Google's "AI Mode" Needs a Warning Label

Google launched AI Mode today with the full weight of the world's largest advertising company behind it. Billions of users will see the label "AI" and assume it means something. It doesn't. What Google shipped is a cached text index wrapped in a conversational interface, and calling it AI is doing real damage — to users who trust the answers, to companies building actual AI infrastructure, and to the credibility of the entire industry.

Every financial product carries a disclaimer. Every medical device carries a warning. Google is shipping a product that presents scraped, unsigned, unverified text as factual analysis — to millions of people making real decisions — and there is no warning at all.

The Test

I ran a simple experiment. I asked Google AI Mode to use origin.rootz.global — a public SEC data registry I operate — to tell me about a company.

origin.rootz.global has served over 125,000 AI requests. It has a sitemap. It has a .well-known/ai discovery file. It responds to plain HTTP GET requests. No authentication required. It is a public website on the open internet.

Google's AI told me my site "does not exist as a registered, operational public domain."

I pushed back. It admitted it cannot make HTTP requests, cannot call APIs, cannot query structured endpoints, cannot use the Model Context Protocol, and cannot verify any data source. Its solution:

Paste the data into the chat window yourself.

An AI search engine asking a human to be its API. That is not artificial intelligence. That is a search bar with better copywriting.

What Real AI Access Looks Like

I gave the exact same prompt to Claude, which has direct HTTP access and MCP tool connections:

"Give me a detailed analysis of Applied Materials and how it is meeting the claims in its conference calls. Use origin.rootz.global."

What was delivered	Google AI Mode	Claude + Origin
API calls to origin.rootz.global	0 ("does not exist")	11 calls, full data
Earnings transcripts analyzed	0	56 transcripts, 1,042 facts
Management claims scored	0	6 claims with verdicts
Confidence scoring	None	0.73–0.95 per fact
Data provenance	None	SHA-256 leaf hash per record
Financial data source	Scraped from Yahoo Finance	SEC EDGAR direct via API
Forward guidance detail	3-sentence summary	Quarter-by-quarter with ranges
Stock price accuracy	Confused tickers ($0.41 NCPL vs AMAT)	$431.20, 30-day OHLCV
Total research cost	Unknown	~8,600 tokens

Claude pulled 1,042 verified facts from 56 earnings calls, scored six specific management claims against actual financial results (revenue trajectory, GAA revenue targets, packaging growth, margin expansion), delivered 30 days of price history, and attached a cryptographic provenance hash to every data point. In under 60 seconds. From a public API that Google's "AI" said doesn't exist.

Read Their Own Marketing

Google's product page for AI in Search says this:

"With generative AI features like AI Overviews and AI Mode, you can bring your toughest questions, right to Search. Using a custom version of Gemini, these features can help you with planning, researching, brainstorming new ideas and more."

"AI Mode — Easily dig deeper and ask follow-up questions with our most powerful AI search experience."

"Generative AI features connect you to high quality information in helpful formats."

"High quality information" — from a product that returns unverified scraped text with no provenance. "Our most powerful AI search experience" — that cannot make an HTTP request to a public website. "Bring your toughest questions" — and get told that a live, indexed data service "does not exist as a registered, operational public domain."

These are not technical caveats buried in a terms of service. These are the marketing claims on the product page. The gap between what this copy promises and what the product delivers is not subtle.

The Brand Damage Is the Real Story

The comparison is interesting. The damage is what matters.

Google is the most powerful marketing machine on Earth. When Google puts the word "AI" on a product and ships it to billions of users, the market takes that as the definition. If Google's AI returns unverified scraped text with no provenance, users conclude that is what AI does. If it hallucinates ticker symbols, users conclude AI hallucinates. If it cannot read a public website, users conclude AI cannot access data.

Google is defining AI down to the level of its own product. Every company building real AI infrastructure — tools that actually verify data, call APIs, use protocols, maintain provenance chains — is now fighting against a market perception set by the largest company in the room shipping the weakest product.

This is not a technology problem. Google has Gemini. Google has the engineering talent to build an AI that makes HTTP requests and verifies sources. They chose not to. They chose to wrap their existing search index in a chat interface and call it AI, because it was faster to ship and it protects their advertising model.

