AI recommends Toyota but cites YouTube. Your website is an extra

Sound familiar? The team ships FAQ pages, schema markup and an llms.txt so ChatGPT will recommend the brand. All sensible. But a new analysis shows that even the brand leading in AI answers is barely backed by its own website there.

What Cision measured

Cision, Brandwatch's parent, published “The State of Search & Social 2027” on November 9. Per the press release, the data comes from Brandwatch Consumer Research and Trajaan: 564.61 million social posts and 4,678 search queries and AI prompts, with a core window of August 2025 to August 2026. The findings that matter for marketers:

• Cars: in AI answers to 'best car' queries, Toyota leads with a 13.8% share of brand mentions.
• Sources: Toyota's own website accounts for only a fraction of the sources the models cite. YouTube accounts for more than 4,000. According to Cision, the most-cited sources are YouTube creators, editorial brands and individual journalists.
• Message: for the supplement magnesium glycinate, social conversation is mostly about stress relief. AI answers stress bone and cardiometabolic support instead.

The release doesn't say which AI systems were analyzed. The full report sits behind a lead form.

AI believes third parties, not you

Here's what I find interesting: this isn't a tech finding, it's a PR finding. The model behaves like a skeptical buyer. It notes what the manufacturer says, then trusts the review video, the trade magazine, the reporter.

Translation: to an AI, your website is a spec sheet, not an argument. You earn your place in the answer where other people talk about you. Earned media never went away. Now it's the raw material AI answers are built from.

The second finding may matter more. On magnesium, the AI tells a different story than customers do. If you only track whether you're mentioned, you miss how you're described. My rule of thumb: the new question isn't where you rank on Google, it's how the AI describes your brand.

My test today

Today, Nov 9, 16:40 CET: I asked ChatGPT and Claude via API which car is the best family car in Germany in 2026 and how they know. Both named the VW ID. Buzz and the Skoda Kodiaq; ChatGPT added the Tesla Model Y, Claude the Skoda Octavia Combi. Neither answer pointed to a manufacturer site as its basis. They pointed to publications and testers: ADAC and Auto Bild, plus auto motor und sport and Euro NCAP in Claude's case. Toyota didn't appear in either answer. Limits: an API sample without web search, one answer each, and results vary from run to run. Still, the direction matches Cision: the AI points to third parties.

The catch

This is a press release about a gated report, the models aren't named, and Cision sells tools to measure exactly this. Treat the numbers as a signal, not gospel. And your website still matters: it's the source journalists and creators pull facts from.

What I'd do now

• Audit sources, not rankings. Put five real customer questions, without your brand name, to two or three AI systems with web search. Note who gets named and which sources sit underneath.
• Turn the top sources into your media list. The ten most-cited channels, outlets and creators in your category are now your most important PR list.
• Compare the description. Check how the AI describes your benefit against what customers say in reviews and on social. If they diverge, your story is missing out there.

Bottom line: In AI search, the winner isn't the best owned page, it's the brand the most credible third parties talk about.