You Made 400 Videos Last Year. AI Watched None of Them.

Somebody on my team counted it up for me last quarter and the number stuck with me. Between walkaround videos, personalized customer videos from the sales floor, service tips, and the short vertical stuff we cut for social, our store put out somewhere north of four hundred videos in a year. Real production. Real hours. A used car manager standing in the rain in February talking through a certified X5 because that's what it takes now.

Then I did the thing I always do, which is ask an AI assistant a question a customer would ask and see whether any of that work shows up. "What should I check on a used BMW X5 before buying it." We had made that video. It was good. The answer I got back cited two magazine articles and a forum. Four hundred videos, and as far as the machine is concerned, we never said a word about it.

The engines don't watch. They read what's around the video.

This is the part I had backwards for a while. I assumed that as AI got better, it would eventually just watch our videos and understand them, the way it can now look at a photo. And it's getting there in a narrow sense. But that's not how the answer actually gets built today. When an assistant pulls video content into an answer, it's leaning almost entirely on the text wrapped around that video — the title, the description, the transcript or captions, the chapter markers, the page it's embedded on.

So the question isn't really "can AI watch my video." It's "did we give the machine anything to read." And for most dealership video, the honest answer is no. Our titles are things like "2023 X5 xDrive40i Walkaround — Stock #B24117." Our descriptions are the store's phone number and a link. Captions are auto‑generated if they exist at all. That's a video a human can learn a lot from and a machine can learn nothing from.

Meanwhile the guy in his garage with a tripod titled his video "Used BMW X5 — 5 Things to Check Before You Buy," wrote four paragraphs underneath it explaining each one, and uploaded a clean transcript. He has a fraction of our production budget and he owns the answer.

Why this one stings more than the others

Most of the AI visibility gaps I write about are things we never did in the first place. We never wrote real service content. We never described our certified process. Fine — that's work we skipped, and now we know.

Video is different. We did the work. The expensive part is done. The knowledge is sitting there, already spoken out loud by somebody at the store who actually knows the answer. We just left it locked in an audio track nobody transcribed. That's not a content problem, it's a packaging problem, and packaging is cheap.

There's a second cost too. Video is where the genuinely dealership‑specific stuff lives — our recon standards, how our service drive actually works, what our loaner policy really is, the way our people talk. That's exactly the distinctive, first‑person material AI has almost none of about any dealership. It's the most citable content in the store and it's the least readable.

What I'd fix first

Start with transcripts, because that's the whole ballgame. Whatever platform you're on will generate one, and the auto version is usually decent enough to clean up in a few minutes. Get it onto the page where the video is embedded, as actual text on the page — not just as a caption file the player loads. If you do nothing else from this list, do that one.

Then rewrite titles as questions a person would ask. "2023 X5 xDrive40i Walkaround, Stock #B24117" becomes "What's the difference between the X5 xDrive40i and the M50i?" The stock number matters to us. The question matters to the customer and to the engine trying to match one to the other.

Write descriptions like short articles. Three or four real paragraphs covering what the video says, in text, under the video. Not the phone number and a link. If the video answers a question, the answer should also exist in writing directly beneath it.

And put the video on your own site, not only on the social platform. A walkaround embedded on the VDP with a transcript under it strengthens the page the engine is already trying to read. The same video living only on a social feed is doing almost nothing for your store's visibility.

Pick one video and try it this week

Take the single best evergreen video your store has made — not an inventory unit, something that answers a real question — and give it the full treatment. Question‑style title, real written description, clean transcript on the page. Then ask an assistant that question in a month and see whether anything changed.

If you want to see which shopper questions your store is currently missing entirely, the free AEO check at aeowhisperer.com runs them through ChatGPT, Gemini, and Claude and shows you how you come back. Odds are good you'll find a few where the answer is sitting in a video you already made.