Recently, the term “GEO” (Generative Engine Optimisation) has become particularly popular. This is partly because people are increasingly turning to AI rather than searching on Google, and partly because Google itself has incorporated Gemini into its SERPs (search engine results pages).
So it goes without saying that these days, “being visible on social media” is starting to look almost more appealing than “ranking highly on Google”.
So here we go with slides, webinars and courses promising to reveal the incredible secrets of GEO. The problem is that there are also plenty of inaccuracies doing the rounds. And today we’re going to do a bit of fact-checking.
People talk about “optimisation for AI” as if there were already a single, shared and well-established standard. That is not the case. It is a relatively new field. Everything is still being defined and we need to tread carefully. So, without making promises we cannot keep, let’s simply be professional and set out what we know AS OF TODAY, fully aware that we must continue to study, day after day, because it is only natural that things will change.
How is this article put together?
Let’s start by setting the record straight: first of all, I’d like to point out that all the information you’ll find in this article comes from primary sources: material written by Mr Google, Mr OpenAI, Mr Microsoft, and so on. No unofficial or unverified external sources.
And then the structure: This article will be divided into two parts.
- In the first part, I’ll explain the basics and the differences between GEO and SEO, and then I’ll tell you what Google, OpenAI and Microsoft actually say.
- And a second part in which I will discuss the 10 “myths to dispel” and draw some conclusions, aiming to provide you with practical guidelines based on the previous analysis of primary sources.
It’s going to be a long, perhaps tedious process, but I’ve put a lot of work into it and I hope you’ll enjoy it. That said, let’s get started:
The name: GEO
The first problems arise from the confusion surrounding the name GEO.
The formalisation of the term Generative Engine Optimisation stems from a 2024 academic paper; Microsoft now explicitly uses the term in its official communications; Google and OpenAI, on the other hand, publish operational documentation on AI search, crawling, controls and inclusion, but do not offer a single, universal ‘GEO manual’. Taken together, this documentation suggests that ‘GEO’ today is primarily an umbrella term for various practices, rather than a discipline with a single set of codified rules.
SEO or GEO? Are they the same thing? Spoiler: NO!
If one wishes to use the term in a serious context, the primary reference is the paper “GEO: Generative Engine Optimization”, accepted at KDD 2024.
The paper discusses GEO as a paradigm for improving the visibility of content in generative AI responses, not simply as a means of “getting the AI to mention you”.
The central issue addressed in the paper is visibility in the generated responses, which includes citations, the presence of the content within the response, and the way in which the source is presented and cited.
The paper also clearly highlights that generative engines do not function like a traditional SERP: they do not simply return a linear list of results, but instead compile concise answers, with quotes arranged differently and using visibility metrics that differ from the traditional ranking systems on which those working in SEO are accustomed to relying.
So this is already a significant difference: GEO doesn’t mean ‘being mentioned by name’. In some cases, that might be the case and your brand might be mentioned, but it’s also possible that one of your pages will simply be used as a source, cited, summarised, or have key points extracted from it. But we’re not talking about working to get a super-rich snippet that potential customers will click on!
Even Microsoft, when referring to GEO in its official communications, links it to the way in which content contributes to what they call “AI-driven experiences” – that is, experiences in which citations, grounding, reasoning and results matter, not just clicks and traditional ranking.
What Google really says
Google’s official documentation is, at present, far less sensational than many of the slides circulating online.
On the page “AI features and your website”, Google makes one thing very clear: SEO best practices still apply to AI Overviews and AI Mode; there are no additional requirements for appearing in these experiences; and no special optimisations dedicated to AI features are needed. Google then goes into a lot of technical detail, mentioning coding elements that are not relevant to us at this stage, but the summary is what we have already discussed, namely that there is no specific schema.org “to appear in AI”.
So that alone should help you steer clear of those dodgy gurus we’re so fond of.
Google tells us that, even for AI, the basic fundamentals we should already be familiar with are key: allowing crawling, ensuring content is discoverable via internal linking, providing a good user experience, making important content available in text form without embedding text in images, using high-quality images and videos where appropriate, ensuring that structured data matches the visible content, and keeping information up to date on other Google services such as Merchant Center and Business Profile.
Furthermore, Google reiterates its commitment to unique, useful and satisfying content for users! This means that the work involved in planning and creating editorial content remains important.
Google continues to follow the same path it has been on for at least 10 years now: content quality! Ranking systems are designed to prioritise useful, reliable content created for people, not content produced to manipulate rankings. So all the ‘tricks’ and what were once known as ‘black hat’ techniques are useless.
So what exactly is quality content? Google itself tells us : originality, comprehensiveness, added value, clear sourcing, trustworthiness, evidence of expertise, and transparency regarding who created the content.
Above all, it clarifies two points that are often misunderstood:
- primo, E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) non è un fattore di ranking specifico;
- Secondly, trust is the most important factor among those Google uses to evaluate content.
When it comes to the relationship between AI and automatically generated content, however, the picture is more nuanced and the information changes from day to day: at times, it seems that Google is less dogmatic than many claim, given that the official documentation states that the focus is on the quality of the content, not on how it was produced.
However, articles are also appearing (with increasing frequency) that discuss Google’s intention to penalise automatically generated content. Google also adds a key condition: using AI or any other form of automation to generate a large volume of pages without adding value for users may violate the policy on scaled content abuse.
So where exactly is the dividing line? I don’t know. As I said, things are still a bit vague at the moment, so we’ll just have to wait and see.
