GEO (generative engine optimisation): what it is and how to get cited by AI Overviews, AI Mode and ChatGPT

For years SEO had one goal: to climb into Google’s ten blue links. Today a growing share of searches end before those links, inside an answer written by artificial intelligence. It happens in the AI Overview at the top of the results, in AI Mode, in ChatGPT and in Perplexity. Only a few sources appear in those answers, cited by name and with a link. Generative engine optimisation, or GEO for short, is the work it takes to be one of those sources.

In this guide we bring together what we’ve learnt working on our clients’ websites and on our own: how each engine reasons, what changes compared with SEO, how to write and structure pages and how to tell whether the work is paying off.

What generative engine optimisation is

GEO is the set of practices that make content understandable, trustworthy and easy to cite for the generative AI systems that answer users’ questions. In classic SEO you work for the click; in GEO you work for the citation, meaning you aim to be the site the AI takes its answer from and shows as a source.

Other acronyms circulate around the same goal:

• GEO (generative engine optimisation): the most widespread term, coined for generative engines such as AI Overviews and Copilot.
• AEO (answer engine optimisation): puts the emphasis on conversational “answer engines” such as ChatGPT and Perplexity.
• LLMO (large language model optimisation): the more technical version, referring to language models in general.

The names change, the work is the same. In the rest of the article we use GEO.

GEO and SEO: what stays and what changes

GEO is built on top of SEO, not instead of it. AI Overviews and AI Mode draw on Google’s index, and sites that rank well already have an advantage. So these remain essential:

• keyword and search intent research: knowing what people ask is still the starting point;
• useful, comprehensive content: shallow copy gets picked neither by Google nor by a language model;
• a technically sound website: pages that are indexable, fast and stable (here’s our guide to Core Web Vitals).

Three things do change, however:

1. The unit of measurement is the passage rather than the page. Generative engines extract individual blocks of text to compose their answer. A paragraph that answers clearly and stands on its own is worth more than a long page in which the answer is diluted.
2. Questions are longer and more conversational. People writing to ChatGPT or AI Mode describe their situation (“which CRM for a 20-person company that sells abroad”) instead of typing two words.
3. Who you are matters, even beyond your own website. A recognisable author, brand mentions on third-party sites, reviews and consistent profiles all weigh on the choice of sources.

How generative engines choose their sources

Each platform has its own logic. Knowing it helps you understand where to focus your efforts.

Google’s AI Overviews

The AI Overview is the box with an answer generated by Gemini that Google shows at the top of the results, before organic links and ads, with the cited sources alongside. It has been live in the United States since May 2024 and arrived in Italy on 26 March 2025, together with other European countries.

It appears mainly on informational searches: direct questions (“how do you”, “what is”, “how much does it cost”), comparisons, explanations. It appears much less on transactional searches (“buy running shoes”) and on many local searches, where the classic results remain front and centre. To choose its sources, Google starts from the pages it already considers relevant and trustworthy for that query, so organic rankings still matter a great deal.

Google AI Mode

AI Mode, unveiled at Google I/O 2025 and then extended to other countries and languages, is a separate conversational search screen: the user asks a question, gets a detailed answer and can continue with follow-up questions.

The mechanism to understand is called query fan-out: AI Mode breaks the question down into many sub-questions, searches for them in parallel and combines the results. For a website, this means it can be cited even for a single aspect of a complex question, provided it has a section that answers that aspect well. AI Mode also shows its sources with links, and Google has already started testing ads within it: organic space is shrinking, so SEO and paid campaigns need to be planned together.

ChatGPT search

ChatGPT’s search (which began as a prototype called SearchGPT) consults the web in real time when a question requires it and shows the sources used. To read websites, OpenAI uses its own crawler, OAI-SearchBot: if your robots.txt blocks it, your pages can’t appear among the sources.

Two aspects carry more weight here than elsewhere. The first is freshness: on topics that change often, a page that hasn’t been touched for years loses ground to an updated one. The second is precision: verifiable claims, data with sources and clear definitions are picked up more readily than vague or promotional copy.

