How to Choose a GEO Agency: 12 Criteria

Choosing a GEO agency is not procurement; you are deciding who shapes how your brand is represented inside AI answers. Base that decision on evidence an agency can show, not on the polish of its pitch.

GEO (Generative Engine Optimization) means getting your brand cited as a source in answers produced by ChatGPT, Gemini, Perplexity and Google's AI Overviews. The field is new, so two kinds of supplier sit side by side: teams that do the work, and teams that changed the cover of an SEO proposal and typed "GEO" on it. From outside, both say the same sentences.

This is not a "best agencies" list — those are subjective, mostly sponsored, and none knows your business. It is a framework for your own assessment: 12 criteria, 8 questions, and the red flags worth stopping on.

What a GEO Agency Actually Sells

A GEO agency does not sell traffic. A large share of AI answers end without sending anyone to your site; the answer arrives in the chat window. What is sold is representation: asked "what should someone getting a dental implant in Istanbul watch out for?", the AI cites your clinic, gets your price range right, does not confuse you with a competitor.

Its concrete components:

Entity data: brand name, services, location, opening hours, price range — on the site in machine-readable form (structured data / JSON-LD).
Accessibility: AI crawlers can reach the site — robots.txt permissions, server-rendered HTML, blocking firewall settings reviewed.
Quotable content: text answering the question directly, in short, clear paragraphs.
Off-site consistency: directories, industry sites, forums and review platforms that do not contradict each other.
Measurement: regular testing of whether you appear in answers at all.

An agency that talks only about the last of the five, or only about content, hands you an incomplete proposal. The whole is set out step by step in how to build a GEO strategy.

The Assessment Framework: 12 Criteria

Use the 12 items below as a checklist. Score each 0 to 3, judge the total out of 36, and sign with nobody scoring under 25.

Technical audit? Do they examine your site before quoting and bring concrete findings, or does everyone get the same deck?
Structured data. Who writes the Schema.org markup, which types, how is it validated?
Crawler access. Do they name GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot?
Rendering. Do they ask whether content is built in the browser or on the server, and advise accordingly?
Who writes the content? An editor who knows your sector, or unreviewed generated text?
Domain expertise. Have they worked in your sector; which regulatory limits do they know?
Measurement method. How is visibility measured, with which query set, how often?
Reporting transparency. Raw data access, or only a summary slide?
Ownership. When the contract ends, who keeps content, accounts and technical configuration?
Team visibility. Can you meet the people doing the work, or does nobody appear after the salesperson?
Scope clarity. Hours per month, pieces of content, technical items — in the contract as numbers?
Ease of exit. Term length, early-termination condition, exit within 90 days?

These ask not how well an agency markets itself but what it can document. That is where the difference sits.

How to Tell Whether an Agency Is Technically Competent

Technical competence is the fastest thing to test in a meeting: the vocabulary is either known or it is not. Have them open your site and ask, "what on this page can an AI crawler not see right now?"

A competent team makes observations like these within ten minutes: content loads later via JavaScript and is absent from the raw HTML; product pages carry no Product markup; robots.txt blocks an AI bot nobody meant to block; two pricing pages contradict each other on the same service; the FAQ sits inside an accordion with no heading tag.

An incompetent team answers with "we will make your content AI-friendly" — a sentence with no checkable meaning.

A small practical test: check the agency's own site. JSON-LD in the page source? A file like nextgeoagency.com/llms.txt? An agency whose own site is closed to AI crawlers cannot teach you this work.

Measurement and Reporting: Which Numbers Are Real?

Measurement in GEO is hard because AI answers vary by person, by session and by model. Ask the same question twice and you can get two answers. One screenshot is not evidence.

A serious agency builds it like this:

Defines a fixed query set describing your business — usually 30 to 100 questions — and shares it with you.
Runs those queries at regular intervals across more than one engine.
Records three things per run: was the brand mentioned, in what context, which sources were cited.
Reports competitors' mention rate on the same queries.
Shows from server logs whether AI crawlers really visited.

For the method in detail and the tools behind it, see how to measure AI visibility. Notice how many of these the agency lists unprompted.

Eight Questions to Ask the Agency

Ask these as written and write the answers down. What matters is less the answer than how concrete it is.