And then there’s another point that’s often misunderstood: controls. If a website wants to limit what Google can display in its results or previews, it doesn’t need to use any new or mysterious tools: for the most part, these are the same controls that have been available for years for standard Google searches.
However, there is another setting, called Google-Extended, which serves a different purpose. Put simply, it is used to tell Google whether a website’s content can also be used for certain functions related to its artificial intelligence, such as the development and use of certain systems based on Gemini.
The important thing is this: this choice does not determine whether the site will appear in Google’s standard search results, nor does it help improve or worsen its ranking. It concerns a different use of the content.
This detail is crucial, because many discussions lump Google Search, Gemini Apps and other systems together as if they were a single entity. They are not!.
Finally, the issue of measurement must also be addressed with precision: Google states that sites featured in AI features are included in the overall traffic shown in Search Console, in the Performance report and under the ‘Web’ search type. Furthermore, information can also be found within Analytics by analysing the ‘referral’ sources. It is therefore incorrect to say that “the impact of AI on Google cannot be measured”. It is more accurate to say that it can be measured, but within aggregated metrics, which are not always separated in the ideal way for those carrying out analysis.
What OpenAI really says
In the case of OpenAI, the official documentation does not offer a general theory of GEO, but it does provide some fairly concrete practical guidance on ChatGPT Search, crawlers and publisher controls. OpenAI states that any public website can appear in ChatGPT Search.
To ensure that your content can be discovered, cited and clearly linked to in ChatGPT’s summaries and snippets, OpenAI recommends that you do not block OAI-SearchBot. If, on the other hand, you do not want a page to appear even as a link and title, the recommended approach is to use the noindex tag.
OpenAI also clearly distinguishes its user agents.
- OAI-SearchBot is used to ensure visibility in ChatGPT search results;
- GPTBot is used for crawling, with a view to potentially using the data to train foundation models;
and tells us that the two settings are independent. So we can decide, for example, that our website will be available to ChatGPT as a source for users, but not for training the AI model, and vice versa.
So, as we have seen with Google too, there isn’t just ‘one AI’; there are some bots used for ‘indexing’ and others used for ‘training’ the various models we use.
When it comes to the logic behind the ranking, OpenAI is rather vague: the official page on ChatGPT Search states that the ranking is based on a series of factors designed to help users find reliable and relevant information, and, as far as I am aware, there are no official communications or sources explaining strategies for securing a top ranking.
When it comes to analytics, OpenAI offers a useful – and often overlooked – tip: for users who allow the OAI-SearchBot, traffic from ChatGPT can be tracked because ChatGPT automatically adds the parameter utm_source=chatgpt.com to referral URLs. So, here too, it is incorrect to say that AI traffic is invisible. It’s not perfect (at least for now), but if we spend a bit of time working with Analytics, some data does emerge.
And then, still on the subject of ChatGPT, there is one important exception: e-commerce. OpenAI has launched a programme for merchants that includes a whole range of features to improve product visibility. Could ChatGPT be set to become a sort of personal shopper to whom you explain exactly what you need, and who then suggests the best results? It’s possible, but it’s still too early to say.
What Microsoft/Bing actually says
Of the major players, Microsoft is the one that currently uses the term GEO most explicitly in its official communications. In a post on the Bing Search Blog, Microsoft defines GEO as the practice of understanding how content contributes to AI-driven experiences, and explains that, in an AI-first world, visibility is no longer defined solely by rankings or clicks, but also by answers, citations, reasoning and results.
The most significant new feature on the Bing side is AI Performance within Bing Webmaster Tools. According to the official documentation, this tool tells you when content from your site is cited in AI responses on Microsoft Copilot, in AI summaries on Bing, and in certain partner integrations.
So, from a measurement perspective, this is a significant step forward compared to what Google and ChatGPT tell us. The metrics included cover total citations, average cited pages, grounding queries, page-level citation activity and visibility trends over time.
But Microsoft also sets out some very clear guidelines: the average number of pages cited has nothing to do with the concept of ranking, authority or the page’s role.
Microsoft’s operational guidelines are probably the most interesting because they are the most specific and practical. The documentation suggests enhancing depth and expertise, improving the structure and clarity of content, using clear headings, tables and FAQs where appropriate, supporting claims with examples, data and cited sources, keeping content up to date, aligning text, images and videos so that they consistently represent entities, products or concepts, using IndexNow to signal updates and, for local listings, managing data in Bing Places for Business. Do read the page I’ve linked to because it’s very interesting! At last we have a practical guide, but it should be taken for what it is: a set of practices to promote inclusion and citability, not a formula for ‘mastering AI’.
And above all, it is very interesting to note that what we read in Microsoft’s guide is not all that far removed from the best practices Google recommends for SEO.
There is no single “official GEO guideline”
If we consider Google, OpenAI and Microsoft together, the picture that emerges is quite clear: there is currently no single, cross-platform technical standard or official guideline known as GEO, with universal rules applicable to Google AI Overviews, ChatGPT Search, Copilot and all the rest.
Above all, it is a field that is constantly evolving: we mustn’t get stuck in the past, but must always keep up to date, because this is a field that is still in its very early stages.
That’s all for today. In the second and final part of this article, I’ll be debunking 10 myths surrounding GEO and will try to provide you with a summary of the guidelines to follow, which are as closely aligned with the primary sources as possible.
See you next time!
