AI Certificates: A Paper to Hang or a Proof to Show?

Credential Engine counted 1,850,034 unique credentials in the United States alone in 2025, offered by more than 134,000 providers: about 1.02 million digital badges, 486,000 certificates and 264,000 degrees. These are figures for all credentials, not for artificial intelligence alone, but they are enough to show the problem: when papers multiply, any single paper persuades less.

A certificate is a signal

The economist Michael Spence called this mechanism "signalling". An applicant bears a visible cost, such as years of study or an examination, to convince an employer who cannot see their abilities directly. A signal works under the condition summarised on the Nobel Prize page: its cost must differ sufficiently between senders, so that it is easier for the more able to obtain. If everyone obtains it with the same effort, it reveals nothing. That is why one may doubt a certificate given to everyone who finishes watching the lessons. This is the editors' inference from the theory, not a measurement of any particular programme.

What does the evidence say?

Most of the available evidence concerns university degrees, not short AI certificates, so it should be read with care. In a correspondence experiment with fictitious CVs (Deming and colleagues), an applicant with a business bachelor's degree from a for-profit "online" institution was 22% less likely to receive a callback than one with the same degree from a non-selective public university. In the other direction, the Hechinger Report relays a Harvard Business School-led study, Dismissed by Degrees (2017), in which 61% of 600 business and human-resources leaders surveyed said they had discarded the CVs of qualified applicants because they lacked a university degree. So a signal is sometimes read for more than it is worth, and sometimes a capable person is screened out for lacking it.

As for the short certificate itself, the closest we found is a field experiment registered with the AEA RCT Registry by researchers at Georgia State University and the Georgia Institute of Technology. It planned about 4,000 job applications with fictitious CVs, some listing an online specialisation certificate. At the registry's last update, in May 2023, no results had been published. We do not claim that no other studies exist, but what we opened does not support a reliable figure for the value of these certificates.

The signal that is hard to forge

If a signal's strength lies in its cost being higher for those who lack the ability, the strongest signals are those whose forgery costs as much as mastery: a model you built on real data and tested against a stated measure, or a solution to a specific problem that someone else can run and review. That is an editorial view supported by the logic of signalling, not the result of an experiment. Nor does it mean a certificate is worthless; it may help a beginner with no record. But its quality can be tested with questions:

• Does the programme have an assessment the learner could fail?
• Does it end in a project reviewed by someone other than the learner?
• Does it teach a lasting foundation, such as statistics, model evaluation and data handling, or the interface of a tool that will change within months?
• Who awards it, and does the awarding body profit from the number enrolled or from graduates' success at work?

Before you spend your time and money

Treat the certificate as a stopping point, not a destination. Choose a programme that obliges you to produce work, keep what you produce where others can see it, and write in your CV what you can do, not just the name of the course. Look at your own market: what do employers in your field actually ask for? To go further on building a portfolio, see the article "Building an AI Project Portfolio: What Convinces an Employer?", and on making learning last, "How Should You Learn AI? What the Evidence Says About Effective Learning". A seeker of knowledge keeps their intention sound: they seek it for benefit and trustworthiness, not to collect papers. Success comes from God, together with taking the means.