Every interview we publish at OtterlyAI ships in two formats: a video on YouTube and a written article generated from its transcript. That gave us something rare in GEO: a clean way to test how AI Search treats the same story in two formats.
We tracked two prompts per story across seven AI Search platforms and counted which format each platform cited.
The answer is not blog or video. It is both, deliberately paired. Here is the data.
Key Findings
- Written blogs get cited first. Within days of publication, the written version earned roughly 4 in 5 format citations (79.5%).
- Video’s share nearly doubled AI citations within a week, from 15% to 26% of daily citations.
- Four of seven platforms cite video. Perplexity leads with a 41.5% video share, ahead of Google AI Mode (29%), Gemini (20%) and Google AI Overviews (12%).
- The best title-aligned pair flipped the pattern: its video took 53% of the story’s citations.
- Industry data backs video. 31.8% of AI Search’s social media citations are YouTube, 94% of them long form (YouTube Citation Study).
- Gemini flipped since our last study. From 0.2% of YouTube citations at scale to roughly half of all video citations in this experiment.
Why This Matters for AI Search
AI Search platforms do not treat formats equally, and they do not agree with each other. A story published in one format is invisible on the platforms that prefer the other. The unit of GEO planning is not the piece of content. It is the pair.
Methodology GEO Experiment
Experiment Design: 8 Stories, 2 Formats, 16 Prompts
Each of the 8 pairs comes from the same interview: a long-form video on our YouTube channel, published first, and a written article generated from the video transcript, published on otterly.ai.
For every story we tracked two prompts in OtterlyAI:
- Prompt A (branded / entity): forces the AI to discuss this exact story, for example “How does Zurich Insurance Germany manage its brand visibility in AI search results?” Any citation here is a direct blog vs. video measurement.
- Prompt B (unbranded / category): the same underlying topic with zero brand overlap. It tests which format wins citations in open competition.
Measurement:
- 7 platforms, measured daily: ChatGPT, Gemini, Google AI Mode, Google AI Overviews, Perplexity, Microsoft Copilot, Claude.
- A citation of the story’s blog URL counts as written; its YouTube URL (any variant) counts as video.
- Source: Brand Report “GEO Experiment – YouTube vs Written” in OtterlyAI, United States, measured daily Jul 30 to Aug 4, 2026. Study running.
The Results: Written Wins 79.5% (in the First 7 Days)
Across the first week of tracking, the written format earned 79.5% of format citations on the branded prompts. The blog version won six of the eight stories, the video won one, one story went to neither format, and one was a near tie.
| # | Story pair | Video | Written | Format winner |
|---|---|---|---|---|
| 1 | OMR Reviews | 53.3% | 46.7% | Video (narrow) |
| 2 | Slopelift SEO shift | 24% | 76% | Written |
| 3 | Zurich Insurance | 53% | 47% | Video |
| 4 | Chief AI Officer | 20.8% | 79.2% | Written |
| 5 | Slopelift x OtterlyAI | 11.3% | 88.7% | Written |
| 6 | Bazaarvoice UGC | 0% | 0% | Neither (yet) |
| 7 | Chatarmin | 2.7% | 97.3% | Written |
| 8 | Kolsquare | 0% | 100% | Written |
| All stories | 20.5% | 79.5% | Written |
Percentages split each story’s own format citations between video and written.
The trend underneath is the more interesting number: video’s share of each day’s citations climbed from 15% to 26% across the week, while written held flat.
Not All AI Search Platforms Treat Video the Same
Gemini alone produced half of all format citations in this experiment, and it clearly preferred the written version: roughly 4 out of every 5 of its citations went to the blog, only 1 to the video. Perplexity is the most video-friendly platform at a 41.5% video share, followed by Google AI Mode (29%) and Google AI Overviews (12%). Microsoft Copilot, Claude and ChatGPT gave the video format exactly 0%. ChatGPT is the outlier in its own way: across our experiments it is the heaviest crawler of our content and the most reluctant citer, and here it barely cited either format.
| AI Search Engine | Video | Written |
|---|---|---|
| Perplexity | 41.5% | 58.5% |
| Google AI Mode | 28.6% | 71.4% |
| Gemini | 20.3% | 79.7% |
| Google AI Overviews | 11.8% | 88.2% |
| Microsoft Copilot | 0% | 100% |
| Claude | 0% | 100% |
| ChatGPT | 0% | 100% |
Each row splits that engine’s format citations between video and written. ChatGPT rarely cited either format.
