AI search programme has nothing to hide at the shortlist stage. One that resists sharing its prompts, avoids naming its platforms or wants to own your accounts is already telling you how the relationship is likely to go.