Which Content Formats Do AI Search Engines Cite? What 610 Videos and 571 Articles Show

The video and article numbers in this guide come from one published study. For the other formats, the honest answer is that it depends on your questions: the sites an engine cites for "best payroll software for startups" are not the sites it cites for "how to file a VAT return". That is why the first step is always to look at the actual answers.

What the study measured

German finance publisher onvista turned its daily market shows into single-stock YouTube clips and written articles. The Otterly.AI and onvista study, updated September 3, 2026, tracked 125 German-language prompts from May 12 to August 26, 2026, across seven engines: ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Copilot, Gemini and Claude. The campaign covered 610 videos (140 full shows and 470 clips) and 571 articles. The study counted a citation as one prompt citing one page on one day in one engine, and recorded 2,319 of them: 1,617 for videos and 702 for articles. Here is where each format's citations came from:

• Videos: 56.1% from Google AI Overviews, 40.0% from Google AI Mode, 3.9% from ChatGPT.
• Articles: 43.0% from Copilot, 39.2% from ChatGPT, 17.8% from the other engines. The lesson is not that video beats text. It is that each engine has its own habits, so the format you pick decides which engines can cite you. These figures come from one finance publisher in one language, so treat them as a strong signal, not a law.

Short, focused clips beat full recordings

In the study, 27.2% of topic clips were cited at least once, against 8.6% of full shows. Clips averaged 4.36 citations each; full shows averaged 0.56. The likely reason is focus. A clip about one stock answers one question directly. A full broadcast covers many topics, so it is a weaker match for any single question. What to do: when you have a webinar, demo or interview, cut the parts that answer a specific buyer question into their own clips, and give each a title that states the question it answers.

Public beats unlisted

From July 13, onvista kept three clips per day public and set the rest to unlisted. The 118 unlisted clips received no measured citations. In a matched group of 33 public and 33 unlisted clips, 42.4% of the public clips were cited and none of the unlisted ones. This fits how YouTube works. YouTube's visibility help page says unlisted videos can be watched by anyone with the link, but they do not appear on your channel's Videos tab and generally stay out of YouTube search. If you host video on your own site, Google's video guidance separates a watch page, where the video is the main point of the page, from an article that simply embeds a video. Google needs a valid, stable thumbnail and recommends consistent video details. A dedicated watch page can sit alongside an article that uses the same video.

Pairing an article with a clip covers more questions

For 571 topics that had both an article and a video, 31.5% earned a citation for at least one of the two. Articles on their own covered about 24.5% of topics, videos 17.5%, and 10.5% were cited in both formats. So publishing both formats reached more questions than either alone. It also spread the work across engines: the article was more likely to show up in ChatGPT and Copilot, the clip in Google's AI answers. The study also reported that cited articles picked up their first citation after a median of two days, and cited videos after 23 days. Use that as a rough guide to when to check, not a promise. Pair formats when each one genuinely helps the reader. An article is best for sources, numbers and precise terms. A clip is best for showing a process or keeping an expert's spoken explanation. Repeating the same material twice adds little.

How Xtrusio picks the right format for each question

The study shows that format matters, but your buyers ask different questions from a German stock investor. You need the same view for your own market. That is what Xtrusio does. It shows which formats the engines cite for your questions. Xtrusio asks ChatGPT, Claude, Gemini, Perplexity, Grok, Google AI Overviews and Google AI Mode the questions your buyers ask. It keeps every full answer, every brand named and every cited page, so you can see whether an engine leans on comparison pages, Reddit threads, YouTube videos or documentation for each question. It sees what the engines search for. In Xtrusio's checks, ChatGPT and Claude search the web on every answer; Gemini and Grok decide when to search. Xtrusio records the searches ChatGPT, Claude, Gemini and Grok run, so you know the exact phrasing the engine is looking for. It plans the format and the channel. For each buyer question, Xtrusio drafts a content plan from those searches and the sites the engines cite, with the right channel for each piece: your blog, LinkedIn, Medium, Reddit, YouTube, guest posts or PR. You approve the plan. It produces the pieces. Xtrusio writes and scores articles, checks every link before approval, and publishes to your blog where it is connected. From each article it drafts LinkedIn posts, YouTube scripts, Medium essays and press releases. Link Strategy tracks each guest post, PR placement and published piece through to its live URL and shows whether an engine cites the live URL for the question it targets. It confirms the citation. A page counts as a win only when an engine's answer to the question it targets cites that exact URL in a later scan. You see results by engine, topic and buyer persona, so you know which format worked where. Xtrusio has a free self-serve plan, and a managed service with a free trial, then tiered plans (details on request). For the full process from question to published page, read our answer engine optimization step-by-step guide.

How will you know your format choice worked?

Run this test on any new piece:

1. Pick 10 questions your buyers really ask.
2. For each, record which engines answer it and which formats and sites they cite today.
3. Publish the format that matches what the engine cites, and keep it public.
4. Ask the same questions again later. Count a win only when the engine cites your exact URL.
5. Keep the pieces that were not cited in your count, so your hit rate is honest. If an article is cited but its clip is not, keep the article and check the clip: is it public, does it answer the question on its own, and does it have a clear title? For more on reporting results, see our guide to measuring GEO campaign success.

See how Xtrusio works and find out which formats the engines cite for your buyer questions.