"What are three technical problems on our site that need fixing right now?"
"Which query set will you measure visibility with, how often, and will we see it?"
"Who writes the content, and what is that person's record in our sector?"
"Do you use AI-generated text? If so, how does editorial review work?"
"What concrete deliverables land in the first 90 days?"
"If the contract ends, who keeps the content, the schema code and the accounts?"
"Can we speak to a business in our sector you have worked with?"
"What is the hardest part of this work, and where could you fail?"

The last is the most revealing. An experienced team lists risks: "few authoritative publications in your sector, so citations come slowly"; "you will not publish prices, a disadvantage on some queries"; "your infrastructure does not support server-side rendering, a development cost". No risk in a proposal means nobody thought about it.

Red Flags and Good Signals

Red flag	Good signal
"We will get you to number one in ChatGPT" — AI answers have no rank	"We will measure your mention rate on this query set and raise it"
A guaranteed outcome on a fixed date	A range, with assumptions written out
A generic proposal, sent without opening your site	Findings specific to your site, named in it
Measurement as "we will send screenshots"	Fixed query set, repeatable test, raw data
Volume as the only argument: "30 articles a month"	Fewer pieces, each tied to a query
Long contract, penalty for early exit	Three-month pilot, then mutual review
Accounts and content stay with the agency	Every asset opened in your accounts on day one
Its own site has no schema, no server-side HTML	Its own site applies what it describes

The pattern: good signals are verifiable, red flags are not. A sentence you cannot check is decoration, not a commitment.

If you already work with an agency, run the same audit on your supplier: how to tell whether your current agency is really doing GEO from its reports is a separate piece.

What to Look for in Pricing and the Contract

In Turkey, GEO is mostly sold on a monthly retainer, with a one-off setup fee on top for larger projects. Figures move so much with company size, sector and technical debt that a single price here would mislead you. Look not at the price but at how the price defines what you get.

Three things belong in the contract:

Unit of scope: pieces of content per month, technical work items, consulting hours. "As needed" cannot be measured.
Delivery schedule: what will have been delivered at 30, 60 and 90 days.
Ownership and handover: transfer of content, code and accounts when the contract ends.

Ask also that the first month be an audit month. A good agency does not produce in month one anyway; it measures your site, your data and your current visibility. A team publishing on day one does not know what it is writing or why.

The Decision: Build Your Own Scorecard

Meet two or three agencies, no more; decision quality falls as comparisons multiply. Score the 12 criteria the same day as each meeting — a week later what you remember is the salesperson's energy, not the substance.

One last check: show the proposal to a friend who knows your sector but nothing about marketing. If they cannot answer "what exactly will these people do?", it was not written clearly enough. GEO may be technical, but what it does and why should be explainable in plain language.

Frequently Asked Questions

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

An SEO agency targets your ranking on the results page; a GEO agency targets your brand being cited as a source in AI-generated answers. They overlap heavily on technical groundwork — structured data, site speed, content quality. The difference is in measurement and in how content is built: the goal is not the click but the citation.

How long does GEO work take to show results?

Technical fixes take effect relatively quickly; content and authority effects surface more slowly, and the field generally accepts that the first meaningful change starts to be measurable within a few months. Timing depends on your site's technical state and on competition in your sector. An agency guaranteeing a firm date is promising what it does not control.

Does a small business need a GEO agency?

Some basics can be handled in-house: keeping contact details consistent, publishing clear answers to frequent questions, updating directory listings. Structured data setup, server-side rendering and regular measurement need technical support. The decision depends less on budget than on how much your business is searched for through AI.

Is it a bad sign if the agency produces content with AI?

On its own, no; unreviewed production is what is bad. Ask who verifies the text, where sector-specific facts come from, and what happens when something wrong is published. An agency that cannot describe its editorial review cannot manage the risk of publishing incorrect information in your name.

How do I check whether my current agency is really doing GEO work?

Ask for three things: the query set they use, the last three months of measurement data, and the server-log record showing AI crawlers reaching your site. You can also open the raw HTML of one page and check for structured data yourself. If those requests get no concrete answer, the service you pay for cannot be documented.

Still undecided after filling in your scorecard? A second opinion is the cheapest next step. At Next GEO Agency the first thing we do under our GEO and SEO services is always the same: measure how AI systems see your site and share the findings as they are. Start with a free business analysis and use the report whichever agency you choose.

GEO Ajansı Nasıl Seçilir? 12 Kriterlik Rehber