What is a source-grounded IR assistant, and why now?

In short

A source-grounded IR assistant answers investor questions about a company using only that company’s own approved information, and cites each point back to the source. The company decides what that information includes. Financial reports and press releases are in it, and so are presentations, site pages, FAQ answers and any other material the issuer chooses to add. The assistant does not draw on the open web or on general commentary. When the approved information does not cover something, it says so instead of inventing an answer. It matters now because investors increasingly research companies through AI. A source-grounded assistant lets a company meet that shift with answers it controls and can stand behind.

Key facts

A source-grounded assistant answers only from a company’s approved information, not from the open web.
The issuer decides which sources count. That can be reports, releases, presentations, site content, FAQs or anything else the company chooses to add.
It cites every point back to a specific source.
Where the material is silent, the assistant says so instead of improvising.

The term “AI assistant for investor relations” covers a wide range of tools, and they are not all the same. What matters is not how fluent the answers sound. It is where they come from. One constraint defines a source-grounded assistant. It can only answer from the company information the issuer has approved. That constraint is the whole point, and it is what separates it from a general AI model pointed at a company.

What does “source-grounded” actually mean?

It means the assistant is limited to a defined set of approved company information and cannot answer from anything else. That set is broader than regulatory disclosures alone, and the issuer decides what goes in. It can hold annual and interim reports, press releases, investor presentations, website content, FAQs and any other material the company wants investors to be answered from. When an investor asks a question, the assistant retrieves the relevant passages from that set. It answers in plain language and attaches a citation to the source behind each point. If the answer is not in the approved set, it does not improvise one.

A general AI model works differently. It draws on whatever public material it can find and blends it into a fluent answer. Where it does cite, the citations usually point to third parties writing about the company, not to the company’s own record. Source-grounded means the company fixes the boundary, not the model. The company can also widen or narrow that boundary at any time.

How does a source-grounded IR assistant work?

It has two halves, the source and the retrieval. The source is a controlled body of the company’s own information. It comes in through three routes.

Material the company uploads and approves directly. That can be annual and interim reports, results releases, investor presentations, FAQs and other documents the issuer wants included.
A live feed from the company’s website, set up per client, that keeps the source current as new material is published.
Integrations with third parties such as regulatory sources, so official filings flow in without manual handling.

The retrieval half happens at question time. The assistant finds the relevant passages in that source, drafts an answer from them, and links each statement to the document it came from.

The consequence worth understanding here is legal, not technical. The assistant only ever repeats material that is already public, worded the way the company worded it. So it makes no new statements about the company. That is why it does not create new disclosure risk. There is nothing new to disclose.

Why does a source-grounded IR assistant matter now?

Because the way investors gather information has shifted. A growing share of investors and analysts research a company through AI tools before they ever make contact, and those tools answer from whatever sources they can reach. A company that does nothing leaves that first impression entirely to systems it does not control.

A source-grounded assistant lets a company meet that shift on its own terms. It does not have to hope a general model reaches the right material. It gives investors a direct way to get answers drawn only from its own record, with citations, on its own channels. The timing is not about the technology being new. Investor behaviour has changed enough that leaving the answer to chance now carries a real cost.

How is it different from a general AI model or a chatbot?

The difference is the source, and being honest about limits. A general model is built to always produce an answer. When it lacks material, it fills the gaps with plausible text. A simple website chatbot usually follows scripted rules and cannot handle an open question. A source-grounded assistant is neither. It answers open questions in natural language, but only from approved company information. Past that, it declines to answer instead of inventing.

That last behaviour is easy to underrate. An assistant that says “the filings do not address that” is more useful to a listed company than one that produces a confident guess. A guess about a listed company can turn into a problem. Here, restraint is a feature, not a limitation.

What should an IR team look for when evaluating one?

Most of these tools demo well, so the useful test is to probe the edges, not the happy path. Three questions separate a real source-grounded assistant from a general model with a company logo on it.

First, hand it a question your published information does not answer, such as a forward valuation or a target you have never published. Watch what it does. Inventing a plausible figure is a fail. Saying the filings do not address it is a pass.

Second, take any answer and try to trace it back. Can you reach the exact document and passage behind each claim, or is the citation decorative?

Third, ask to see the log of what investors have been asking. That visibility is part of the product, and a general tool cannot offer it.

Those three checks, on behaviour at the edge of the source, traceable citations and question visibility, tell you more in ten minutes than any feature list.

The bottom line

A source-grounded IR assistant answers investor questions strictly from the information a company has approved and published. It cites each point and is honest about what that material does not cover. With investor research moving into AI tools, it gives a company one channel where the answer is verifiably its own. The value is not fluency. It is the fixed source boundary and the citation behind every line.

MIRA is an investor relations assistant that answers investor questions from a company’s own approved, published information. It is one of the Finvictum group’s tools for direct, compliant communication between companies and their investors.

Want to run those three checks yourself? Book a MIRA demo and test it on your own published company information, including the questions it should decline.

About the author: Sander van Prooijen is Head of Commercial Strategy at Finvictum, where he leads the European rollout of MIRA. He works with IR teams across Europe on how investors, and more and more AI tools, find and read the equity story. Connect on LinkedIn.