# How to get cited by ChatGPT and Google's AI Overviews When Google shows an AI summary, users click a result half as often. What gets a page quoted instead, what the research measured, and what I got wrong on my own site. - Author: Radwan Altaf (https://radwanaltaf.com) - Published: 2026-10-08 - Topics: Growth, AI transformation - Canonical URL: https://radwanaltaf.com/writing/how-to-get-cited-by-chatgpt-and-ai-overviews - Disclosure: Every external figure is linked to its original source. The changes described are to this site, and each one can be checked in its public pages or markup. No client detail is included. --- In this article I will explain what makes an AI answer engine quote one page rather than another, what the research has actually measured, and what I changed on this site so that it can be quoted, including two changes I got wrong. To begin with, the problem. Pew Research Center tracked the real browsing of 900 American adults through March 2025, covering 68,879 Google searches. When a search showed an AI summary, they [clicked a traditional search result in 8% of visits](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/). When no summary appeared, they clicked one in 15%. They clicked a link inside the summary itself in 1% of visits, and they were more likely to end their browsing session altogether, 26% of the time against 16%. Thus, on a search where a summary appears, the realistic prize is being the source the summary is built from, and being named as that source. A reader who sees your name next to the answer may never visit, but he or she now associates the answer with you. That is worth something, and it is the only thing on offer when the click does not happen. ## What is AI engine optimisation? AI engine optimisation, also called generative engine optimisation (GEO) or answer engine optimisation (AEO), is the work of making a page likely to be retrieved, quoted and credited by AI systems that answer questions directly, such as Google's AI Overviews and AI Mode, ChatGPT search and Perplexity. In simple words, SEO tries to get your link clicked. AIEO tries to get your sentence quoted with your name on it. The two overlap far more than the people selling GEO services tend to admit. An answer engine does not read the whole web for every question. It searches, pulls passages from the pages it finds, and writes from those. Google's own [guide to AI search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) puts it plainly: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO." In a nutshell, AIEO is mostly SEO with a stricter editor, and the editor reads one passage at a time. ## What has actually been measured? A lot of AIEO advice online is opinion. These are the findings I trust, because I read the source rather than the summary. **Specific claims get quoted more.** The paper that introduced the term, [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735) by Aggarwal and colleagues at Princeton and partners, published at KDD 2024, tested several ways of rewriting a source page against GEO-bench, a set of 10,000 queries across 25 domains. The three best were citing sources, adding quotations and adding statistics, with "a relative improvement of 30-40%" on the paper's main visibility measure. Keyword stuffing, the oldest SEO trick there is, offered "little to no improvement", and on Perplexity it did 10% worse than leaving the page alone. **Ranking helps less than it did.** In July 2025, Ahrefs [found](https://ahrefs.com/blog/search-rankings-ai-citations/) that 76.1% of pages cited in AI Overviews ranked in Google's top 10 for the query. If you have seen that figure quoted, it is out of date. After AI Overviews moved to Gemini 3, Ahrefs [re-ran the study in March 2026](https://ahrefs.com/blog/ai-overview-citations-top-10/) across 4 million cited URLs and found that only 37.9% came from the top 10. Roughly 31% came from beyond the top 100. In other words, Google now quotes passages from pages that do not appear on the first page of results, which makes what the passage says matter more, not less. **Google says there is no special trick.** Google's [documentation for AI features](https://developers.google.com/search/docs/appearance/ai-features) says: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A page has to be indexed and eligible to be shown in search with a snippet. Since 31 August 2026 there is one more condition: a [Search generative AI control](https://support.google.com/webmasters/answer/16908024) in Search Console, set to include your site by default, which you can switch to exclude it. **The click is shrinking.