Generative Engine Optimisation (GEO)
The term came out of a research paper in 2023. Since then it has been resold as a discipline with its own budget line. Here is the honest version: GEO describes how generative systems retrieve and cite sources, that selection is probabilistic, and the only lever you genuinely own is how clearly your business is described across the properties you control.
Work directly with the person doing the work
Meet the team →AI Overview
Generative engine optimisation (GEO) is the practice of improving how often and how accurately a website is retrieved and cited by generative search systems such as Google's AI Overviews, ChatGPT, Gemini, Copilot and Perplexity. The term was coined in a research paper submitted to arXiv on 16 November 2023, which studied a purpose-built research prototype rather than any live product. In practice the levers are ordinary SEO levers: crawlable pages, plainly stated facts, standard schema.org markup, and a business entity that stays consistent across the properties you control. Loudachris sells a free measurement diagnostic, not a separate GEO retainer.
Provenance
GEO is not a marketing word. It started as a paper.
“GEO: Generative Engine Optimization” was submitted to arXiv on 16 November 2023 by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande. It is public and free to read at arxiv.org/abs/2311.09735. The authors named the problem before an industry existed to sell the solution: content creators have very little control over when and how a generated answer surfaces their work, and they wanted a framework for measuring visibility inside those answers.
Read the method section before quoting the headline. The paper evaluates against GEO-bench, a benchmark of queries and web sources the authors assembled, and it measures a generative engine they built for the study. Whatever visibility lift it reports is a finding about that harness in November 2023, not a measurement of Google's AI Overviews, ChatGPT, Gemini or Perplexity as they ship today, and it is not a result anyone can promise you.
2023
The year the term was coined. Every product sold as GEO since is younger than the paper that named it.
6
Authors on the paper, working in academic research rather than agency sales. Their goal was a measurement framework.
0
Live commercial generative engines the study optimised against. It used a prototype the authors built themselves.
How a generative system actually reaches your page.
4 stages sit between a question and a citation. Each one narrows the field, and you have progressively less influence at each step.
1. Query fan-out
The question a person types is rarely the query that runs. Generative systems rewrite it, often into several queries at once, to cover the ways an answer might be phrased. This is the first place visibility is won or lost, and it happens before your site is involved at all.
2. Retrieval
Each query hits an index and returns candidate passages, not whole websites. A page that is blocked, slow to render, buried behind a script, or duplicated across 4 near-identical URLs is a page the retriever quietly skips. Nothing downstream can rescue it.
3. Grounding
From a long candidate list, a handful of passages are selected to write the answer from. Selection favours text that states something specific and checkable. A paragraph that says quality workmanship and competitive rates contains nothing to ground an answer in, so it loses to one that names the service, the suburb and the price.
4. Attribution
The visible citations are a subset of what the system read, chosen while the answer is written. Being read and being credited are 2 different outcomes, which is why a business can influence an answer without ever appearing as a link in it. Nobody outside the vendor controls this step.
The Part Nobody Sells
Ask the same question twice and you will get 2 answers.
A ranking is a position on a page that anyone can check and re-check. A citation is an outcome of a generation, and it moves for reasons that have nothing to do with your website. That is not a flaw in the systems. It is how they work, and it is the single most important thing to understand before you sign anything sold as GEO.
6 variables can change the sources in front of you without a single line of your site changing:
What follows from this: a screenshot is not evidence. If someone shows you a single assistant naming a business, that is 1 sample from a distribution, and the next run may look nothing like it. The only defensible way to talk about generative visibility is to fix the prompts, run them repeatedly, log every result including the ones that went against you, and describe the pattern rather than the best screenshot. That is a measurement job, not an optimisation product.
Entity Clarity
The only lever you genuinely own.
You cannot reach inside a retriever. You can decide what it finds when it looks. Every generative system has to work out whether the records it holds describe 1 business or several, and it resolves that question using the properties you control: your website, your Google Business Profile, and the listings and profiles carrying your name.
When those agree, a system can state a fact about you. When they contradict each other, the safest output is an answer about somebody else. That is the whole mechanism, and it is why entity work outperforms every tactic invented specifically for AI.
Yours to control
- ✓A single canonical page per service, reachable and rendered in the HTML
- ✓The same trading name, address and phone number, everywhere
- ✓The same words for the same service across every page
- ✓Standard schema.org markup describing the business and its services
- ✓A complete Google Business Profile with real categories and hours
- ✓Prices, service areas and process stated plainly instead of implied
Not yours to control
- ✕Which index an assistant retrieves from, and how fresh it is
- ✕How your customer's question is rewritten before retrieval
- ✕Whether the system searches at all or answers from training
- ✕Which of the passages it read receive a visible citation
- ✕The model version, the session history and the personalisation
- ✕What a forum thread or a third-party listicle says about you
Every item in the left column is standard SEO work. Nothing in it was invented for generative search, which is the point. Rename the category whatever you like: underneath the acronym this is SEO, delivered inside the retainer you would be paying for anyway, alongside local SEO.
