AI SEO agency: what the work is, and what to ask before you hire one

Four workstreams, the deliverable each one should produce, and the six questions that tell real measurement from a dashboard with a score on it.

Published by AI Knows Us · Updated 12 October 2026

An AI SEO agency, also sold as a generative engine optimisation or GEO agency, works to get your business named inside the answers ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews give when buyers ask. The real work is four things: a repeatable measurement of whether you are named today, technical access so the engines can read your site, pages written to be quoted, and getting your name onto the third party sites those engines cite. A proposal that does not name all four is covering part of the job. This page gives you the questions that tell you which kind you are talking to.

The four workstreams, and the deliverable for each

Ask for the deliverable by name. Agencies that do this work can produce these without hesitating.

WorkstreamDeliverable you should receiveHow you verify it
MeasurementA written prompt set, the engines tested, the number of runs per prompt, and the raw answers saved with datesRe run five prompts yourself and compare
AccessA robots and access report naming each AI crawler, plus a server log check showing which ones actually arrivedOpen your own robots file and read the log
ContentPublished pages, one per buyer question, answer in the first paragraph, each with a checkable factRead three of them and see if you could quote one
Third party presenceA named target list of directories, review platforms and publications, with status per targetSearch for your own name and see where it appears

What the measurement work actually involves

This is the part that separates a working agency from a reseller, because it is the part that is hard to fake once you know what to ask for.

A defensible measurement has a fixed list of buyer questions, written in the words a buyer would use, with your company name nowhere in them. It runs each question several times in each engine, because the same question asked twice can return different companies. It records which sources the engine cited, not only whether you were named. And it is dated, so the next run is a comparison rather than a fresh opinion. We published the format we use in how to report AI search visibility, and the cadence question in how often you should measure AI visibility.

The citation list is the most valuable thing in the report. If an engine cited four directories and a trade publication when answering your category question, you have just been handed your third party target list for the next quarter.

How to tell real measurement from a dashboard

Plenty of dashboards in this category show a score going up. Six questions tell you what is behind it.

  • Can I see the prompt list? If the prompts are hidden, the score cannot be checked or repeated.
  • Is my company name in any prompt? It should not be. Naming yourself primes the answer and inflates every number after it. The fair protocol is in should you ask ChatGPT about your own company.
  • How many runs per prompt, and what is the spread? One run per month is an anecdote. Several runs give you a range you can act on.
  • Which engines, and through which interface? An answer from an API with no browsing enabled is a different thing from the product your buyer uses. We covered why one app is not enough in why one AI app is not enough to measure with.
  • Are the raw answers kept? You want the text, not just the count, because the wording tells you what to fix.
  • Can I reproduce last month's figure? Ask them to show you how.

The questions to ask in the first meeting

  • What is my mention rate today, on which prompts, in which engines, and when did you run it?
  • Which of my competitors is named most often, and which sources are cited when they are?
  • How many pages will you publish in the first ninety days, and who writes them?
  • Where will my facts come from? Hand them your real numbers and watch whether they ask for them.
  • Which third party sites are on the target list, and what does each one require?
  • What does the monthly report contain, and can I re run the test myself?
  • Who owns the pages and the data if we stop working together?
  • What is the price, in writing, and what is included at each tier?

On the last one, published pricing in this category varies widely by scope and market. We surveyed what is actually published in GEO agency pricing, a dated survey of 30 published offers, so you can walk into the conversation with a reference point.

Signals worth acting on

Three claims should make you stop and ask for evidence.

A promised placement. Walk away from anyone who guarantees a position in an AI answer or a set number of mentions. Google's own guidance on hiring an SEO says plainly that no one can guarantee a number one ranking and tells site owners to beware of providers who claim to guarantee rankings. The same reasoning holds here, because the engine decides per question and per run.

A claimed special relationship. That Google page also tells readers to beware of anyone alleging a special relationship with Google. Apply it to this category: ask to see the agreement with OpenAI or Google that is said to place your business inside an answer.

A single file as the headline deliverable. An llms.txt file takes an afternoon and is worth having. It is not a programme. Google states plainly in its AI features documentation that you do not need to create new machine readable files or AI text files to appear in AI Overviews or AI Mode.

Agency, freelancer or in house

The split follows where your constraint is.

  • In house suits you when you have someone who can write with authority about your category. The measurement is a disciplined hour a month with a spreadsheet, and the writing is the real cost.
  • A freelancer or consultant suits a narrow category with a short question list, where one expert writer beats a team.
  • An agency earns its fee on the parts that are mechanical at volume: running the prompt set across engines every month, chasing dozens of third party listings, and keeping pages shipping on a schedule.

Whoever you hire, go into the conversation with a number. Ask ten questions a buyer would ask, leave your brand name out of all of them, and write down who gets named. That takes an afternoon and it changes the meeting: you are discussing a gap you can both see rather than a proposal. AI Knows Us produces that reading and keeps the cited sources alongside, which is also the thing to ask any agency how it will move.

This page is about hiring for AI answer visibility. For choosing a traditional search agency, see how to choose an SEO agency. For the questions to put in any marketing agency brief, see before you hire a marketing agency.

The AI SEO agency questions worth an answer

What is the difference between an AI SEO agency and a GEO agency?

In practice, the name only. Ask both for the four deliverables above and compare the answers rather than the labels. The term a firm picked tells you about its marketing, not its method.

Can my current SEO agency do this?

Often yes for three of the four workstreams, because crawl access, content and third party presence are adjacent to what they already do. The measurement is the piece most SEO agencies have not built, so ask that question first.

How long before anything moves?

Access fixes land within days of the next crawl. New pages tend to appear in live, searched answers over a few weeks. Answers that come from model training move on a longer horizon, so judge a programme on a quarter rather than a month.

What should the first ninety days contain?

A baseline measurement with the prompt set written down, a crawler and render access fix, five to ten pages published against real buyer questions, and a third party target list with work started on it. Anything less is a report.

Do I need to stop my existing SEO work?

No, and stopping it would make this harder. Google requires a page to be indexed and snippet eligible before it can be used in AI Overviews or AI Mode, so ranking work is the foundation this sits on.

What to do first

  • Get a baseline before you brief anybody. Ask ten buyer questions in two engines, without naming your company, and save the answers.
  • Note which competitors come back and which sources are cited. That is the brief.
  • Send the four deliverable names to every agency on your list and compare the replies side by side.
  • Ask each one to re run five of your own prompts in front of you.
  • Agree the report format and the price in writing before the first invoice.

See what AI says about you.

The first scan is free, it needs no account, and it asks the questions your buyers ask.

Free. No card. We ask 5 real buyer questions on 2 AI apps.

Already sure you want this? Join the waitlist. We open on 31 October 2026. It costs an email address. No card, and nothing is charged. Or talk to a person if you have a question first. Or read how we fix what it finds.