** Beyond the Pew finding, Ahrefs [measured in February 2026](https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/), using December 2025 data, that an AI Overview "now correlates with a 58% lower average clickthrough rate for the top-ranking page". Thus, even the page in first position gets far fewer clicks when a summary sits above it, and the citation is what is left. ## Why Paul Graham gets quoted without trying Paul Graham's site is a useful comparison, because it does none of the technical things the AIEO checklists recommend. When I fetched his essays on 8 October 2026, the pages had no meta description, no structured data, no Open Graph tags and no H1. The title of each essay is a GIF image. The robots.txt points to a sitemap that returns a 404. There is no llms.txt. Even so, people search for his sentences by name. Typing "what is a startup paul" into Google completes to "what is a startup paul graham". The answer they are looking for is the first sentence of his essay [Startup = Growth](https://paulgraham.com/growth.html): "A startup is a company designed to grow fast." That sentence answers the question in eight words and needs nothing before or after it. You could lift it out of the essay, put it in an answer box with his name next to it, and it would still be complete and still be his. That is exactly the shape an answer engine needs. Many of his essays open the same way. [How to Get Startup Ideas](https://paulgraham.com/startupideas.html) begins: "The way to get startup ideas is not to try to think of startup ideas." [How to Write Usefully](https://paulgraham.com/useful.html) contains the rule behind it: "Useful writing makes claims that are as strong as they can be made without becoming false." A strong claim is quotable. A vague one, such as "the issue is complex and there are many factors", gives an engine nothing to quote, and Paul Graham makes exactly that point about academic writing in the same essay. My take, and I hold it firmly, is that Paul Graham has been doing AIEO for more than twenty years without the term, and his essays are the best evidence that the writing does more than the markup. He wrote for readers who might give up halfway. Answer engines read the same way, one passage at a time, and keep the passage that stands up on its own. ## The idea in one sentence **Answer engines quote sentences that answer a question without needing the paragraph around them, so the most valuable AIEO work happens in the first sentence of each section, and the technical layer only helps a page that is already worth quoting.** ## What I changed on this site, and what I got wrong I will use this site as the worked example, because every change below can be checked in its public pages and markup. **1. I connected the pages to one person.** Every page used to emit its own standalone `Person` block, with nothing telling a crawler that they all described the same individual. I gave the person and the website stable identifiers and pointed every article's author and publisher at them, so the whole site resolves to one connected entity. The `sameAs` list, which links the person on this site to the same person elsewhere, went from one profile to five, and later to seven. Then I got one of them wrong. From 22 September to 7 October 2026, the Instagram link in that list pointed at `instagram.com/radwan_altaf`, a profile that is not mine. The correct handle is `radwan.altaf`. For fifteen days the markup was telling every engine that read it that I was also someone else. It was a one character difference, and no build check could have caught it, because the link was valid. It was just somebody else's. Thus, an entity graph is only as good as the least checked link in it. **2. I added FAQ markup, which Google no longer shows for most sites.** Each article's questions and answers live in the article's front matter and are emitted as `FAQPage` structured data. In August 2023 Google [announced](https://developers.google.com/search/blog/2023/08/howto-faq-changes) that FAQ rich results "will only be shown for well-known, authoritative government and health websites". This site is neither. I keep the markup anyway, for a reason that has nothing to do with Google: writing five questions with answers that stand alone forces me to restate every figure with its source, in a form that can be extracted without the article around it. The discipline is worth more than the markup. **3. I published an llms.txt file and markdown copies of every article.** The [llms.txt proposal](https://llmstxt.org/) asks sites to "provide LLM-friendly content" through a plain markdown index and markdown versions of their pages. This site has both: an `/llms.txt` generated from the same configuration the site renders, and a markdown copy of each article at its URL plus `/index.md`. I have no evidence it has changed anything, and the evidence against it is direct. Google's AI search guide lists llms.txt under things you can ignore, because "Google Search itself doesn't use them", and adds that creating one "will neither harm nor help your site's visibility or rankings in Google Search". John Mueller of Google was blunter in June 2025: ["FWIW no AI system currently uses llms.txt."](https://bsky.app/profile/johnmu.com/post/3lrshm4gggs2v) I have not found a major AI search provider that says it reads the file. The markdown copies cost nothing to keep, so they stay, but they are not a strategy. **4. I let the AI crawlers in by name, and wrote one comment wrong.