AI Search Now Trusts Video Before It Trusts Your Website

For the last two years, the advice on getting cited by AI search has focused on writing. Clear structure, direct answers, FAQ blocks, citable facts on your own website. That advice is still correct. It is no longer complete.

New research from Adweek, tracking citation patterns across ChatGPT, Google Gemini and Perplexity, found that YouTube's share of AI citations has jumped from 18.9% to 39.2%, a 107% increase. Over the same period, Reddit's share fell from 44.2% to 20.3%, a 54% drop. YouTube is now cited roughly 40% more often than Reddit across the three platforms studied, and Gemini in particular favours it, citing YouTube in around 48% of relevant answers.

The pattern that mattered most in AI search a year ago, first-hand community discussion, is being overtaken by a different one: video with a real voice attached to it.

Why video is winning the citation race

The shift is not really about YouTube as a platform. It is about what large language models can now do with video. Transcripts, captions, chapter markers and structured descriptions give AI tools a text layer to parse and a person, tone and context to attribute the answer to. A well-produced video effectively hands an AI model two forms of evidence at once: what was said, and who is saying it.

That second part is doing more work than it looks like. AI tools are being asked to make a judgement call every time they choose a source: is this trustworthy enough to repeat as an answer? A named person on camera, with a channel history and a recognisable voice, is easier for a model to treat as a credible source than an anonymous forum post or a page of unattributed marketing copy.

This is part of a wider move researchers are calling search everywhere optimisation. Visibility is no longer decided by one algorithm. It is split across Google, TikTok, YouTube, Reddit and a growing list of AI chat tools, each weighing sources differently, and each increasingly rewarding content that demonstrates a real person stands behind it.

The money is already moving

Budgets are catching up with the citation data. The IAB's creator economy report puts US creator advertising spend at $37 billion in 2025, up 26% year on year, a rate close to four times faster than the media industry's overall growth. The same report projects $44 billion in 2026. That is not a niche influencer-marketing line item any more. It is becoming a core channel budget, sitting alongside paid search and paid social rather than beneath them.

Put the two data points together and the direction is hard to argue with. AI tools are citing creator and video content more. Advertisers are funding creator and video content more. Brands whose content strategy still stops at the blog post are optimising for a smaller and smaller share of how they actually get found.

What this does not mean

It does not mean your website has become irrelevant. It remains the place AI tools check to verify specific facts about your business: what you charge, where you operate, what you actually offer, and whether your claims are backed by schema and structured data. We have covered that groundwork in how to appear in AI search and LLM results and generative engine optimisation, and none of it stops being necessary.

It also does not mean every business now needs a YouTube channel with a content calendar and a studio budget. What it means is that a business with zero video presence is now missing one of the two evidence types AI models look for before they will repeat what you say as fact.

What to change in your content plan

Three practical shifts follow from the data, without requiring a full production department:

1. Put a face and a voice on your best written content

Your highest-performing article, FAQ or explainer is the natural first candidate for a short video version. It does not need reshooting from scratch. It needs someone credible on camera saying the same thing in their own words, with a transcript published alongside it.

2. Treat transcripts and chapter markers as required fields, not extras

A video without a clean transcript and timestamped chapters gives AI tools far less to work with. This is a production checklist item now, on the same level as captions were five years ago.

3. Make the presenter identifiable, not anonymous

Named, consistent presenters build the channel history that makes a source look citable over time. A rotating cast of unnamed voiceovers does not build the same signal, even if the information is accurate.

Trust is still the point

Underneath the platform statistics, this is the same story we keep coming back to in brand work. Trust used to be won through polish. It is increasingly won through evidence: who said this, what is their track record, and can it be checked. Written content proves your facts are correct. Video, done properly, proves a real person and a real business are willing to put their name behind them.

The businesses that treat this as a production checklist item, rather than a strategic one, will keep publishing content that AI tools have less and less reason to trust.

If you need help turning your existing expertise into content that AI search, and actual humans, are more likely to trust, talk to Ayuda.

Sources and further reading

• Adweek/DigitalApplied: YouTube overtakes Reddit in AI citations
• IAB: 2025 Creator Economy Ad Spend & Strategy Report
• B2the7: Marketing trends, search everywhere optimisation