Trust is moving outside the website – AI Search Intelligence Briefing Issue 05

01The Big One

Trust is moving outside the website

AI is becoming the middle layer between brands and the people looking for them. But nobody in that chain is taking information at face value. Not shoppers. And, it turns out, not entirely the machines either.

1. People shop with AI. Then they check its work.

Here’s the part worth sitting with.

AI isn’t replacing the information ecosystem. It’s becoming one more layer inside it.

Source:Product.ai, The Trust in AI Commerce Report, Edition 2

2. People want to know where the answer came from.

AI interfaces hand you the answer first. The publisher, expert, or review behind it can fade into the background. As AI does more of the synthesizing, the identity of the source may matter more, not less.

Source:Raptive, Sources still matter

3. A citation isn’t the same as evidence.

This one is fascinating.

Each platform leans on its own corner of the web. Among the English-language platforms, ChatGPT cited Wikipedia most. Google AI and Grok favored Reddit. Perplexity favored YouTube. The overlap at the top was surprisingly thin.

Then the researchers ran an experiment. They invented concepts that didn’t exist, published them on platforms AI already likes to cite, and watched.

Placement did most of the work. One article on a high-preference platform outperformed more than 20 matched posts on low-preference ones. A self-hosted WordPress test stayed uncited after 40 posts in a week.

None of this makes Reddit, YouTube, or Wikipedia good or bad sources.

It means retrieval preference and source trustworthiness are two different things.

AI Search IntelligenceKinetic Pencil

Source:Liu et al., From Public Posts to AI-Search Citations: Measuring the Fragility of AI Search

Why this matters

Now, people discover brands through AI synthesis. Luckily, AI users want independent evidence… but the systems doing the synthesizing have uneven, platform-specific tastes.

02The Signal

Google wants to keep the whole shopping trip

Google launched Universal Cart in the U.S., rolling out across Search (including AI Mode) and the Gemini app.

Shoppers can save products from multiple retailers into one cart, get alerts on price drops and restocks, and see relevant promo codes. When they’re ready, they check out with Google Pay or hand the cart off to the merchant’s own site. Checkout runs on Google’s Universal Commerce Protocol, and the merchant stays the seller of record.

The bigger signal isn’t the cart. It’s that the intermediary is starting to hold onto the journey.

AI Search IntelligenceKinetic Pencil

Discovery, comparison, price tracking, and the purchase itself can now happen without a string of separate visits to retailer sites.

• discovery
• comparison
• price tracking
• the purchase itself

That raises the bar for product data. Readable on a webpage isn’t enough.

It has to be structured well enough for an outside system to compare it, monitor it, and act on it.

Source:Google, An intelligent new shopping cart

03The Data Point

Delegating and double-checking aren’t opposites

Meanwhile, nearly 90% verify AI recommendations before buying.

Contradiction? Not quite.

Trust isn’t a single threshold. People hand off parts of a task while still checking the recommendation underneath. That line probably moves with price, category, and how easily the decision can be undone.

So “do consumers trust AI?” is starting to be the wrong question.

The better one: what will people let AI do, under what conditions, and where do they still want proof?

Source:Product.ai, The Trust in AI Commerce Report, Edition 2

04Connecting the Dots

Trust is becoming distributed

A company’s AI representation isn’t built by its website alone.

AI systems pull from the company’s own content, third-party publishers, reviews, community threads, and product data. Each system weights those differently. Then the person reading the answer goes somewhere else to check it.

That’s the chain now:

1. Brand claims
2. external evidence
3. machine retrieval
4. AI synthesis
5. human verification

So the strategic question shifts. Not “how do we get cited by AI?”

But “what evidence about us exists out there, how consistently does it represent us, and can both machines and people verify it?”

AI Search IntelligenceKinetic Pencil

The citation study is what makes this sharp. Being retrievable isn’t the same as being reliable. A system can strongly prefer a source without independently checking every claim it finds there.

The long-term advantage likely belongs to organizations with a coherent, corroborated, machine-legible body of evidence about who they are.

Emerging hypothesis (accumulating evidence, not yet established): as synthetic content gets easier to produce, verifiable evidence becomes the scarce thing. Original research, clear product and service data, consistent entity information, and real customer proof help people trust a brand. Increasingly, they help machines represent it accurately too.

• original research
• clear product and service data
• consistent entity information
• real customer proof

05On the Radar

Deleting it doesn’t un-cite it

One detail from the citation study deserves its own watch list.

The researchers deleted their experimental content after Day 11, then kept testing. Some systems kept citing it anyway. Caching, slow index refreshes, stale retrieval: any of them could explain it.

That matters for anyone trying to correct the record. Updating a page doesn’t mean every AI system updates what it believes about you.

Which leaves an operational question few teams can answer yet. Once stale information enters the AI layer, how do you spot it, and how do you get it corrected?

Source:Liu et al., From Public Posts to AI-Search Citations