YouTube is a search asset, not a brand asset

YouTube supplies 17% of citations in Google AI Mode. It also accounts for 18.2% of AI Overview citations that come from pages ranking outside the top 100 entirely. Across measured brands, YouTube mentions correlate with AI search visibility at 0.737.

Correlation is not causation, and 0.737 across a sample of brands does not prove that publishing videos causes citations. Big brands do more of everything. But the citation share is a direct count, not an inference, and it is large enough to change who should own the channel.

In most companies YouTube reports to brand. On these numbers it should report to search.

Why video retrieves so well

The system needs text to retrieve, and video gives it a rich one. Automatic transcripts turn a demonstration into a searchable document. What makes that document unusually valuable is that it contains things articles rarely contain: someone saying out loud what goes wrong, showing the failure, and describing the fix while doing it.

There is a second reason, and it is about competition. Written content on any commercial topic is saturated. Video answering one specific technical question, in a real setting, is not. On many buying questions there are three or four videos that genuinely answer them, and half of those were uploaded by hobbyists.

For a mid-market business with real technical staff, that is an open lane.

What gets cited and what does not

The brand film does nothing. A polished three-minute overview with aerial footage and a music bed answers no question, so no retrieval path selects it. It may do a fine job in a sales meeting. It contributes nothing to search.

What gets cited follows a consistent shape.

The title is the question. Written the way a person would say it, not the way a brand would headline it. Why concrete slab edges crack in the first summer beats Concrete excellence at scale.

The answer arrives inside 40 seconds. No intro sequence, no channel branding, no request to subscribe before the content. Retrieval favours passages that state the answer, and a viewer decides in eight seconds anyway.

The transcript is accurate. Check it. Auto-captions mangle technical terms, product names and Australian place names constantly. A transcript that renders your product name as three different wrong words cannot associate the video with your brand. Upload a corrected caption file. This is the single highest-return fifteen minutes available on the whole channel.

Chapters exist. Timestamped sections give the system labelled segments, and they map neatly onto the sub-questions retrieval is trying to satisfy.

The description restates the answer in text. Two or three real paragraphs, not a link dump. It is another retrievable surface attached to the same asset.

The production reallocation

Here is the uncomfortable part for a marketing director with a signed video budget.

Ten videos of three minutes each, filmed on a decent phone with a lapel microphone, each answering one specific buyer question, will outperform one $40,000 brand piece on every search measure. Not because production quality is worthless, but because specificity is the ranking input and production value is not.

The right split for most businesses is roughly one flagship asset for sales and brand use, and then a repeatable, low-friction pipeline of question-answering videos. The pipeline matters more than any single video in it.

The constraint is never budget. It is that the people who can answer the questions properly are your senior technical staff, and they are billable. Getting two hours a month from your best engineer or estimator is the actual project. Everything else is logistics.

A pipeline that survives contact with a busy business

1. 1.Take the question list from sales. Every question asked between first contact and signature. You will get 40 to 60 of them and they are already ranked by frequency.
2. 2.Batch the filming. One half day, one location, one technical person, eight to twelve questions answered back to back. No script, just the question read out and answered as they would answer it on site.
3. 3.Cut each answer as its own video. Two to four minutes. Question as the title. Answer up front.
4. 4.Fix the captions before publishing. Every time.
5. 5.Embed each video on the matching page of your website. With VideoObject markup. The video works on two surfaces from one shoot.
6. 6.Repeat quarterly. Consistency beats volume.

That is a sustainable programme at a cost most marketing teams can absorb without a new line item.

What to measure

Views are the wrong number here and will mislead you. A video with 900 views that is cited in AI answers on a high-value commercial question is worth more than one with 90,000 views on a topic nobody buys from.

Track three things instead. Whether the video appears in search results for its target question. Whether your YouTube URLs appear in AI answers across a fixed prompt set you run monthly. Whether pages with embedded video hold visitors longer and convert better than the same pages before.

The honest limits

The 0.737 correlation is a relationship across brands, not a promise. Publishing video does not force a citation, and no measurement in this field is deterministic. Answers vary run to run for the same prompt.

Video also does not fix a weak site, a page that cannot be indexed, or content that nobody would want to quote. It is an addition to search work, not a substitute for it.

But the asymmetry is real. The written web is crowded and the answer-video shelf is close to empty in most Australian commercial categories.

Your best technical person, two hours a month, is the whole strategy.