What Gets Sold As GEO
The deliverables, checked against the documentation.
These are the line items that turn up on GEO quotes. We are not naming firms, because the tactic is the problem and it is being sold by plenty of them. Check each one against what the platforms have actually published.
Sold as
Status
llms.txt file
Google states no AI text files are needed to appear in AI features
AI-specific schema vocabulary
Google states there is no special schema.org structured data to add
Content chunking “for AI”
No requirement to break content into small pieces
Writing specifically for AI
Not required. Write for people
Speakable schema as an AI lever
Limited availability, news publishers only
HowTo rich results
Retired from Google Search
Guaranteed AI citations
Citation selection is probabilistic and varies by phrasing, session and model version
Source for the first 4 rows: Google's AI features and your website documentation. The speakable row comes from Google's speakable structured data reference. The longer version of this teardown, with the reasoning behind each row, sits on what actually works in AI search.
The Commercial Bit
A diagnostic with published scope, not a GEO retainer.
We do not sell generative engine optimisation as a product, because we cannot name a deliverable inside it that is not already SEO. What we do sell, at no cost, is the measurement step almost nobody runs before spending: an honest read on whether generative systems can find and describe your business at all.
Here is the entire scope, published rather than described on a call.
01
We fix the prompts before we run anything
A set of questions in the wording a real customer would use, agreed with you in advance so the test cannot be steered toward a flattering result.
02
We run each prompt repeatedly, not once
Across the assistants that matter to an Australian buyer, logging every run including the ones where nobody mentions you.
03
We record which domains got cited
Not just whether you appeared. Who did appear, how consistently, and what their pages said that yours did not.
04
We audit the entity layer
Name, address, phone, categories, service naming, canonical pages and schema across your site, your Google Business Profile and the listings carrying your details.
05
We separate the fixable from the noise
Some gaps trace to something you can change. Others are variance. We will tell you which is which, and we will tell you if the answer is that you have no problem worth paying to solve.
The diagnostic is free. If it finds work worth doing, that work goes into website SEO, which starts from $1,500+GST/month, month to month with no lock-in. There is no second invoice with “GEO” written on it, no guaranteed placement, and no promised date, because none of those 3 things can be delivered honestly. See the SEO cost guide for what moves that figure, or the audit page for how the diagnostic runs.
About this page
Written by Chris Lourenco, Director of Loudachris Digital Marketing, a Google Partner agency based at 21-22 Greenhill Road, Wayville SA 5034. ABN 73 630 143 190. Every claim on this page is either linked to a primary source or stated as our own opinion. Where we do not have evidence, we have said so instead of filling the gap with a statistic.
We have deliberately published no AI-search case studies here, because we do not have outcome data we would be willing to defend. When we do, it will appear on our results page with the client named, the same way every other result we publish is. More on how the wider hub is structured on AI search visibility, and more on us on the about page.
Last reviewed: 9 August 2026.
Questions about generative engine optimisation.
What is generative engine optimisation?
Who coined the term GEO?
Is GEO just SEO with a new name?
Is GEO the same thing as AEO?
How do generative engines choose which pages to cite?
Why do we get cited for one phrasing of a question and not another?
Can anyone guarantee a citation in ChatGPT, Gemini or Perplexity?
What is entity clarity and why is it the lever that matters?
Should we publish an llms.txt file or add AI-specific schema markup?
Does the 2023 GEO paper give us a playbook we can follow?
How do we find out whether we have a GEO problem at all?
What does GEO work cost with Loudachris?
Book Your Free AI Search Diagnostic
We'll agree the prompts with you, run them repeatedly across the assistants your customers use, and give you a straight answer on whether you have a retrieval problem worth fixing. 30 minutes. No cost. No obligation.
Or call 0403 454 199 or email chris@loudachris.com.au
Keep Reading
GEO is 1 page of a larger argument. The rest of it is here.
What Actually Works
The full teardown of the tactics sold as AI optimisation, with the documentation behind each verdict.
Answer Engine Optimisation
The other acronym, where it came from, and how it differs from GEO in practice rather than in a pitch deck.
Choosing a GEO Agency
The questions to ask before you sign, and the answers that should end the conversation.
Find out whether you have a GEO problem at all.
Before anyone sells you a solution, measure the problem. The diagnostic is free, the scope is published above, and the honest outcome is sometimes that there is nothing here worth paying for.
Call 0403 454 199
Adelaide-based? AI search in Adelaide covers the local picture. Working in a specific vertical? Start at industries.