** The robots.txt names each AI crawler explicitly instead of relying on the wildcard. The one that matters most for ChatGPT is `OAI-SearchBot`, which [OpenAI's documentation](https://platform.openai.com/docs/bots) describes as the crawler used to surface websites in ChatGPT's search features. It is separate from `GPTBot`, which is used for training, so a site can allow one and block the other. In the same file I allowed `Google-Extended`, and the comment I wrote next to it said it controls "Gemini and AI Overviews grounding". The AI Overviews part was wrong. Google's [crawler documentation](https://developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers) says: "Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search." AI Overviews are part of Search, so they depend on ordinary Googlebot access, indexing, and now the Search Console control described above. Google-Extended governs training and grounding for Gemini and Vertex AI, and since August 2026 Google says it is also how you limit training of the models behind Search's AI features. It has never decided whether a page appears in an AI Overview. Blocking it would not have removed this site from AI Overviews, and allowing it did not add the site to them. I only found the mistake because I checked the documentation while writing this article. The comment is corrected in the same change that publishes it. **5. I link every figure to the source I actually read.** This is the change I would keep if I had to drop all the others. When I wrote the AI-native engineering article on this site, two of the five figures I first took from search snippets were wrong. One was a "45%" that was really 66%, because the snippet had swapped two survey questions. Both errors were caught only by opening the primary source. An answer engine that quotes you passes your mistakes on with your name attached, which is a worse outcome than not being quoted at all. ## How to write a page that gets quoted **1. Get indexed, then rank where you can.** A page has to be indexed and eligible for a snippet before any of this matters, and ranking still helps, even if the March 2026 Ahrefs data shows it helps less than it did. The first job is still ordinary SEO: a page that answers a real query, with a title in the reader's words and links from elsewhere. I wrote about finding those queries in [How to get content ideas people are already searching for](/writing/how-to-get-content-ideas) and about the title in [How to write headlines that get clicked](/writing/how-to-write-headlines-that-get-clicked). **2. Answer in the first sentence under every heading.** Write the heading as the question, then make the first sentence the answer, the way "A startup is a company designed to grow fast" answers "what is a startup". The rest of the section can explain, qualify and give examples. The first sentence has to work alone. **3. Write one definition block that stands on its own.** Each researched article on this site, including this one, has a section that defines its core term, restates it in simpler words, and contrasts it with what it is usually confused with. It must make sense with no preceding paragraph, because the extractor will not include one. **4. Put the number, the source and the year in the same sentence.** A claim with no named study behind it gives an engine nothing to cite. "Pew Research Center found in 2025 that users clicked a result in 8% of visits with an AI summary" is a sentence that can be lifted whole. Adding statistics and citing sources were two of the three best methods in the GEO paper. **5. Use the questions people actually type as your headings.** Typing "how to get cited by" into Google on 8 October 2026 returned "how to get cited by ai", "how to get cited by chatgpt", "how to get cited by llms", "how to get cited by perplexity" and "how to get cited by claude". Those are the questions, in the asker's words. **6. Let the right crawlers in.** Check robots.txt for `OAI-SearchBot` and Googlebot at minimum. OpenAI's documentation says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers". A blanket rule written to stop AI training can also block the search crawlers that would cite you, so read the file line by line rather than assuming. Then open Search Console and check that the Search generative AI control is still set to include your site. **7. Keep one name and one set of profiles everywhere.** Use the same name, the same job description and the same `sameAs` links across the site and your profiles, and check every link by hand. My Instagram mistake is the example. **8. Check the answers yourself, on a schedule.** Pick ten questions your page should answer. Once a month, ask them in ChatGPT, Google and Perplexity and write down who is cited. It is crude, and the answers vary from run to run, but it is the only measurement most of us have. ## Habits that keep a page out of the answer **Keyword stuffing.** The GEO paper found it did not help, and it makes the page worse to read. **Treating llms.txt as the strategy.