AI visibility for your equity story: what IR teams should check

In short

Investors and analysts now research listed companies through AI tools such as ChatGPT and Perplexity as well as through the IR website. That leaves an IR team with three questions few can answer today. Which sources do the models actually use? Can you steer what they say? Can you still see what investors ask? This guide takes each question in turn and lists the practical steps that follow. None of them require rebuilding your website.

Key facts

• You do not choose the sources a general model uses. The citations investors see usually name third parties writing about you.

• Your site only enters the answer if the crawler can reach it, and that is not guaranteed.

• Once investor research moves into general AI tools, IR can no longer see what investors ask.

• The fixes are infrastructure work and do not require rebuilding the site.

Institutional investors already work this way. In Brunswick’s 2026 survey of 100 US institutional active-equity investors, 54% rated GenAI as at least moderately important in their investment research.^[1]

Which sources does a general AI actually draw on?

Sources you did not choose. The model works from whatever public material it can reach. Your own pages compete with aggregators, reference sites, news and third-party commentary. Where those answers come from, and what a citation in one actually covers, is set out under when investors ask an AI about your company, where do the answers come from.

One point matters for everything below. The summary is often broadly accurate, yet neither you nor the investor can check where it came from. A listed issuer must publish information that is fair, accurate and backed by evidence. An answer nobody can trace fails the last test.

Can you steer what a general AI answers about your company?

Not directly. On the open models you cannot edit how they summarise your story, and you cannot push an update the day you release results. Below that sits a technical layer that rarely reaches the IR team at all.

Settings made for unrelated reasons can close your site to AI crawlers entirely. When that happens your own pages never enter the answer, however clear the equity story on them is. Rendering does the same thing more quietly. In December 2024, Vercel and MERJ measured crawler traffic across Vercel’s network. The OpenAI, Anthropic, Perplexity, Meta and ByteDance crawlers all fetched JavaScript files without running them. OpenAI’s crawler fetched them in about 11.5% of its requests. Google’s Gemini and AppleBot were the exceptions.^[2] A site that builds its content in the browser shows the other crawlers an empty shell. All of this comes before any content work. If a crawler cannot reach your content, the content does not count.

Can you still see what the market is trying to work out?

Less and less. The topics investors probe and the points where they hesitate used to reach IR through calls, emails and meetings. Once the research happens inside a general assistant, IR loses that signal. What it costs, and how to get it back, is covered under what investors are asking about your company, and why IR stopped seeing it.

It belongs in a guide about visibility because the two losses arrive together. Sourcing is about what investors are told. Question visibility is about what you learn. Most teams only notice the first.

How do you check what AI can read on your site?

You can do most of it in an afternoon, without a developer next to you.

• Read your own robots.txt. Open yourdomain.com/robots.txt and look for the AI agents by name, such as GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended. A missing name does not mean permission, and a listed name does not mean a block. Check what the file actually says about each one.

• Ask separately about the security layer. Bot protection and CDN rules sit in front of the site and can refuse a crawler that robots.txt allows. The file will not tell you this. Only whoever administers that layer can.

• Look at the page without JavaScript. Disable it in your browser and reload an IR page, or view the raw source. Whatever is missing is roughly what the non-rendering crawlers see.

• Ask an assistant about your own business and read the citations. The answer’s quality is not the point. Look at which sources it used and whether any of them are yours.

• Run the same question against two peers. The comparison tells you whether the problem is yours or common to your sector.

Read the results together. If robots.txt names and allows the agents but an assistant still cannot cite you, the block is almost certainly in the security layer. Take that to whoever administers it. If the page comes back empty with JavaScript disabled, the crawlers get nothing, and no permission setting will change that. If both checks come back clean and the assistant still cites third parties, access is fine. The question then becomes what your pages actually say. That is the one case where content work is the right next step.

So the order matters. Run together, the checks show which of the two problems you have before you spend money on the wrong one.

What can an IR team actually do about it?

Treat it as an infrastructure problem, because that is what it is. It does not require rebuilding the website.

If the checks show that the site turns crawlers away, fixing that is a configuration job, and an expert review of the site will scope it. Start with that baseline.

The second step is structural. Give investors somewhere to ask that draws on your own record. That means a source-grounded assistant, limited to your published material, with a reference behind each point. MIRA is one, and the mechanics sit under what is a source-grounded IR assistant. Because the questions then run through your own channel, IR sees them again.

The bottom line

Investors now form first impressions of a listed company inside AI tools. The models build those impressions from material the company did not choose, sometimes without reaching the company’s own pages, and IR never sees the questions. The fix is infrastructure.

The good news is that most of it is not IR’s job to fix. Ask one question of whoever owns your website, internal IT or an external agency. Are the AI crawlers reaching our IR pages today, and can you show me the evidence? IR does not need to answer that. IR needs a name next to it and a date. The peer comparison and the assistant on your own site follow from the answer.

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. MIRA won Best New Digital Finance Product of the Year at the Digital Finance Awards Belgium 2026, organised by FinTech Belgium and judged by an independent jury.^[3]

References

1. Brunswick Group, 2026 US Investor Survey. 100 US institutional active-equity investors, half long-only managers and half from hedge funds. Brunswick Review, 9 February 2026. review.brunswickgroup.com/article/investor-survey-2026

2. Vercel and MERJ, The Rise of the AI Crawler, 17 December 2024. Crawler requests measured across Vercel’s network. vercel.com/blog/the-rise-of-the-ai-crawler

3. FinTech Belgium, The Digital Finance Awards Belgium 2026 Celebrates the Best and Brightest in Digital Finance, Ghent, 23 June 2026. fintechbelgium.be/news

See where the answer stops. Book a MIRA demo on your own published documents, and we will show you the questions it declines as well as the ones it answers.

About the author: Sander van Prooijen is Head of Commercial Strategy at Finvictum. He leads the European rollout of MIRA and works with listed companies on investor communication in the age of AI. Connect on LinkedIn.