How to Choose an AI SEO Company: 12 Questions to Ask Before You Sign

Every SEO agency now calls itself an AI SEO company. Most of them added the phrase to a service page sometime in the last two years and changed nothing else about how they work. That is not a criticism of the industry so much as a warning to you: the label on the proposal tells you almost nothing about what you are actually buying.

This matters more than a normal vendor decision. An SEO retainer is usually a six to twelve month commitment, paid monthly, with results that lag the invoice by weeks or months. Pick wrong and you do not find out for a quarter, by which point you have paid for three months of work you cannot see, and you have to start the search over with less budget and less patience than you started with.

This guide is a practical checklist, not a sales pitch for any one agency. Ask a shortlisted agency all twelve questions on a single call. Their answers, more than their pitch deck, will tell you whether you are hiring a specialist or renting a keyword.

What "AI SEO" actually means, before you evaluate anyone

Strip away the marketing and AI SEO is two things stitched together. The first is answer engine optimisation, sometimes called AEO: the practice of getting a brand named and linked when someone asks ChatGPT, Perplexity, Gemini or Google AI Overviews a question. The second is generative engine optimisation, or GEO: structuring content, schema and entity signals so a language model can extract and quote it accurately. Neither replaces traditional SEO. Both depend on it, because an assistant that cannot crawl your site, parse your data or trust your entity has nothing to cite in the first place.

A competent AI visibility service treats classic SEO and AI citation work as one system, not two separate line items. If an agency pitches AI SEO as a bolt-on module sitting on top of a weak technical foundation, question one below will expose that fast.

The 12 questions to ask before you sign

1. Do you work with our industry specifically, or is it one vertical among many?

A generalist agency spreads the same playbook across a real estate client on Monday and a SaaS client on Tuesday. That is fine for low-stakes categories. It is a real risk in regulated or high-consideration industries, where the content, claims and even the schema markup carry compliance weight a generic template was never built to handle. Ask how many current clients operate in your exact category, and ask what they had to unlearn from a different vertical to serve it properly. An agency that works with lenders and fintech brands exclusively, for example, starts a lending engagement already fluent in RBI fair-practice language and financial-ads policy, rather than discovering those constraints in month three.

2. Can you show baseline-versus-current numbers, not just a percentage?

"We grew organic traffic 184 percent" means very little without knowing the starting point, the time window, and which system the number came from. A serious agency shows you the baseline month, the measurement window, and the exact metric definition, footnoted. Ask to see a full case study with baseline-versus-current data rather than a highlight reel of percentages. If every case study on their site ends in a happy number with no starting figure attached, that absence is itself the answer.

3. How do you verify your own case studies, and will you let me apply the same test to theirs?

This is the question that separates agencies who expect scrutiny from agencies who avoid it. A trustworthy partner can name the specific things you should check: whether the baseline is a single month or an average, which system the conversion number came from, whether the cost metric includes fees, and whether new coverage, like newly claimed branch listings, is inflating the growth figure. Some agencies publish this exact test as a public checklist against their own results. That willingness to be checked is a stronger signal than the results themselves.

4. Which AI platforms do you actually track, and how often?

"AI SEO" without a named tracking cadence is a slogan, not a service. Ask specifically which assistants they monitor, since ChatGPT, Google AI Overviews, Gemini, Perplexity and Copilot are the current set worth asking about by name, how many prompts are in the tracked set, and how often the set re-runs. A monthly re-run with a fixed prompt list and a fixed competitor set is the current baseline for defensible reporting. Anything vaguer than that is guesswork dressed as a service.

5. How do you make our content quotable, not just indexable?

Ranking and being cited are different outcomes. A page can rank on page one and still never get quoted by an assistant, because the actual fact, whether that is a rate, an eligibility rule or a fee, is buried in an image, a PDF, or a JavaScript-only table the model cannot parse. Ask what specifically changes in how content is written and marked up, including schema type, sentence structure and table format, to make a fact both correct and liftable by a model without hedging.

6. Do you manage crawler and bot access, or does that sit outside your scope?

This is a technical question most SEO conversations skip entirely, and it is often where the real problem lives. AI crawlers such as GPTBot, PerplexityBot and Google-Extended get blocked by default in a lot of CDN and WAF configurations, sometimes without anyone on the client side realising it. An agency doing this work properly reviews robots.txt, CDN rules and llms.txt as a standard part of an AI visibility engagement, not as a separate technical audit you have to request and pay for again.

7. How do you measure AI-referred traffic, given most assistants strip the referrer?

A meaningful chunk of AI-driven research never shows up cleanly in Google Analytics, because the assistant sending the visitor does not always pass a referrer. An honest agency tells you this upfront and explains how they estimate the gap rather than ignoring it or, worse, presenting an undercounted number as the whole picture. If an agency claims perfect attribution for AI-referred sessions, that claim itself is a red flag.

8. What happens when a model states something factually wrong about us?

For any regulated or reputation-sensitive brand, this is arguably the highest-stakes question on the list. Language models repeat outdated facts confidently, often long after the source page has been corrected, because the model or an aggregator page is quoting stale information. Ask for the actual remediation process: how errors are found, how the source is traced, and how long correction typically takes once the source is fixed. In one lending engagement, an AI visibility programme traced 21 factual errors, including a discontinued fee that assistants kept repeating, back to two outdated aggregator pages, and cut error-containing answers by 86 percent within a few months of fixing the source. That is the kind of specific process answer to listen for, not a vague assurance that "we monitor for accuracy."