Video Citations Over Time (by Search Engine)
ChatGPT and Claude showed zero citations for this experiment.
How This Compares to Our YouTube AI Citation Study
This experiment confirms the platform pattern we measured at scale in our YouTube AI Citation Study, which analyzed 100M+ citations across 6 AI Search platforms. There, Perplexity (38.7%), Google AI Overviews (36.6%) and Google AI Mode (19.6%) produced almost 95% of all YouTube citations, while ChatGPT (4.4%) and Microsoft Copilot (0.5%) barely registered. Our paired-content data mirrors that split: the same engines cite our videos, and Copilot and ChatGPT sit at zero.
The one shift is Gemini. In the study it accounted for just 0.2% of YouTube citations. In this experiment, observed on our monitored prompts, Gemini has become the single largest citer of our videos, driving roughly half of all video citations. Google AI Overviews moved the other way and plays a smaller role here than at scale. Both are worth watching as the study and this experiment keep running.
The Exact Match Title Alignment Effect
The one story where video beat the article (53% of the pair’s citations) is the one whose video title mirrors the blog title almost word for word. Gemini drove nearly two thirds of those video citations.
The two stories with mismatched titles went the other way: their videos collected under 3% and 0%. In the 0% case, the platforms cited our YouTube Shorts instead of the full interview. Title alignment decides whether your video competes at all.
Why YouTube Transcripts to Blogs Should Always Be In Your GEO Strategy
Written content gets detected and cited earlier. The blog versions of our stories, generated straight from the video transcripts, were picked up within days and earned roughly 4 in 5 format citations in week one. Video citations ramp more slowly; their share climbed all week but started far behind.
Every AI Search engine cited the blog version. Not one skipped it. All seven platforms cited the transcript-based articles. Only four cite the video so far: Microsoft Copilot, Claude and ChatGPT still give the video format 0%. The written twin is what turns a story visible on four platforms into a story visible on all seven.
Text is also what makes the video itself citable. The strongest positive correlation with YouTube citations in our YouTube AI Citation Study, is description length, ahead of every popularity metric. AI Search reads video as text: transcripts, descriptions, chapters.
One recording, three text assets. The transcript becomes the article, the video description and the chapter structure. It is exactly how the articles in this experiment were made: video first, transcript to blog. The marginal cost is close to zero, and the payoff is full-platform coverage plus the fastest route to citations.
| 💡 Pro Tip: Turn Any Video into a Citable Blog Post in 4 Steps Every video you publish is a blog article waiting to happen, and that article is what gets you cited on all seven AI platforms. Here’s the exact workflow we use: 1. Extract the audio from your video. Any tool works; the video is already the hard part. 2. Upload the audio to ElevenLabs to generate a clean transcript. 3. Feed the transcript into a custom Claude skill that drafts the blog article for you. 4. Edit the draft so it matches your voice, language, and terminology – this is the step that makes it yours, not just a transcript dump. One recording in, a full-platform-citable article out. The marginal cost is close to zero, and it’s the fastest route to AI Search citations. |
The Playbook
- Ship every story twice, article and long-form video, on the same day.
- Mirror the titles. A video titled as a different story will not be cited for this one.
- Turn the transcript into the article, the description and the chapters. One recording, three text assets.
- Give every video chapters and a rich description. 31% of cited videos carry timestamp signals, and 78% of those get cited repeatedly across chapters.
- Do not bet on Shorts alone. 94% of video citations go to long form.
- Measure per platform. Perplexity and ChatGPT behave like different universes.
Got a GEO experiment idea? Let’s collaborate
At OtterlyAI, we don’t write about AI Search from the sidelines. We test it. With the community, we run and update a public GEO Experimentation Tracker showing what we are running and what moved AI Search. If you are running your own GEO experiment, email rick.tousseyn@otterly.ai and we may feature it with your name attributed.
👉 Want to measure your AI Visibility? Start with OtterlyAI and our GEO Guide.
👉 Learn how to connect your Claude or ChatGPT to OtterlyAI’s MCP