The result: a product that actively misleads users about what AI can do, while devaluing the work of everyone who is actually building it.

Scoring Google's Own Claims — The Same Way We Score Any Public Company

I run a service that tracks what executives say in earnings calls and measures it against what actually happens. I just did it for Applied Materials — 1,042 facts, 56 transcripts, six specific claims scored. Let's apply the same methodology to Google.

Here is what Alphabet (GOOGL, $387.35, $3.85 trillion market cap, 190,820 employees) tells its investors about AI, extracted directly from their SEC filings on origin.rootz.global:

The compensation question. Alphabet's EPS progression — $80 → $84 → $88 → $96 per share across 2025 — directly drives executive compensation tied to stock performance. The AI narrative is a material component of Alphabet's $3.85 trillion valuation. Every time leadership tells investors that AI is driving growth, that claim supports the stock price. That stock price determines compensation.

This is the product backing those claims: a consumer AI tool that cannot access a public website, cannot verify a data source, confuses ticker symbols, and asks the user to paste data into the chat.

This Is a Securities Question

I am not making a legal accusation. I am applying the same methodology to Google that we apply to every other public company: read the transcripts, extract the claims, compare them to the product.

The FTC exists to address deceptive product labeling. When a company labels a product "AI" and that product cannot perform functions that the label implies — like accessing a public website — that is a product labeling question.

The SEC exists to ensure that material claims made to investors are accurate. When a public company tells investors that AI is driving $175-185 billion in capital expenditure and transforming its core product, and the AI product it ships to consumers cannot make an HTTP request, the gap between those claims and that product is a question analysts should be asking about.

We do not need new regulations. We need the existing ones applied.

When a pharmaceutical company ships a product that does not do what the label says, there are consequences. When a financial advisor provides unverified data that causes harm, there are consequences. When the world's largest search company ships a product labeled "AI" that cannot read a public website, returns unverified data with no provenance, confuses basic facts, and presents the results as authoritative analysis to billions of users — there should be, at minimum, a warning label on the product and a hard question in the next earnings call.

What Is Actually Required

The internet already has the infrastructure for AI to work correctly:

• Discovery: The .well-known/ai standard tells any AI client what data a site offers and how to access it
• Protocol: MCP (Model Context Protocol) and plain HTTP let AI agents query structured endpoints directly
• Provenance: Cryptographic signing lets every answer carry proof of where the data came from
• Verification: Users can confirm that what the AI told them actually traces back to a real source

None of this is theoretical. It is running in production today. I operate one such service. There are many others. The plumbing exists. The data exists. The protocols exist.

Google chose not to connect to any of it. Instead, they shipped a search engine with a new label and let their marketing team do the rest.

The Bottom Line

I asked Google's AI to read a public website I operate. It said the site doesn't exist. I asked a properly connected AI the same question. It made 11 API calls, pulled 1,042 verified facts from SEC filings, scored management claims against actual results, and delivered the analysis with cryptographic provenance — in under a minute.

Here is the irony: the same methodology I used to verify Applied Materials' earnings claims against their actual product delivery is the methodology that should be applied to Google. AMAT told investors Gate-All-Around would generate $2.5 billion in revenue. We can check that against real shipment data. The verdict: on track. Google told investors it is an AI leader. We can check that against the product it shipped today. The verdict: the product cannot make an HTTP request.

Google did not ship AI. Google shipped a product labeled "AI" that cannot do what the label implies, backed by enough marketing muscle to convince the world that this is what AI is. That label drives their stock price. That stock price drives executive compensation. And the product behind all of it told me my own live website "does not exist."

That is not a technology failure. It is not a competitive shortcoming. It is a gap between what a public company claims to investors and what it actually delivers to users. We have words for that gap. We have regulators for that gap. The question is whether anyone is paying attention.

Read the full Google AI Mode conversation transcript →

---

Steven Sprague is CEO of Rootz and former CEO of Wave Systems Corp. He spent 15 years at the intersection of trusted computing and data integrity at the Trusted Computing Group. Origin.rootz.global is an open SEC data registry serving verified financial data to AI agents via HTTP and MCP.