# How do you write content AI quotes? Answer first, standalone paragraphs, sourced figures > Writing rules for quotable content: give the answer in the first sentence, paragraphs that stand alone, sourced figures, an FAQ and server-rendered HTML. Published: 2026-10-07 Content that AI quotes is content whose sections stay accurate and clear even when read without the rest of the page. To get there, the answer is given in the first sentence, every paragraph opens with its subject and reads on its own, and figures are written with their unit, date and source. The page is also delivered as HTML generated on the server, in a form a crawler can read without running JavaScript. This post goes through these rules one by one, with the sources that say why they work. None of them is a guarantee of being quoted: whether an engine quotes you can only be seen by measuring. ## What does the research say, and what doesn't it say? The most cited study in this field is the [GEO paper](https://arxiv.org/abs/2311.09735) by Aggarwal and colleagues, published at KDD 2024. The paper builds its own question set, which it calls GEO-bench, and for each question picks one of the top 5 sources returned by Google at random and rewrites it with different writing methods. It then measures how that source's visibility in the generated answer changes. The condition on the result matters as much as the result itself. The paper reports that, on its own benchmark and for sources already pulled into the answer, adding statistics and quotations increased a source's visibility by between 30% and 40% in relative terms. It measures keyword stuffing as bringing almost no improvement. In practice this condition says: good writing can help a page the engine already reads take up more room in the answer. It does not bring into the answer a page the engine never found or could not read. That is why the rules below fall into two groups: the page being readable, and being quotable once it is read. > [!UYARI] > A sentence that says "you will be quoted 40% more with this writing technique" promises something the research does not say. The measured increase applies to sources that were already in the answer and to a single benchmark. ## Rule 1: Give the answer in the first sentence The first paragraph of a page should answer the question in the title directly. An opening such as "Today the digital world is changing fast" carries no information worth quoting. The sentence "Corporate payroll software is software that calculates salaries, bonuses and statutory deductions in one place", on the other hand, can be quoted on its own. Putting the answer first does not lose the reader either. A reader looking for detail keeps going; a reader looking only for the answer finds it in the first sentence. The first paragraph of this post was written by this rule too. ## Rule 2: Open every paragraph with its subject and make it stand alone When a paragraph is quoted, the reader sees it without the paragraphs around it. References such as "this method", "as mentioned above" or "these" become meaningless at that moment. Open the paragraph by saying what it is about: for example, "The most often overlooked feature of payroll software is..." Writing the subject explicitly matters for your brand as well. In our own measurement, across the 60 analyses we ran up to 3 August 2026, 341 (12.4%) of the 2,748 responses that cited the brand's own site did not mention the brand's name anywhere in the answer text. The measurement covers ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews; the brands are our own client and comparison brands, so not a market sample. We determined whether the name appeared with automatic name matching; namesakes and spelling variants can affect the result. This number does not tell us why. One possible reason has to do with writing: if a page talks about "we" and "our company", the engine may take the information and pass it on without naming its owner. Writing the brand's name once in the paragraph that carries the information is a simple precaution against that possibility; we have not measured its effect separately. We cover the difference between being read and being mentioned separately in our post on [the difference between brand mentions and source citations](/en/blog/marka-anilmasi-ve-kaynak-gosterilmesi-farki). ## Rule 3: Write figures with their unit, date and source A figure quoted without saying what it measures turns into false information. The sentence "35% faster" cannot be defended, because it is unclear compared with what, when and who measured it. An e-commerce site could write the same claim in a measurable way like this: "The average delivery time we measured across 120 orders in September 2026 was 2.1 days." A good figure sentence has four things in it: | Element | Example | Why it is needed | |---|---|---| | What was measured | average delivery time | The meaning of the figure | | Unit | days | Comparability | | When, and how large the sample | September 2026, 120 orders | Its currency and its weight | | Who measured it, or the source | our own records; or a linked source | Verifiability | Two of the methods the GEO paper tests are close to this rule: adding statistics and citing sources. However, the paper measures visibility in the answer, not whether the numbers are correct. An incorrect figure harms your brand even if it is quoted. {{cta}} ## Rule 4: Add a frequently asked questions section A frequently asked questions section turns the buyer's real questions into headings and gives each one a short answer that reads on its own. This format naturally applies all three rules above: the question in the heading, the answer in the first sentence, the paragraph standing on its own. Choose the questions from real sources, not guesswork: questions that reach customer service, objections the sales team hears in every meeting, what people type into the search box. A question such as "Why should you choose us?" is not a buyer's question; "How many days does setup take?" is. > [!NOT] > Google [states clearly](https://developers.google.com/search/docs/appearance/ai-features) that no special structured data or separate "AI file" is needed to appear in AI Overviews and AI Mode. The value of a frequently asked questions section comes not from its markup but from the text read on the page. ## Rule 5: Deliver important information in server-rendered HTML The precondition for being quoted is