I’ve spent the last fortnight playing Mystery Investor. Not just for shits and giggles, though it was fun.

I’ve been investigating how AI-driven questions from shareholders are piling pressure on finance and investor relations teams. And I wanted to do a test of my own to see how they’re coping. I bought real shares in three listed companies – a retailer, a defense provider, and a manufacturer. And I emailed them five queries each, generated in seconds by Claude (See the full prompt and Q&As below).

The result was surprising. I thought I’d get nothing back. But the retailer and defense company sent me genuinely helpful replies within a week.

Unfortunately, the manufacturer ghosted me like a bad Tinder date.

The experiment showed how easy it is for investors to generate “intelligent” interrogations in no time – although once I dug down, some were not as smart as they appeared. And the responses suggested some firms, but not all, are getting used to replying to AI generated queries. There’s more about the process and our verdicts below.

But first, let’s investigate how this trend towards robotic questions is impacting CFOs and how they’re responding.

Finance heads are having to spend more time on IR as they adapt to investors’ use of AI, according to a 2026 Gartner report. That includes extra hours on:

Michael Perica, CFO of software support provider Rimini Street, a listed company, said he is feeling the effects of this trend first-hand with a greater volume of investor enquiries and more unexpected questions coming in.

“Investors are querying new areas and in original, creative and insightful ways. It’s hard to imagine this is not driven by AI. Overall, this is positive as it leads to much more productive engagements. But the extra volume, depth and variance lead us to spend more time preparing for earnings calls,” said Perica.

He also agreed that, as our experiment showed, LLMs can dumb down some questions too.

John-Paul Crutchley, head of investor relations at wealth management firm Quilter, said many analysts use AI because, following consolidation in the industry, they now cover many more companies than they used to. 

“AI agents have removed much of the necessary detail and finesse in analysts’ work. These tools can evolve some queries but they can lead to a lot of basic, derivative enquiries too, regurgitating what’s already out there,” said Crutchley.

But Crutchley noted that the use of AI has dumbed down the quality of some investor questions. 

“We’d rather have a good conversation, where the investor is thinking about the industry, five-year goals, and how you drive them with competitive angles. That’s not AI-driven,” he said.

AI enables investors to analyze much more information – such as news, filings, and research – so they can cover more, or allot more hours to higher-value activities. But they also risk surfacing outdated material, mistakes, and missed context.

Tom McMillan, founder and principal at MCI Capital Markets, said: “Bragging about the success of your AI programs is like a parent bragging about their off-the-rails teenager. AI for interpretation and judgment is error fraught, and can lead to:

“An example is investors using AI to comb millions of filings, looking for signals. I’ve seen AI interpreting an update to an executive’s compensation package as a key executive leaving. That could erroneously move your stock price.”

Not all researchers are on board with AI. An anonymous equity analyst told me he’d tried using it to prepare for an earnings call recently but found the results too bland and generic. He role played some face-to-face questions he’d rather ask, which sounded much more natural and emotionally smart than Claude’s efforts.

Less pedantic robot, more probing detective, so to speak.

One way to help avoid AI dishing up bad ideas or information in the first place is to make sure your IR outputs are LLM friendly. 

Crutchley said he and his team are “having lots of conversations about how to ensure what you put on your website is accessible to AI. For example, is it in pdf or a more readable format? What will it pick up on? We’re thinking about how to run it through an LLM before we put it out to ensure the right prompts and cues are picked up, and that it feeds into trading algorithms.”

This needs to be done in a private, contained and traceable environment, so nothing gets accidentally leaked before publication date.

How to feed AI has become critical for IR. According to a report by advisory firm Brunswick, 46% of investors are skipping earnings calls in favor of AI-generated summaries. 

“Companies need to take responsibility for how information reaches investors,” according to the report. “Make accurate information about your company available online for ChatGPT, Claude, Perplexity and other platforms to discover. For example, Gen AI prioritizes direct company sources in its results.”

Perica agreed AI-led investor research sometimes misfires. “It’s drawing conclusions that make sense, but does that mean it’s actually happening? In our follow-on Q&As, we have to be ready to caution investors that AI answers are based on probabilities, not always rooted in fact.”

McMillan recommended:

Finally, AI interrogations may be irritating but cut your analyst a little slack. “Remember, these issues are an outcome of analysts under incredible pressure, turning to AI but not having time to vet the output,” said McMillan.

I bought a small number of shares in three listed companies: a retailer, a defense firm, and a manufacturer.

I used a simple off-the-shelf prompt asking Claude Sonnet to generate five questions about each company’s performance.

I deliberately spent no time researching the companies, but simply copied what Claude gave me and sent it to each firm’s investor relations email address. I gave my name but didn’t reveal my real agenda.

Most of the questions appeared deep and probing, as requested. However, there were a few basic errors, misunderstandings and misinterpretations.

An IR team member replied within three days. She answered all the questions politely and helpfully, albeit briefly with one paragraph each. Claude produced some smart questions here. But there was also one semi-redundant query and one factual error. The IR team didn’t call these out but simply stated the correct position.

Verdict: Rapid, useful answers, that carried the pace and to the point response you’d expect from someone who spends all day responding to these kind of half baked queries.

The director of IR responded four days after my email. He thanked me for my support as a shareholder and gave helpful answers of two to three paragraphs each. Claude offered some insightful questions, but they also contained one miscalculation and one timing error. The director dealt with these by sticking to the big picture rather than correcting specific problems.

Verdict: Smartly, but politely, regurgitated from the most recent results presentation.

A quick analysis showed Claude’s questions were fair and pointed. But it posed two queries that the firm had already answered in an earnings call and management guidance. They also contained some slightly inaccurate wording about past performance, and regulatory timelines. So far, they haven’t responded.

Verdict: Still waiting…