What changed: from ranking to being cited
For years the goal was to put your page at the top of the results. Today an AI writes a full answer above those results — in Google's AI Overviews and AI Mode, in ChatGPT, Perplexity and Copilot — with a handful of links to its sources beside it. Whoever is named inside that answer gets the attention. Whoever is not may never be seen.
The effect shows up in behaviour. In a study published in July 2025 using real browsing data from US adults, the Pew Research Center found that users clicked a traditional result in 8% of visits where an AI summary appeared, against 15% of visits where none did. The figure covers one market and one period, but the direction is clear: the answer itself is now part of the competition for visibility.
What is generative engine optimization (GEO)?
Generative engine optimization is the work of shaping your site's content and technical foundation so that AI answer engines can find it, understand it, and choose it as a source they cite. It extends search engine optimization rather than replacing it: a page that is not indexed or not understood will not be cited by anyone.
The term comes from a peer-reviewed paper presented at KDD 2024. Its authors tested different ways of modifying content and measured the effect on how visible a source was in generative engine responses. They found these methods could raise visibility by up to 40%, and that adding citations to credible sources, quotations and statistics were among the most effective. Keyword stuffing — the old reflex — offered little to no improvement. Notably, citing sources helped lower-ranked sites far more than the sites already at the top, which means well-documented content gives a smaller site a genuine opening.
What Google officially says about AI Overviews
Google's documentation is direct: established SEO practice remains the foundation for appearing in AI Overviews and AI Mode, and there are no additional requirements or special optimizations needed. The real condition is that a page is indexed and eligible to be shown in Google Search with a snippet.
One consequence is easy to miss. Preview controls apply here too. Directives such as nosnippet, max-snippet and noindex limit what Search shows from your pages, and therefore what these features can use. A site that restricted snippets years ago for some forgotten reason may have removed itself from the answers without realising it.
A May 2025 post from Google's Search team on succeeding in AI search repeats the same themes: unique content that is genuinely useful to people rather than a copy of what exists everywhere, a good page experience, pages that Googlebot can reach, deliberate use of preview controls, and structured data that matches what is visibly on the page. As for the Google-Extended token, which governs use of content for Gemini training, Google states it does not affect a site's inclusion in Google Search and is not a ranking signal.
ChatGPT, Perplexity and Copilot each use their own crawler
The other engines do not rely on Google's index, and each has a crawler that must be able to reach you. OpenAI documents three: OAI-SearchBot, used to surface websites in ChatGPT's search results; GPTBot, which collects content that may be used to train models; and ChatGPT-User, which fetches a page when a specific user asks for it during a conversation. OpenAI says allowing OAI-SearchBot is what lets a site appear in those search results, and that robots.txt changes can take about a day to take effect.
Perplexity likewise documents PerplexityBot for surfacing and linking sites in its results, and states that it is not used to crawl content for training foundation models. On the Microsoft side, Bing explains that grounding answers in sources builds on the same crawlers and quality signals as its search index, and its webmaster tools now report how often your pages are cited in Copilot — so a site that is poorly indexed in Bing starts from behind.
The practical result is a two-part decision that should be made on purpose, not by accident. Allowing search and answer crawlers is one choice; allowing training crawlers is another, and robots.txt lets you separate them. Check your firewall and CDN settings as well: some offer a single switch that blocks AI crawlers, and it can be turned on without the marketing team ever hearing about it.
What makes a page citable
Answer engines lift passages, not whole pages. So write each section to open with the direct answer in two or three sentences, then expand. A paragraph that makes sense without the one before it is a paragraph that can travel into an answer.
Back every claim with a source the reader can open. That is what the research rewards, and it protects your reputation: an unsourced figure may be repeated in your name and later turn out to be wrong. Show publication and update dates; undated content is hard to trust as current.
Keep the basic facts about your company consistent everywhere: your name, exactly what you offer, to whom, and where. Engines build their picture of you from your own site and from the profiles and directories that mention you, and contradictions between them erode trust. Add structured data — organisation, articles, FAQs — that matches exactly what the reader sees.
In the Saudi market, write the Arabic version natively rather than translating it. Someone asking in Arabic gets an answer assembled from Arabic sources, and specialised, well-sourced Arabic content is still scarce in many technical fields. That is an opening for whoever moves first.
Common mistakes that waste the effort
The first is writing a separate version “for AI” that differs from what people read. No official guidance asks for it, and copy written for machines loses the human who will make the buying decision.
The second is producing dozens of thin, near-identical pages to catch every question. Google's guidance points explicitly toward unique, useful content, and the research found that stuffing keywords does not raise visibility.
The third is treating an llms.txt file as a shortcut. It is a community proposal for summarising a site for language models; some tools may use it, but it is not an official standard for any of the major engines, and it does not make up for good, indexable pages.
