An AI "Answer" Is a Confident Guess Until It Cites a Source. A Data Firm Just Built Its Fix Into Copilot

The picture most investors carry: type a business question into an AI assistant and you're getting the answer a research subscription costs thousands for, because the model has read the whole internet. The costly conclusion that follows: chat is free, and any data vendor whose product a chatbot can imitate is doomed.

Both are wrong, and the September 1, 2026 announcement from Euromonitor International is a tidy window onto why. Euromonitor connected its Passport market-intelligence database to Microsoft 365 Copilot so that customers who are licensed for both can ask a question in plain English and get an answer that carries a traceable citation from Euromonitor's proprietary data, right inside their normal workflow.

Before the deal makes sense as an investment event, understand what is actually being sold. It's not a fancier chatbot. It's a fix for a specific, expensive failure.

The intern who wings it

Imagine you've just hired a brilliant intern. Whip-smart, reads faster than anyone you know, speaks in complete, winning sentences. There's one problem: the intern has no sources. Ask for the "fastest-growing beauty category" and the intern will give you a number, confidently, with a straight face — then hand over another number tomorrow that contradicts it, also with a straight face. The intern isn't lying. The intern is improvising: generating the most plausible-sounding words, not retrieving a fact it verified.

That is a large language model. Fluent, fast, and capable of stated confidence in a fabricated figure. Take a business decision off a made-up number and you pay twice: once for the time you spend checking it, once for the decision you made while wrong.

Now hand the intern a locked reference library — a few thousand dollars' worth of verified reports — and add one rule: never give me a number unless you can point to the shelf and the page. The intern's fluent answers are suddenly footnoted. You can walk over and check. That is what Euromonitor just did inside Copilot, and the financial industry has a name for it: grounding.

Now label the props, because each one maps to real money.

• The intern is the language model inside Microsoft 365 Copilot — the engine that generates fluent text.
• The intern's improvisation is a hallucination: a plausible but invented "fact."
• The locked reference library is Passport, Euromonitor's proprietary database of verified market data.
• The rule "point to the shelf and page" is Copilot naming the source dataset, coverage period, and methodology behind each answer.
• The catalog system the intern uses to fetch pages is the Model Context Protocol, or MCP — the technical standard that lets an AI connect to an outside data source.
• The office where this all happens is Microsoft 365: Word, Excel, Teams. The place the work actually gets done.

Ask both versions the same question. The ungrounded intern gives you a smooth percentage and no page — unverifiable, maybe invented. The grounded agent pulls from Passport, names the category and the growth figure, and cites the dataset behind it.

A private library, but a big one

The toy version helps you see the machine. Euromonitor's real numbers show why anyone pays for the library in the first place. Passport covers 210 countries and jurisdictions — about 99.9% of the world's consumers — and Euromonitor calls the ground-up rebuild of the platform, of which this Copilot integration is one piece, "the biggest single investment in Euromonitor's history," a business founded in 1972. That library is not scrapable from the open web, which is why "just use a free chatbot" fails for the decisions that actually matter.

Euromonitor says the integration is available to organizations that already hold a Passport subscription, listing roughly 60,000 Passport users across 110 countries. Euromonitor's CTO, Chris Fosberry, frames it as connecting decision-making to 50 years of human-led insights; Microsoft's Rob Howard, a CVP for Microsoft 365 Copilot extensibility, echoes the "trusted intelligence" pitch. In other words, this is one data vendor plugging its verified library into one AI platform — and it is not alone.

Where this breaks — and what it doesn't fix

The analogy has done its job; here is where it ends, because two easy misreadings would cost you.

First, a citation is not proof. Grounding tells you which source a figure came from; it does not tell you the model interpreted that figure correctly, or that the interpretation fits your situation. A "sourced" answer is still a claim you should read, not a verdict to obey.

Second, this is not a Microsoft-exclusive moat. Euromonitor made Passport "MCP-ready," meaning the same verified library can be connected to other assistants — Claude and ChatGPT among them — and MCP is an open standard that Microsoft itself now supports in Copilot. The data owner did not lock itself to one platform; it built a way to be handed to any intern who follows the catalog protocol.

Now bring the model back to the stock

Here is the part investors actually can act on, because the company in the headline — Euromonitor — is privately held. You cannot buy the data owner in this deal. So read the announcement as evidence about two things you can.

The first is Microsoft. This deal is a concrete sign that third parties will build agents on Copilot, turning it from a chat feature into a distribution channel for outside data. Microsoft's incentive is clear: every integration like this makes a paid Copilot seat more valuable and stickier, and routes the work through the tools Microsoft already owns. That fits the bull narrative for the stock — which, at roughly $497, trades near its 52-week high of about $554 and at a forward price-to-earnings multiple in the mid-30s. The mechanism is real; whether seat adoption converts into revenue at a multiple like that is the open question, not this announcement.

The second, and the one worth more of your attention, is the class of companies that own the grounding. As AI shifts value from the fluent model to the verified data beneath it, the scarce inputs are the ones that are licensed, correct, and hard to replicate — credit ratings, financial data, real-estate data, consumer surveys. Analysts at Morgan Stanley, Morningstar, and S&P have all written in 2026 that proprietary data is exactly the moat that survives the rise of AI. Public names sitting on that kind of data — firms like Moody's, S&P Global, FactSet, CoStar — are the tradable cousins of a private Euromonitor, and they are where this story shows up in a portfolio.

But keep the two-way risk visible. Moody's, for one, talks openly about defending its proprietary data from large language models even while it builds with them — co-opetition, not a one-way win. AI can raise the value of verified data and commoditize the periphery of it at the same time.

The one test to keep

If you remember one question when the next "AI meets data" headline lands, use this one: who owns the grounding, and is it genuinely hard to copy?

A chatbot that merely sounds smart earns nothing; its confident guesses are everywhere and free. The durable value sits with the platform that distributes the work and the firms that own a library voters can't rewrite. And one warning comes with the model: an AI answer with a citation is evidence you can check, not a reason to stop checking. That is the difference between the intern you can trust and the intern you merely can't shut up.

That's the real product being launched here — and it isn't Euromonitor's to own as a ticker. It belongs to you, the moment you remember that an answer and a source are not the same thing.