being read. According to the [analysis](https://vercel.com/blog/the-rise-of-the-ai-crawler) Vercel published on 17 December 2024, based on traffic across its own network, none of the major AI crawlers ran JavaScript: this includes OpenAI's OAI-SearchBot, ChatGPT-User and GPTBot crawlers, Anthropic's ClaudeBot and PerplexityBot. The ChatGPT and Claude crawlers downloaded JavaScript files but did not run them. The same analysis notes that Gemini uses Googlebot's infrastructure and can process JavaScript. That means one more difference between engines: ChatGPT may be unable to read a page that Google can read. Vercel's recommendation follows from this: the main content, information such as titles and descriptions, and navigation links should be generated on the server; content generated on the client side should be used only for nonessential extras. This analysis is dated December 2024, and crawler behaviour can change. The safest assumption is still this: prices, product specifications, contact information and the answers to frequently asked questions should be present as text in the page's first HTML response. ## Rule 6: Check your preview settings The only basic requirement for Google's AI features is that the page is indexed and can be shown with a preview in search results. On the same page, Google writes that the `nosnippet`, `data-nosnippet`, `max-snippet` and `noindex` settings limit how your content can appear in AI features. These settings can be used deliberately, but they are often left over from a template or an old plugin. If all of your product pages carry `nosnippet`, even the best-written paragraph cannot be quoted in Google's AI Overviews. ## Quotable content checklist Before publishing, ask these questions for every page: 1. Does the first paragraph answer the question in the title in two or three sentences? 2. Does every paragraph open with its subject and make sense when read without its surroundings? 3. Is the brand's name written at least once in the paragraph that carries the information? 4. Does every figure have its unit, date and source next to it? 5. Is there an unmeasured figure or an unverified claim? If so, remove it. 6. Is there a frequently asked questions section made up of buyers' real questions? 7. Does important information such as prices, specifications and contact details appear in the page's HTML without JavaScript running? 8. Is there a `nosnippet` or `noindex` setting on the page that you are not aware of? 9. Are AI crawlers blocked in your robots.txt file? 10. Did you measure with the same questions before and after the change? The first six items are about writing, the next three about access. To check the access items on your site by hand, our [AI visibility checklist](/en/kontrol-listesi) explains each check together with how to fix it; our [free scan](/en/tarama) runs the same checks automatically for your domain. ## After writing: measure to see The effect of these rules on a page is seen by measuring, not by guessing. For our own product, on 27 August 2026 we asked five engines 26 Turkish-language questions in the GEO category. Of the 116 measurable responses, 108 showed at least one source, but our product was mentioned in none of them; 9 observations could not be measured and were left out of the count. This is one category and one day, but it shows this much: the engines read sources in this category, and ours was not among the pages they read. After a result like that, the job is to rewrite the pages by these rules and run the next measurement with the same questions. If the questions change, the difference may come from the questions rather than the writing. Our guide [how to measure brand visibility](/en/blog/yapay-zeka-marka-gorunurlugu-nasil-olculur) explains step by step how to set up the measurement. ## Frequently asked questions ### Is structured data required for quotable content? No. Google writes that no special structured data or additional file is needed to appear in AI Overviews and AI Mode; the basic requirement is that the page is indexed and can be shown with a preview. Structured data can still be useful if it is consistent with the visible information on the page, but it does not replace the text. ### Does the 40% increase in the GEO paper apply to my site too? Not directly. The paper reports that, on its own benchmark and for sources already pulled into the answer, adding statistics and quotations increased visibility by up to 40%. It says nothing about a page the engine did not find or could not read; you can only see the effect for your site by measuring before and after with your own questions. ### Does long content or short content get quoted more? We do not recommend a length without being able to point to a reliable measurement on the subject. What matters is not length but that each section makes sense on its own: a long page can answer many questions if every paragraph is written to stand alone. A short page full of references such as "this" and "above", by contrast, loses its meaning when a single paragraph is quoted. ### My page runs on JavaScript; can AI read it? According to Vercel's December 2024 analysis, the crawlers of OpenAI, Anthropic and Perplexity did not run JavaScript, while Gemini could run it through Googlebot's infrastructure. If important information loads only through JavaScript, some engines may not see it. You can check this yourself by looking at the page's source HTML, or by turning JavaScript off and on. ## Sources - [GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024, arXiv:2311.09735)](https://arxiv.org/abs/2311.09735), arXiv / KDD 2024, 2024 - [The rise of the AI crawler](https://vercel.com/blog/the-rise-of-the-ai-crawler), Vercel, 17 December 2024 - [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features), Google Search Central, updated 10 December 2025 - Responses that cite the brand's own site but do not name the brand, 2026-08-03: 2,748 responses citing the brand's own site in the 60 analyses up to 3 August 2026 (recounted on 7 October 2026); ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews; our own client and comparison brands, not a market sample; name matching is automatic - GeoMono's own visibility: Turkish-language questions in the GEO category, 2026-08-27: 26 questions, 5 engines, 130 planned observations; 116 measurable responses, 9 observations could not be measured; one category, one day (https://geomono.com/tr/geo-nedir)