Perplexity

Perplexity was born as an answer engine: every answer is built with numbered citations that link to the pages consulted. The crawler rule applies here too (PerplexityBot must be able to access your pages), and it rewards content that answers directly, with data and sources.

What they have in common

A study by Profound that analysed 680 million citations between August 2024 and June 2025 shows differing preferences: ChatGPT tends to cite encyclopaedic content, authoritative directories and ordered lists, while Google’s AI Overviews favour pages that already rank well and add structure, up-to-date data and immediate answers.

Differences aside, the same criteria keep coming up:

• clarity: the answer is explicit, close to the question and understandable even when lifted out of context;
• authority: the author and the publisher are recognisable, and the site is cited elsewhere;
• technical readability: the page is accessible to crawlers, with hierarchical headings and structured data.

Who risks losing traffic

With generated answers, many searches end without a click (so-called zero-click searches). The most exposed sites are those that live on generic informational content: blogs answering “what is” and “how does it work”, glossaries, basic guides, Q&A portals. This is precisely the content AI summarises best.

Less exposed are sites with strong commercial intent (product and service pages, landing pages), local pages with specific information, brands that people search for by name, and those who produce original material: their own data, real cases, expert opinions. Even for the exposed sites, though, being among the cited sources brings visibility and high-quality visits, because people who click through from an AI answer already know what they’re looking for.

How to optimise content for AI

1. Answer straight away, section by section

Every section should open with the answer, in the first two or three lines, and then go deeper. Short paragraphs with a single idea, blocks that make sense even when read on their own: that’s how a passage becomes extractable. Lists, comparison tables and definitions in bold help, because they’re formats a generative engine picks up easily.

An example: instead of a generic H2 such as “SEO strategies”, write “How is artificial intelligence changing SEO?” and answer in the first sentence under the heading.

2. Headings as questions, and an FAQ section

Phrase your H2s and H3s as the questions people ask search engines, using Google’s related searches, the “People also ask” section and the questions your sales team receives. End articles with an FAQ: each question corresponds almost one-to-one with a possible prompt. Before publishing, ask yourself what five questions a reader would ask about the topic and check that the answers are in the text.

3. Use structured data

Structured data (Schema.org in JSON-LD format) tells search engines explicitly what’s on the page: who the author is, when it was updated, which questions and answers it contains. The most useful types are Article, Person for the author, Organization for the company, FAQPage, BreadcrumbList and, where relevant, Product, LocalBusiness, Event or HowTo. On WordPress, plugins such as Rank Math or Yoast handle much of this; either way, it needs validating with Google’s Rich Results Test and must match the visible content.

4. Build authority on a topic

A single isolated article is rarely enough. A cluster of linked content works better: a main page (pillar) covering the topic as a whole and several in-depth articles on individual aspects, connected with descriptive internal links. This structure fits AI Mode’s query fan-out: the more sub-questions you cover well, the more chances you have of being cited.

5. Demonstrate experience and trustworthiness (E-E-A-T)

Generative engines, like Google, prefer sources that can be identified and verified. In practice:

• sign your content with real authors, with a bio describing their specific experience and links to their professional profiles;
• include first-hand data, even small pieces: a campaign result, an observation from a project, a figure you collected yourself;
• cite authoritative sources for factual claims;
• show publication and update dates, and update key pages when something changes;
• make it clear who is behind the website: an about page, contact details, legal information.

6. Work on your reputation beyond your website

The choice of sources also depends on what’s said about you elsewhere. Brand mentions in publications and industry sites (even without a link), appearances on podcasts and at events, reviews, a complete Google Business Profile, consistent profiles on LinkedIn and in industry directories. User-generated content platforms such as Reddit and YouTube are often cited in AI answers: being there with useful contributions helps. For local searches, Google’s engines draw directly on the business profile, so up-to-date information and reviews remain decisive.

7. Avoid mass-produced content

The most common mistake is responding to AI by producing text written entirely by AI, without adding anything. A language model has no reason to cite a page that says the same things as a hundred others. The content that gets cited brings something unique: its own data, real cases, hands-on experience, a reasoned point of view. Our rule is simple: every article must contain at least one element that can’t be found elsewhere.

