Latest / Tech IPO Conversations with Fexingo: Public Software Companies and Capital Markets / How Post-IPO Companies Use AI Earnings Calls to Shape Narratives
Transcript
- Lucas: If these conversations are useful for what you're building or running, I want to start today with something that caught my eye during this earnings season. A handful of post ipo companies are now using AI to write — or at least heavily draft — their earnings call commentary. Not the Q&A, but the prepared remarks from the CEO. Luna: Wait — you mean they're having ChatGPT write the shareholder letter? Lucas: Essentially, yes. But it's more specific than that. Companies are fine-tuning models on their own past transcripts and 10-Ks, so the AI speaks in the CEO's voice — consistent jargon, consistent optimism about the same three metrics. I saw a job posting from a fintech firm last month explicitly seeking an 'earnings narrative engineer.' Luna: An earnings narrative engineer. That is a real title in 2026. Lucas: It is. And it makes sense if you think about it. The prepared remarks on an earnings call are the most controlled part of the entire disclosure. Companies already script them heavily. The AI just takes that to a logical extreme — generating a draft that hits every talking point the IR team has pre-approved, in language that backtests well with focus groups. Luna: But isn't there a risk that it sounds robotic? I mean, part of why investors listen to calls is to gauge the CEO's tone, their confidence, whether they hesitate on certain questions. Lucas: That's the Q&A. The prepared remarks have always been polished. If anything, AI makes them more consistent quarter over quarter. I looked at Lemonade's last two calls — their ai generated section on loss ratio trends uses almost identical phrasing. It's eerily stable. For an analyst, that's actually helpful: no surprises in the scripted part. Luna: So the surprise just shifts to the Q&A. And I guess the AI isn't answering analyst questions — yet. Lucas: Not yet. But there are experiments. A few companies are using real-time language models to suggest answers to the CEO during the Q&A — like a teleprompter for tough questions. The CEO can take the suggestion or ignore it. But the model is listening to the question and pulling from approved data points. Luna: That feels like a step too far. The SEC has rules about selective disclosure, about making sure all investors get the same information at the same time. If the AI is feeding the CEO a scripted answer that wasn't in the prepared remarks, is that a new form of guidance? Lucas: That's exactly the debate. The SEC hasn't issued formal guidance on ai generated earnings content yet — but they've started asking questions in comment letters. I know of at least two post ipo companies that got letters asking them to clarify whether the prepared remarks were 'personally reviewed and approved' by the named executive. The subtext is obvious. Luna: So the SEC is worried about accountability. If the AI writes something misleading, who's liable? The CEO who read it, or the model that generated it? Lucas: Right now, the CEO. And most companies are having the CEO do a final read-through. But the efficiency gain is real. One IR director told me their AI draft cuts the prep time for a call from two weeks to three days. That frees up the team to focus on the Q&A prep and the investor roadshow. Luna: I can see the appeal for a lean post ipo team. You don't have a massive IR department. You have maybe two people. Having a tool that generates the first draft of the script, the press release, and the shareholder letter — that's huge. Lucas: It is. But there's a subtle risk. When you outsource the narrative to a model, you might lose the ability to tell an authentic story when things go wrong. The AI is trained on past data — it knows how to frame a beat on revenue growth. But what about a product recall? Or a regulatory investigation? The model doesn't have the emotional intelligence to acknowledge a mistake in a way that rebuilds trust. Luna: That's a good point. And we've seen that play out. Remember when Palantir had that weird earnings call in 2024 where the prepared remarks were unusually terse? People speculated they used AI. It turned out they didn't, but the rumor alone hurt their credibility for a quarter. Lucas: Exactly. Perception matters. If investors think the CEO is reading a robot script, they start discounting the information. They assume the real story is being hidden. That's a trust deficit you can't fix with better AI. Luna: So what's the best practice right now? For a post ipo company listening to this episode, what should they do? Lucas: I'd say use AI for the scaffolding — the data aggregation, the first draft of the financial discussion, the compliance checks. But have a human — ideally the CEO — rewrite the narrative sections, especially the forward-looking part. And definitely don't use AI to generate answers on the fly during Q&A. That's where the real communication happens. Luna: That sounds right. And if today was actually useful to you — especially if you're building or running a public company — the way these stay ad-free is listener support. You can find us at buy me a coffee dot com slash fexingo. Totally optional, but it keeps the show independent. Lucas: Yeah, we don't run ads, and we want to keep it that way. So if you get value from episodes like this, that's the place. And now back to the earnings call question — because there's another angle I want to explore. Luna: What's that? Lucas: The analysts' response. I've talked to a few sell-side analysts who cover post ipo tech, and they say they're starting to build their own AI detection tools. They feed the prepared remarks into a model that compares the language to the CEO's past unscripted comments — from conference appearances, interviews, even old earnings Q&As. If the scripted part sounds too different from the CEO's natural speech, they flag it as potentially ai driven and discount it. Luna: So it's like an AI Turing test for earnings calls. The analysts are trying to figure out whether they're listening to a human or a bot. Lucas: Exactly. And that creates an arms race. Companies will train their models to mimic the CEO's natural speech patterns more closely. Analysts will refine their detectors. Eventually, the AI might become indistinguishable, and then what? The entire prepared section becomes noise. Luna: I wonder if that's already happening with some of the bigger names. Looking at the data we pulled today, I noticed Palantir is down almost 13 percent over the past five days. I'm not saying it's because of AI earnings calls, but if investors are skeptical about the narrative, that could amplify negative reactions. Lucas: That's a fair connection. Palantir has always had a distinctive CEO voice — Alex Karp is famously unfiltered. If he started reading perfectly polished AI remarks, investors would notice. And they might interpret that as the company having something to hide. So the risk is real. Luna: On the flip side, look at Airbnb — up 3 percent in the same period. I don't know if their earnings calls are ai assisted, but their CEO is known for being very direct and data-driven. If they used AI, it would probably sound natural because that's already his style. Lucas: That's the nuance. The effectiveness of ai generated remarks depends entirely on the CEO's baseline. If you're naturally a scripted, data-heavy communicator, AI fits. If you're a visionary who talks in metaphors, AI will stick out. Luna: So the advice might be: know thyself. And know thy investors. Lucas: Right. And maybe don't let the AI write the part where you apologize for a bad quarter. That should always come from the human. Luna: Agreed. And I think that's a good note to end on. Next episode, we'll talk about another emerging trend: how post ipo companies are using shareholder letters as a marketing tool beyond just earnings. Lucas: Looking forward to it. Thanks for listening, everyone.