What is a GEO agency and what does one do

A GEO agency gets a brand named and cited in the answers AI tools hand people. Someone asks ChatGPT which payroll provider handles contractors in Spain, or types the same question into Google and gets an AI Overview sitting on top of the results instead of ten blue links. Perplexity, Gemini, Claude and Google’s AI Mode all field that question too. A GEO agency’s job is to make sure the brand turns up in those answers with an accurate description attached, and that the pages the answers cite say what they ought to say about it. GEO stands for generative engine optimization.

Most of that work is ordinary SEO. ChatGPT, Gemini and the rest build answers out of pages that already rank, plus a small pool of third-party sources they’ve come to trust on a given topic (listicles, comparison pages, review sites, trade publications, Reddit threads). So a brand that wants to be in the answer has to rank its own pages the normal way, and it has to get itself named, correctly, on the third-party pages the models keep going back to. An agency that skips both of those and sells you a text file for your site root is selling you a parking permit for a building with no parking lot.

Where the term came from

Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande posted a paper called “GEO: Generative Engine Optimization” to arXiv in November 2023 (arXiv 2311.09735) and presented it at KDD 2024. They were working out of Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi. They tested content edits against a benchmark of generative answers and reported visibility gains of up to 40% from moves like adding quotations and statistics to a page.

That’s where the acronym starts, and it sat in academic territory for a while, because nobody outside the paper walks around calling ChatGPT a generative engine. What put GEO in front of the people who sign invoices was Andreessen Horowitz publishing “How Generative Engine Optimization (GEO) Rewrites the Rules of Search” in May 2025, and New York Magazine running “SEO Is Dead. Say Hello to GEO” that August. Then Google settled the naming fight in its own 2026 guide, “Optimizing your website for generative AI features on Google Search,” which says that from Google Search’s perspective, optimizing for generative AI search is optimizing for search, and so it’s still SEO.

None of this is old. Google launched AI Overviews to every US user on May 14, 2024, and AI Mode followed on May 20, 2025. OpenAI opened ChatGPT search to paid users on October 31, 2024 and to everyone that December. Perplexity has been citing its sources inline since it started in 2022, and Microsoft shipped the AI-powered Bing on February 7, 2023. Anyone claiming a decade of expertise here is doing arithmetic we’d like to see.

What the work consists of

Here’s our method, in the order we run it.

1. We audit first. That means running the brand’s category queries across ChatGPT, Perplexity, Gemini, Claude and Google’s AI Overviews and AI Mode, over a few weeks rather than in one sitting, because answers rotate and a single check will lie to you. We record which brands get named and which pages get cited. We also write down how the brand gets described when it does appear, since stale pricing and a product name you retired two years ago are both fixable once somebody notices them.
2. Then we go after the sources. The audit hands us a list of pages the models keep citing, and we work that list to get the brand added where it’s missing and corrected where it’s wrong. That’s outreach to publishers, one publisher at a time. Some of them want something back (a link, a mention, money, product access), so this is negotiation work, and it moves at the speed of other people’s inboxes.
3. We build citable content on the brand’s own site, which mostly means pages that answer the question in the first two sentences instead of warming up for four paragraphs. The “best X” pages and the comparison and alternatives pages all need to name the brand and say plainly what it’s for. The headings stay clean and the facts stay specific, without the padding a model has to dig through to reach the claim.
4. We clean up the entity. The company gets the same one-sentence description everywhere it appears, meaning the site, LinkedIn, Crunchbase, G2, the directories, and Wikipedia or Wikidata where the company warrants an entry. Organization schema goes on the site. The About page says what the company is without a paragraph of throat-clearing first.
5. We check technical access. GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended all need to reach and render the pages you want cited. Google-Extended is the one that gets misread, because it controls whether your content trains Gemini and has no effect on AI Overviews, which run off Googlebot’s index. If your product comparison table only exists after JavaScript runs, assume nothing is reading it.
6. Then we track it. We re-run the query set on a schedule and report share of answers by platform, plus which citations we gained and which wrong ones we got fixed.

Underneath all of that sits ordinary SEO, because every step above rides on the brand’s pages being able to rank in the first place. That’s why our services page reads like an SEO scope with a citation program attached.

What a GEO agency shouldn’t be selling you

Google’s 2026 guide is useful here, because it rules out several things that get sold as GEO. It says llms.txt does little for Google’s AI features, and that you don’t need to chunk your content or rewrite it for AI. Its earlier documentation, “AI features and your website” from 2025, said there are no additional requirements for showing up in AI Overviews or AI Mode beyond normal Search fundamentals. The 2026 guide also tells you not to manufacture mentions from trusted sites to build authority, which is the polite version of a warning.

“AI-specific schema” is the one we get asked about most. Schema.org has Organization, Product, FAQPage and Article, and Google documents which of them earn rich results. There’s no hidden AI-only type. Mark up what you have, correctly, and move on.

A guarantee of placement is the other thing to walk away from. Nobody controls what ChatGPT says about a category next Tuesday, and any agency showing you one screenshot of a good answer is showing you one roll of the dice. Ask for the same query run across a month. We wrote out how we score that in our piece on the AI visibility audit.

How a GEO agency differs from an SEO agency

Mostly it doesn’t, and the honest ones will tell you that before you ask. Google said the same thing in its own words in 2026, and Ahrefs published a post in 2025 making the argument that GEO is mostly SEO. Ranking, technical work, content and links all carry over, because the systems generating those answers are reading an index built by the same crawling and ranking machinery.

The real addition is the third-party half. A traditional SEO retainer cared about a listicle mostly as a link source. A model reading that same listicle cares whether your brand is in it at all, and what the sentence next to your name happens to say. Getting the brand added to the pages where it’s missing, and getting the outdated description fixed on the pages where it’s wrong, didn’t have a line item on an SEO scope five years ago. That’s the part of what a GEO agency does that earns the separate name, and it’s where most of our hours go. We went through the overlap in more detail in GEO vs SEO, and the mechanics of getting into a specific model’s answers in how to get cited by ChatGPT.

So the practical difference between hiring a GEO agency and hiring a good SEO agency comes down to whether that SEO agency is doing the citation and mention work, and whether anyone there is reading AI answers every week. Plenty of SEO shops now do both, and plenty will tell you AI search is a rounding error right up until a client shows them a chart.

What we’d bet on

We don’t know how long the current arrangement holds. Google can change how AI Mode picks sources next quarter, and OpenAI has already changed how ChatGPT search works more than once since it opened up in late 2024, so anyone selling you certainty about how any of this looks in 2028 is guessing with a straight face.

What we’d bet on is narrower. The pages these models cite keep turning out to be pages that rank, and a brand that’s described accurately across the sources those pages draw from keeps getting named. That’s held since AI Overviews went wide in May 2024, and it’s what our method is built around. If it stops being true, we’ll say so here before we say it to a client.