The technical foundations

• Allowed crawlers. Check that your robots.txt doesn’t block Googlebot, Bingbot, OAI-SearchBot or PerplexityBot. A useful note: the Google-Extended token concerns the use of content to train Gemini and has no effect on AI Overviews and AI Mode, which rely on Googlebot’s normal crawling.
• Content in HTML. Important information must be in the page text, not only in images, PDFs or elements loaded by heavy JavaScript. Use descriptive alt text for images and transcripts for videos.
• Site structure. A single H1, H2s and H3s in logical order, semantic HTML tags, important pages reachable in a few clicks, an up-to-date XML sitemap.
• Speed and mobile. A slow website, or one that’s awkward on a smartphone, is a poor landing place for people clicking through from an AI answer, and one quality signal fewer.
• Snippet control. The nosnippet and max-snippet directives also apply to AI Overviews and AI Mode: they limit how much text can be shown, for example for restricted content. Use them judiciously, because they also reduce your chances of being cited.
• llms.txt. This is a Markdown file that summarises a website’s main content for language models. It remains an unofficial proposal, and there’s no confirmation that engines use it to choose their sources. It costs little to prepare (we publish one on this site), but it shouldn’t be treated as a priority.

How to measure results

GEO is measured with tools that are partly different from those of traditional SEO.

• Google Search Console. Impressions and clicks from AI Overviews and AI Mode are included in the web search performance report, with no dedicated filter. The signal to look for is the gap: stable or growing impressions and falling clicks on the same query often indicate that an AI box has appeared.
• Referral traffic in GA4. Visits from ChatGPT, Perplexity, Copilot and Gemini arrive as referrals from their respective domains: create a dedicated channel group to track them over time and compare their conversions.
• Manual checks. Once a month, put the key questions in your industry to AI Overviews, AI Mode, ChatGPT and Perplexity. Are you cited? Who appears in your place? How is your brand described?
• Dedicated tools. Platforms such as Profound and the AI monitoring features in Semrush and Ahrefs systematically measure citation share, meaning how often you’re cited compared with competitors across a set of questions.

Alongside clicks, qualitative metrics matter too: brand mentions, the tone the AI uses to describe you, conversions generated by AI traffic, enquiries in which the customer says they found you through ChatGPT.

Where to start

1. Identify the 10–20 most important questions for your business and check who is cited in AI answers today.
2. Check robots.txt, indexing and structured data on your main pages.
3. Rewrite the opening of key sections in an answer-first format and add an FAQ to the pages that matter.
4. Add authors, update dates and your own data.
5. Set up monitoring in Search Console, GA4 and manual checks, and repeat every month.

If you’d like to do it together, our GEO service starts from exactly this analysis: which pages on your website are most exposed, where you’re already cited and where it makes sense to start. You can also book a call with us.

Frequently asked questions about GEO

Does GEO replace SEO? No. GEO builds on the foundations of SEO: indexing, content quality, authority. It adds a goal, being cited in generated answers, and some specific practices around structure, structured data and reputation.

Are AI Overviews live in Italy? Yes, since 26 March 2025. They appear mainly on informational searches and much less on transactional ones.

What’s the difference between AI Overviews and AI Mode? An AI Overview is a summary box at the top of the normal results. AI Mode is a separate conversational search mode that breaks the question down into several searches (query fan-out) and allows follow-up questions. Both cite their sources with links.

How long does it take to see results? It depends on the website’s starting authority and on the competition. On pages that already rank, the first changes in citations are usually visible within a few months of the main optimisations.

Does content written with AI help you get cited by AI? Not on its own. Generative engines reward what adds something: your own data, direct experience, real cases. AI can help with structuring and reviewing, but the value has to come from someone who knows the subject.

Should you block AI crawlers to protect your content? It depends on your goal. If you block OAI-SearchBot or PerplexityBot, your pages won’t be able to appear among ChatGPT’s and Perplexity’s sources. For most companies that want visibility, it’s better to let them through.