** Google has said its search ignores the file, so whatever it points to has to earn the citation on its own. **Bullet soup written for machines.** Some AIEO advice says to break pages into tiny chunks. Google's guide says there is "no requirement to break your content into tiny pieces for AI to better understand it", and fragments with no complete sentences are hard to quote and unpleasant to read. **Burying the answer under a story.** A long personal opening is fine for a reader who already chose your page. An engine looking for the answer moves on. **Blocking crawlers by default.** Opting out of training is a reasonable choice to make on purpose, but a rule that also shuts out the search crawlers removes you from the answers as well. ## What this does not solve I would rather be straight about the limits. AIEO cannot be measured properly yet. Answers change between runs, between users and between days. Google's Search Console [counts every link in an AI Overview](https://support.google.com/webmasters/answer/7042828) as an impression at the Overview's single position, mixed in with ordinary results, so it cannot tell you how often you were the quoted source. I am not aware of a report from ChatGPT or Perplexity that tells a publisher how often it was cited. My monthly check is a sample, not a measurement. Being cited does not mean being visited. The Pew data shows a link inside an AI summary was clicked in 1% of visits. If your business depends on page views, AIEO may protect your name while your traffic still falls, and I do not have a fix for that. The GEO paper measured its methods mainly on an answer engine the authors built themselves, on GPT-3.5 reading the top five Google results, and only secondarily on Perplexity. The direction of the findings fits Google's own advice and the Paul Graham example, but the 40% figure should not be expected to transfer as a number. Furthermore, Ahrefs' top 10 figure halved in eight months when Google changed models, so any number in this field has a short shelf life, including the ones in this article. Paul Graham's sentences travel partly because of twenty-five years of links and a name people already search for. Writing his way gives a new site the right shape for quoting. It does not give it his history. Lastly, I have no citation data from this site to show you. Everything in the section above is a change I made, not a result I measured. When I have something worth reporting, including evidence that some of these changes did nothing, I will publish it. ## Frequently asked questions **How do I get my website cited by ChatGPT?** Allow OpenAI's search crawler, OAI-SearchBot, in your robots.txt, because OpenAI says sites that opt out of it will not be shown in ChatGPT search answers. Then write pages whose key sentences answer a question on their own, with specific figures and named sources, so there is something worth quoting. **What is AI engine optimisation?** AI engine optimisation, also called generative engine optimisation (GEO) or answer engine optimisation (AEO), is the work of making a page likely to be retrieved, quoted and credited by AI systems that answer questions directly. SEO tries to get your link clicked. AIEO tries to get your sentence quoted with your name on it. **Do I need special optimisation for Google AI Overviews?** Google says no. Its documentation states there are no additional requirements or special optimisations needed to appear in AI Overviews or AI Mode. A page has to be indexed and eligible to be shown with a snippet, and since 31 August 2026 the site must also be included through the Search generative AI control in Search Console, which is on by default. **Does llms.txt help with AI search?** Not in Google. Google's AI search guide says Google Search does not use llms.txt and that creating one will neither harm nor help your visibility. Google's John Mueller wrote in June 2025 that no AI system currently uses it, and I have not found a major AI search provider that says it reads the file. **What content gets cited most by AI engines?** In the GEO study from Princeton and partners, published at KDD 2024, citing sources, adding quotations and adding statistics were the three best methods, with a relative improvement of 30 to 40% on the paper's main visibility measure. Keyword stuffing offered little to no improvement. In practice that means specific, attributed claims in sentences that make sense on their own. **Do AI Overviews reduce clicks to websites?** Yes. Pew Research Center found that Google users who saw an AI summary clicked a traditional result in 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary in 1% of visits. Ahrefs, using December 2025 data, found an AI Overview correlated with a 58% lower click-through rate for the top-ranking page. ## Rewrite one first sentence today Take the page you most want quoted, find its most important heading, and rewrite the first sentence under it so that it answers the heading in under twenty words with nothing before it. If it cannot be quoted alone, it will not be quoted at all.