The fourth is neglecting the technical base: a page that only renders after heavy scripts run, a language version that is not linked to its counterpart, a slow site. Each one stops the crawler from seeing the content before the question of citation even arises.
How to measure your visibility in AI answers
Start with twenty questions your real customers ask before buying, in their wording and in both languages. Put each one to ChatGPT, Google, Copilot and Perplexity more than once, because answers vary between runs, and record whether you were mentioned, which page was cited, and who else appeared. Repeat monthly and compare rates rather than single runs.
Use the official tools that exist. Bing has added an AI Performance report to Webmaster Tools showing how often your pages are cited across Copilot and Bing's AI summaries, with a sample of the queries behind those citations. The IndexNow protocol lets you notify engines such as Bing the moment a page is published or updated rather than waiting to be crawled. In your analytics, watch referrals from AI assistant domains, knowing that part of the effect never shows up as a click: a reader who saw your name in an answer and searched for you directly later.
How DataWork can help
DataWork's search engine optimization service starts from the foundation answer engines depend on: crawl, index and rendering audits, a properly linked bilingual setup, a review of robots.txt, snippet controls and AI crawler access, and structured data that matches the visible content.
We pair it with our content strategy and production service: research into the questions buyers actually ask, a topic map by buying stage, native Arabic and English writing for each market, and editorial review for accuracy, safe claims and sources that can be checked. Measurement sits in our analytics service, so you can see which content earns visibility, citations and real enquiries. We do not promise a specific position in any engine, because nobody controls that, but we build the reasons that make your site a source worth citing.
Key takeaways
- There are no special requirements for Google’s AI Overviews: an indexed, snippet-eligible page on a sound SEO foundation is where it starts.
- Separate search and answer crawlers (such as OAI-SearchBot and PerplexityBot) from training crawlers (such as GPTBot) in robots.txt, and decide each deliberately.
- Citing credible sources and adding statistics and quotations are among the strongest levers for visibility in AI answers; keyword stuffing does not help.
- Open every section with a direct answer that stands on its own, and keep the facts about your company consistent wherever it is mentioned.
- Measure monthly with a fixed question set across several engines, and use the AI Performance report in Bing Webmaster Tools.
Frequently asked questions
How is GEO different from SEO?
SEO aims to rank your pages in search results; GEO aims to have your pages chosen as sources inside AI-written answers. The second rests on the first: a page that is not indexed or is technically weak will not be cited by any engine, and content structure and sourcing come on top of that.
Is SEO dead because of AI?
No. Google states that established SEO practices still apply to AI features in Search, and Bing says its grounded answers build on the same crawlers and quality signals as its search index. What changed is the measure of success: ranking alone is no longer enough; appearing inside the answer matters too.
How do I get ChatGPT to mention my company?
First make sure OAI-SearchBot is allowed in robots.txt and through your firewall, since that is what lets your site appear in ChatGPT’s search results. Then publish pages that answer your customers’ questions clearly with sources, and keep information about your company consistent across your site, profiles and directories.
Does blocking GPTBot or Google-Extended hurt my search visibility?
According to the official documentation, no. Google states that Google-Extended does not affect inclusion or ranking in Google Search, and OpenAI separates GPTBot, which is for training, from OAI-SearchBot, which is for search. You can opt out of training and stay visible in search.
Do I need an llms.txt file?
It is not required. It is a community proposal for summarising a site for language models, not an adopted standard for the major engines. There is no harm in adding one if it is accurate and kept current, but indexable pages and crawler access come first.
Sources
- AI features and your website — Google Search Central · https://developers.google.com/search/docs/appearance/ai-features
- Top ways to ensure your content performs well in Google’s AI experiences on Search — Google Search Central Blog · https://developers.google.com/search/blog/2025/05/succeeding-in-ai-search
- Google’s common crawlers — Google-Extended — Google Search Central · https://developers.google.com/search/docs/crawling-indexing/google-common-crawlers
- Overview of OpenAI crawlers — OpenAI · https://developers.openai.com/api/docs/bots
- Perplexity crawlers — Perplexity · https://docs.perplexity.ai/guides/bots
- GEO: Generative Engine Optimization (KDD 2024) — arXiv — Aggarwal et al. · https://arxiv.org/abs/2311.09735
- Google users are less likely to click on links when an AI summary appears in the results — Pew Research Center · https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
- Introducing AI Performance in Bing Webmaster Tools (Public Preview) — Microsoft Bing Webmaster Blog · https://blogs.bing.com/webmaster/February-2026/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview
- Evolving role of the index: From ranking pages to supporting answers — Microsoft Bing Search Blog · https://blogs.bing.com/search/May-2026/Evolving-role-of-the-index-From-ranking-pages-to-supporting-answers