# Getting cited by ChatGPT and Perplexity: what the data shows By Maxime Mansiet · Published November 3, 2026 To get cited by ChatGPT and Perplexity, a page first has to be reachable by their crawlers, then contain passages that are easy to reuse: sourced figures, quotes, direct answers. That's what the public data we read shows, with limits we go through below. Nobody outside these companies knows their algorithms. ## How do ChatGPT and Perplexity choose their sources? AI engines first look for pages, then a model writes an answer from what they retrieved. The founding paper on the topic describes this two-step process and reproduces it in its test bench: the top 5 Google results for each question, then an answer generated by gpt-3.5-turbo ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), KDD 2024). So your page has two hurdles: get found, then get reused. ## What does the Princeton GEO study show? The GEO study shows that changing a page's content can raise its visibility in generated answers by about 40% at best. The authors, notably from Princeton and IIT Delhi, tested 9 methods on 10,000 questions ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), KDD 2024). On their main visibility score, where the unmodified page gets 19.3 (results table of the same study): - Adding quotes from credible sources: 27.2. - Adding statistics: 25.2. - Improving how fluently the text reads: 24.7. - Citing sources: 24.6. - Using a more authoritative tone: 21.3. - Keyword stuffing: 17.7, below the original page. That goes against an SEO habit. Keyword stuffing lowers the score and an authoritative tone adds little, while verifiable additions count the most. Tested on the live Perplexity, the methods point the same way: quotes gain 22% on the main score, statistics up to 37% on the second score, and keyword stuffing loses 10% (same source). ## Do small sites stand a chance of being cited? According to this study, lower-ranked pages gain the most. When every source in an answer is optimised, citing sources raises the visibility of the site ranked fifth on Google by 115.1%, while the top-ranked site loses 30.3% on average ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), KDD 2024). This comes from an experimental setup, and nothing guarantees it would repeat exactly today. ## What are the limits of these figures? The main limits of these figures are their age and their setting. The study dates from 2023-2024, uses gpt-3.5-turbo on a simulated engine, and measures a share of the answer. It doesn't measure traffic or sales. The engines have changed since. Take the direction from it and leave the percentage. When an agency promises "+40% visibility" on the strength of this paper, ask what it's measuring. ## Is ranking well on Google enough to get cited? No, a good Google ranking isn't enough to get cited. Ahrefs, which sells SEO tools, tested 15,000 prompts in early July 2025: only 12% of links cited by ChatGPT, Gemini and Copilot appear in Google's top 10 for the same prompt, against 28.6% for Perplexity ([Ahrefs](https://ahrefs.com/blog/ai-search-overlap/), 11 August 2025). These are long-tail prompts and other methods give other numbers, but the scale is a reminder that Google rank tracking won't tell you whether AI engines cite you. Google, for its part, announces no special recipe for its AI features: "There are no additional requirements to appear in AI Overviews or AI Mode", and a page has to be indexed and eligible for a snippet ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features), read 3 November 2026). ## Which crawlers should robots.txt allow? Allow OAI-SearchBot for ChatGPT and PerplexityBot for Perplexity. OpenAI writes that sites blocking OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links ([OpenAI](https://developers.openai.com/api/docs/bots), read 3 November 2026). GPTBot is a different crawler used for model training: you can block it and stay in ChatGPT search (same source). Perplexity recommends allowing PerplexityBot, which feeds its results and which it says isn't used to train models ([Perplexity](https://docs.perplexity.ai/guides/bots), read 3 November 2026). Also check your host's or firewall's settings, which can block these crawlers without your robots.txt showing it. ## Does an llms.txt file help you get cited? No, no study we read shows any effect of an llms.txt file on citations. SE Ranking finds no impact across about 300,000 sites, 10.13% of which have the file ([SE Ranking](https://seranking.com/blog/llms-txt/), 15 October 2026). Search Atlas reaches the same conclusion on 13,911 domains ([Search Atlas](https://searchatlas.com/blog/limits-of-llms-txt-for-ai-search-citation-study/), updated 19 June 2026). Both studies come from SEO tool vendors. Google also says its AI features need no dedicated AI text file ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features), read 3 November 2026). We publish an llms.txt on klyx.fr because it costs little, and we expect no gain from it. ## Does being cited bring traffic? Not necessarily: AI summaries reduce clicks. Pew Research tracked 900 US adults and 68,879 Google searches in March 2025: people clicked a traditional result on 8% of visits with an AI summary against 15% without, and a link in the summary on about 1% ([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/), 22 July 2025). That covers Google's summaries in the US. It says nothing about ChatGPT or France. We'd count a citation as awareness as much as traffic, and track both. ## What should you do on your site? Start with the pages customers read before buying, and make each passage quotable on its own: - Answer the question in the first sentence under each heading. - Put figures with their source and date, and show your prices if they can be public. - Cite credible sources, and drop the repeated keywords. - Keep text in the initial HTML, not behind JavaScript. We do this out of caution: we have no solid public number on it. ## How do you measure citations in AI engines? You measure citations by asking the same engines the same questions before and after your changes, and counting who gets cited. Answers can vary from one run to the next, so one test proves nothing: you need several questions and repeats over time. That's the principle of our [AI search visibility offer](/en/services/ai-search-visibility): from €1,200 excl. VAT, 30 questions tested on 4 engines, 5 pages fixed, and the same test rerun at day 30 ([Klyx pricing](/en/pricing), November 2026). For transparency: Maxime Mansiet, our lead engineer, co-founded CiteMe, a tool that measures a brand's visibility in AI engines. You don't need a CiteMe subscription, and the report is yours. The [discovery call](/en/contact) lasts 30 minutes and is there to pick the questions your customers really ask.