# Which AI answer engines cite video? By Tai Nguyen, ExDarkMatter. Published 2026-10-07; updated 2026-10-07. Canonical: https://exdarkmatter.com/notes/which-ai-engines-cite-video/ **Short answer:** Perplexity and Google's engines cite YouTube often. ChatGPT, Claude and Copilot cited only the written versions in the one controlled test we found, where articles made from transcripts earned most of the citations. A creator who publishes only video gives up citations on text-leaning engines; republishing each video as an article on an owned website covers both kinds of engine. ## Which engines cite video and which favor text? Not all [answer engines](https://exdarkmatter.com/glossary/#answer-engine) treat video equally. The evidence points to a sharp division between platforms. Some search-grounded engines cite YouTube often, while other leading assistants rarely link to video at all. The distribution of video citations across the market is highly concentrated. In [a broader study of YouTube citations by OtterlyAI](https://otterly.ai/blog/youtube-ai-citation-study-2026/), the share of total YouTube citations by platform was 38.7% for Perplexity, 36.6% for Google AI Overviews, 19.6% for Google AI Mode, 4.4% for ChatGPT, 0.5% for Microsoft Copilot, and 0.2% for Gemini. Perplexity and Google's surfaces together account for the vast majority of the YouTube citations in that study. ChatGPT, Copilot and Gemini account for only a sliver. That disparity matters for how creators think about their reach. When we began our own research into long-form content, we initially assumed that video was broadly overlooked by machine synthesis, an assumption we re-examined in [we were wrong about YouTube and AI](https://exdarkmatter.com/notes/we-were-wrong-about-youtube-and-ai/). The reality is more specific: someone asking in Perplexity or Google will meet YouTube links regularly, while in the controlled test below, ChatGPT and Claude cited only written pages. ## What happened in a head-to-head test of video and articles? Comparing raw citation counts across platforms can be misleading because video and text often address different subjects. The clearer question is what happens when identical material is published in both formats simultaneously. We found one controlled study that measured this directly. In [a controlled experiment by OtterlyAI](https://otterly.ai/blog/geo-experiment-video-to-blog/), each interview was published twice, as a YouTube video and as an article generated from its transcript, and the written version earned roughly 4 in 5 format citations (79.5%). The written articles did not merely hold an edge; they took most of the citations. Crucially, [the same experiment](https://otterly.ai/blog/geo-experiment-video-to-blog/) found that all seven platforms cited the transcript-based articles. When the experiment broke down results across engines, [the OtterlyAI data](https://otterly.ai/blog/geo-experiment-video-to-blog/) showed that Microsoft Copilot, Claude and ChatGPT gave the video format exactly 0%. On those platforms, every format citation went to the written articles. In our reading, a creator who publishes only video gives those three engines nothing of theirs to quote. Even on platforms known for indexing video, written text performed strongly. On the engines that did cite video, [the same experiment](https://otterly.ai/blog/geo-experiment-video-to-blog/) found that Perplexity was the most video-friendly platform at a 41.5% video share, followed by Google AI Mode at 29% and Google AI Overviews at 12%. Even on Perplexity, the written article captured the remaining majority share. There was only a single scenario in the experiment where video surpassed the written page. In fact, [the OtterlyAI experiment](https://otterly.ai/blog/geo-experiment-video-to-blog/) reported that the one story where video beat the article earned 53% of the pair's citations, and that was the single case where the video title mirrored the blog title almost word for word. One case proves little, but it suggests that aligning titles helps an engine treat the video and the article as answers to the same question. We should be clear about what these findings represent. This was one publisher's controlled test across a specific set of topics, not a universal census of all queries. Search algorithms update continuously, and citation behaviors evolve. Still, it is the cleanest comparison we have found between video and text versions of the exact same underlying knowledge. ## Why does text still dominate across answer engines? Our reading is that the preference for text stems from how modern language models retrieve and synthesize information. Lifting a precise claim from a written page is simpler than finding it in a transcript or a video. Text also allows granular quotation. When an answer engine compiles a response, it seeks a concise paragraph or sentence that directly addresses the user prompt. A written article provides structured sentences that can be quoted verbatim with an exact hyperlink. A video file requires the engine to interpret speech transcription, identify the relevant section, and decide whether a user will accept a media link in place of a direct reading snippet. This mechanical reality reinforces our findings regarding technical optimization. As we observed in our note showing that [schema is not the AI citation lever](https://exdarkmatter.com/notes/schema-is-not-the-ai-citation-lever/), engines reward substance over superficial tags. Adding metadata cannot convince an engine to cite a thin page, and publishing only video cannot force a text-oriented engine to provide citations it is not configured to deliver. If your explanation exists only inside a video, a text-leaning engine is more likely, in our reading, to quote whoever wrote the answer down. ## What should creators do with both formats? The practical conclusion is not to abandon video. Video builds personal connection, community loyalty, and visual demonstration in ways written text cannot replicate. Google and Perplexity show that video citations are a real channel of discovery. Instead, the evidence suggests that creators should adopt a dual-format strategy. Relying solely on YouTube leaves the text-leaning engines with nothing of yours to quote. Publishing only written posts misses the video citations in Google and Perplexity. To maximize visibility across both ecosystems, we recommend three concrete actions: - **Publish both formats systematically.** Keep your video on YouTube, but generate a comprehensive, well-structured written article from the transcript for every single episode. - **Mirror titles between video and text.** Align your headlines closely so engines evaluating the two formats recognize that both assets address the same topical query. - **Host the text on your own domain.** Building a [Sovereign](https://exdarkmatter.com/glossary/#t4) archive on a site you own ensures that citations drive traffic to your brand rather than leaving your authority stranded on third-party platforms. This dual approach covers both kinds of engine. When Perplexity seeks a video demonstration, your YouTube link is available. When ChatGPT or Claude seeks a quotable paragraph, your website provides the necessary text. ## Where are the full numbers? This note focuses on format citation splits across external answer engines. In [The Dark Archive](https://exdarkmatter.com/glossary/#dark-archive), our study of 1,284 long-form YouTube channels, we analyzed where creator knowledge lives off YouTube, how many channels maintain an independent written archive, and how accessible those archives are to automated crawlers. The full breakdown of channel tiers, machine readability scores and our own citation test is in [The Dark Archive report](https://exdarkmatter.com/report/). To evaluate whether your own channel has a written archive that search and answer engines can read, [run the free Dark Matter Check](https://exdarkmatter.com/check/). ## Sources - [OtterlyAI: video-to-blog citation experiment](https://otterly.ai/blog/geo-experiment-video-to-blog/), retrieved 2026-09-30 - [OtterlyAI: YouTube AI citation study 2026](https://otterly.ai/blog/youtube-ai-citation-study-2026/), retrieved 2026-09-30