Say a chief financial officer is asked on an earnings call whether the European business will recover its margin by the second half. She answers for most of a minute about the platform investment now under way and the customers who have stayed through the transition. Every sentence is accurate. The analyst who asked writes down that the question went unanswered, and the next hand goes up.

A paper published this year in the Journal of Behavioral Finance puts a number on what that analyst did by instinct. Its four authors, three of them at the University of Limerick, used language model embeddings to score how close each executive response sits to the question that prompted it.1 Ranking firms within each sector by how closely their executives' answers matched the questions, the authors found that the closest-answering fifth beat the least close fifth by about 3.9% a year in alpha.1 Alpha here means return that standard market factors and sentiment measures do not account for, and the paper reports that the result holds after controls for sentiment, firm characteristics and market factors.1 To test whether the machine's closeness score matches what a listener means by evasive, the authors had 115 finance students rate 1,642 question and answer pairs. In the fifth of responses with the lowest similarity, 67% were rated off topic, against 22% in the fifth with the highest.1

An analyst remembers a handful of calls. The study ran its scoring across 164,944 earnings calls from 6,813 companies between 2008 and 2024, which turns one executive's pattern of answers into a record that can be pulled up and compared.1

What the number does and does not say

The finding describes a pattern across many calls and says nothing about any one executive. The authors read their forward-looking tests as evidence of a causal path, and they also call for further work to establish it more definitively.1 Two of the four authors list S&P Global Market Intelligence, a data provider to investors whose transcripts the study uses.1 My read is that this gives the group a reason to like the idea that answer quality is worth measuring. And a closeness score tells you the answer stayed on the subject. It cannot tell you the answer was candid, since a CFO can address the question squarely and still give a figure that gets revised in March.

Even with those limits, the paper's choice of what to measure is the useful part: the distance between the question asked and the answer given. My read is that the distance is cheap to widen. A direct answer creates a commitment that someone can check next quarter, and a nearby answer creates none, so a leader who has been burned once by a number learns that the nearby answer feels free in the moment. On earnings calls, the return gap suggests investors notice. Inside a company, the bill arrives as a small downward revision in how much the listener trusts the next answer, and nobody sends that revision back.

The same ears work inside the building

Nothing in the paper measures a town hall. My read is that the same listening happens there, and that employees do it with better data than any analyst, because they have years of the leader's answers to compare against. Picture the head of operations being asked at a staff meeting whether the restructuring is finished, and replying with the reasons it was necessary. The room has the experience the analyst had. Nobody objects. People note it, and the next time something goes wrong on their floor they decide how much of it to bring to that leader.

A nearby answer saves the speaker one commitment and costs the listener one reason to ask again.

Leaders who skip a hard answer for their own comfort are making a trade covered in an earlier piece on feedback withheld to spare the person giving it. The pattern in meetings is the same, with a larger audience.

Answering the question as asked

What would I ask a leader to try? Before speaking, restate the question in one sentence, and make the first sentence of the answer address it, including when the honest content is that the answer is not known yet. "I do not know, and I will tell you by the fifteenth" answers the question that was asked, even if a similarity score might not credit it. I would expect listeners to hear it as direct, though the paper does not test that case. The longer context can come after, and it will be heard differently once the question has been met.

The habit takes practice because the instinct runs the other way. A leader who knows the full story feels that the question was too narrow, and the wider answer feels more useful. Sometimes it is. The listener still asked something specific, and the first thing they weigh is whether it was heard.

Which question did you answer in your last all hands meeting, the one that was asked or the one you were ready for, and who in the room noticed first?