Can hiring signals survive when AI makes cover letters worthless?

Kessler's column does more than diagnose the cover letter. It proposes replacements, matched to what each old signal was meant to show. The quality signal is replaced by third-party vouching: "a recommendation signals that a former employer is willing to vouch for a candidate and their skills." What matters is not whether the employer used AI to write the letter, but "that they stand by its content and are willing to be a reference." The interest signal is replaced by time: networking "perhaps by attending industry events, information sessions, or coffee chats," and, as video interviews become routine, volunteering for an in-person interview, which he calls "a costly signal of interest."

The mechanism is accountability or scarcity. A recommendation keeps its value because a named person accepts exposure for the claim, and that exposure does not depend on who typed the words. An hour of coffee keeps its value because it is an hour the candidate "cannot use any other way." Both are harder to fake than prose, which is the excerpt's test for a signal worth sending. The column closes with a general rule rather than a list: "hidden markets change. New signals take the throne," so the task is to find which signals are currently costly.

The only measured evidence in the excerpt is for the vouching half, and it predates generative AI. Kessler and Sara Heller of the University of Michigan randomly gave young workers "a short letter of recommendation from the supervisor at their summer job." Access to the letter "increased youth employment by 4.5% over the next year and increased earnings over four years by 4.9%." They estimate that including the letter in an application would have an effect "on the order of 10% to 15%" on employment and earnings. The Freelancer.com evidence in the sibling note Does AI-generated cover letter access weaken hiring signals? points the same way: employers facing AI letters leaned more on prior work history and reputation. That is the loss the replacement is meant to cover, so this note extends that finding into a proposal rather than a diagnosis.

The excerpt does not establish whether the interest half works. Coffee chats, information sessions and in-person interviews are argued for, but the excerpt reports no data on them and does not show that candidates who do them are hired more often. The recommendation study's sample size, its setting beyond "young workers," and whether the letter works as a signal to employers or through some other channel are not stated, and the 10% to 15% is an estimate, not a measured result. Nor does the excerpt test whether recommendation letters would keep their value once candidates can produce them as easily as cover letters. The implication is that the two-part framework is a hypothesis to test. The vouching half has the only measured evidence in this excerpt; the interest half rests on argument alone.