9. Do you still do the SEO fundamentals, or only the AI layer?

If an agency's pitch is entirely about prompts, citations and AI share of voice with no mention of technical SEO, indexation, Core Web Vitals or content architecture, be cautious. Retrieval and ranking systems both depend on the same underlying signals: a crawlable site, clean structured data, and content that actually answers the question. A proper SEO and AI search practice runs both as one engine, because AI visibility built on a technically broken site does not hold.

10. What is your content and compliance review process?

If your industry carries any regulatory sensitivity, whether that is financial services, healthcare, insurance or anything adjacent, ask exactly who reviews content before it publishes and at what stage compliance gets involved. "We check it before it goes live" is not a process. "Content is briefed against your compliance guidelines and reviewed with your team before publishing" is a process. The difference shows up the first time a piece of content needs a rewrite after legal has already flagged it.

11. Can you tell me something that did not work?

Every real engagement has at least one abandoned experiment, whether that was a content format that never ranked, a messaging test that raised opt-outs, or a tactic that looked good on paper and produced nothing. An agency that cannot name one, unprompted, is either too new to have learned anything or is only showing you the parts of the story that sell. Agencies confident enough to publish what did not work alongside their wins are telling you they measure outcomes honestly, not just favourably.

12. Who owns the dashboard and the data when the engagement ends?

This is the question people forget to ask until it is too late. Some agencies build reporting inside a proprietary tool you lose access to the day you cancel. Others connect your own data sources, such as GA4, Search Console, ad accounts and your CRM, into a dashboard you own outright, in a platform like Looker Studio, that keeps working with or without them. Ask specifically what you keep. If the honest answer is "nothing, it is built inside our internal system," price that risk into your decision before you sign.

Running the actual evaluation call

Do not send these twelve questions by email and accept a written response. Get on a call and ask them live, in order, and pay attention to two things beyond the content of the answer itself: how specific it is, and how comfortable the person sounds naming a limitation. Vague, confident answers to every single question are a worse sign than one honest "we do not do that particularly well yet." Most of the structural problems that sink an SEO engagement, whether that is ranking for nothing but your own brand name or discovering a year into the contract that AI crawlers were quietly blocked the whole time, are visible in how an agency answers a direct technical question, long before you ever sign anything.

If you want a second opinion before committing to any agency, including one you are currently talking to, a free Growth Audit gives you an independent read on your technical SEO, AI visibility gaps and content foundation, delivered in writing within five working days with no obligation attached. It is a useful benchmark to hold any proposal against, ours included.

The short version

An AI SEO company worth hiring can tell you, specifically and without hedging, which assistants they track, how they verify their own results, how they fix factual errors a model is repeating, and what you keep when the contract ends. Everything else in a pitch deck is decoration around those four points. Ask the twelve questions above before you sign anything, and be more suspicious of an agency with a perfect answer to every one of them than an agency honest enough to name where it is still improving.

Frequently asked questions

1. What is an AI SEO company?

An AI SEO company optimises a brand's visibility both in traditional search rankings and inside AI-generated answers from tools like ChatGPT, Google AI Overviews, Gemini and Perplexity. The work combines classic technical and content SEO with newer practices such as schema and entity optimisation, prompt-set tracking, and citation reporting, often called AEO, short for answer engine optimisation, or GEO, short for generative engine optimisation.

2. How much does an AI SEO agency typically cost?

Pricing varies widely by scope and market, but most AI SEO retainers are structured the same way traditional SEO retainers are: a monthly fee tied to a defined set of deliverables, often following an initial paid or free audit that scopes the work. Rather than comparing price alone, compare what is included at that price, particularly prompt-tracking cadence, reporting ownership, and whether technical SEO fundamentals are covered or billed separately.

3. Is AI SEO different from traditional SEO?

AI SEO builds on traditional SEO rather than replacing it. Traditional SEO focuses on ranking pages in classic search results. AI SEO adds the work needed to get a brand named and cited inside AI-generated answers, which depends on the same crawlable, well-structured foundation traditional SEO already requires, plus additional schema, entity and prompt-tracking work specific to how language models select and quote sources.

4. How long does it take to see results from an AI SEO agency?

Most engagements need two to four months before AI citation movement becomes measurable, and six months or more before it compounds meaningfully, similar to the timeline for traditional organic SEO. Citation-specific fixes, such as correcting a factual error a model is repeating, can move faster, sometimes within four to nine weeks of the underlying source being corrected, because that is a narrower, more mechanical fix than building topical authority from scratch.

5. What is the difference between AEO and GEO?

AEO, or answer engine optimisation, focuses specifically on earning mentions and citations inside AI-generated answers on platforms like ChatGPT and Perplexity. GEO, or generative engine optimisation, is the broader practice of structuring content, schema and entity signals so any generative AI system can accurately parse and reuse it. In practice, most agencies use the two terms interchangeably to describe the same overall discipline.

6. Can our existing SEO agency just add AI SEO on top of what they already do?

Sometimes, but ask them the same questions you would ask a new agency before assuming so. The technical foundation of crawlability, schema and structured data genuinely does carry over, so an agency with strong technical SEO fundamentals has a real head start. What often does not carry over automatically is prompt-set tracking, citation reporting, and the specific content restructuring that makes a page quotable by a model rather than merely rankable, so it is worth confirming those pieces are actually in scope rather than assumed.