An AI certification is worth something to an employer in exactly three situations: when the job advert names it, when the role is vendor-specific and the certificate is that vendor's, and when a recruiter needs a fast, verifiable filter on a large pile of applications. Outside those three, it proves that someone studied a syllabus and passed a proctored multiple-choice exam. It does not prove they can do the work. A declaration of interest first: Square 1 AI sells certification prep and it sells graded project work, so we have no reason to flatter either one.
What a certification actually proves
Strip away the marketing and a vendor certification is a claim with a narrow, precise meaning: on a particular day, this person answered enough multiple-choice questions correctly, under proctored conditions, on the vendor's published syllabus. For AWS Certified AI Practitioner, Microsoft Azure AI Fundamentals or Google Cloud's generative AI credential, that syllabus is real and the exam is genuinely supervised.
Three things follow from that definition.
It is verifiable. Every major vendor publishes a verification page or a shareable badge with a unique ID. That matters more than it sounds. In a hiring market where portfolios can be generated and testimonials can be invented, a claim that an employer can check in one click has real value. It is one of the few credentials a stranger can trust without trusting the candidate.
It is knowledge, not capability. Multiple-choice exams can test whether someone knows what a vector database is for and which service does what. They cannot test whether someone can scope a problem, check a model's output, or fix it when it is wrong, because those are behaviours and an exam cannot observe behaviour.
It is vendor-shaped. A cloud certification teaches you that vendor's services under that vendor's names. That is exactly right if the role runs on that platform and mostly irrelevant if it does not.
The three situations where it helps
When the advert names it. Many cloud, security and platform roles list a specific certification as required or preferred, and applicant tracking systems filter on it. If you want that job and you do not have the credential, you may never be read by a human. This is the strongest practical argument for getting certified: it is about the filter, not the work.
When the role is vendor-specific. If the team builds on Azure, Azure AI Fundamentals tells the hiring manager you already speak the platform's language. It saves onboarding time, and the manager knows it.
When the pile is large. A recruiter with four hundred applications for a junior role needs a filter that is fast, fair and defensible. "Holds a verifiable vendor certification" is all three. It is a blunt instrument, but it is honest about being one.
The two situations where it does not
When the question is "can they do the job". Once a candidate is in the room, nobody asks about the certificate. They ask what the candidate has built and how they work. A certificate cannot answer either question, and a hiring manager who has been burned by a certified candidate who could not ship will discount the credential heavily.
When it is a course-completion certificate rather than a vendor exam. A PDF from a learning platform saying someone finished a course is not the same thing. It is not proctored, it is rarely verifiable, and it is rarely checked. Employers have learned to ignore these, and they are right to.
The evidence on what predicts performance
The most cited study of selection methods remains Schmidt and Hunter's 1998 meta-analysis of eighty-five years of research. It reported a validity of 0.54 for work-sample tests and 0.51 for general mental ability, against 0.38 for unstructured interviews, 0.18 for years of job experience and 0.10 for years of education. The numbers have been revised since and argued over, but the ordering has held up: watching someone do a sample of the work predicts performance better than almost anything else, and credentials sit near the bottom.
A certification is closer to the education end of that table than the work-sample end. That is not a criticism of certifications. It is a description of what kind of evidence they are.
How to weigh one against a portfolio
If you are the candidate, the answer depends on the door you are trying to open. If a role you want names a certification, get it. The filter is real and arguing with it is pointless. If no role you want names one, build first. A project someone can open, run and change answers more of the questions a hiring manager actually has, and it is where your time compounds.
The two are not in competition. The certificate opens the door. The portfolio holds the conversation once you are through it.
If you are the employer, treat the certificate as what it is: a verified statement of knowledge on a published syllabus. Use it as a filter where a filter is what you need. Then assess capability the only way capability can be assessed, which is by watching the candidate work on something real for twenty minutes. We have written that exercise up separately, with the six tells to score against.
Where Square 1 AI sits
We built certification prep because a certification is an external anchor we do not issue and cannot influence. AWS grades the exam, not us. When a learner passes, the claim is theirs and the vendor's, and an employer can verify it without trusting a word we say. That honesty is worth a lot to a small company nobody has heard of yet.
We built graded project work because a certificate cannot show whether someone can scope, verify and iterate. Every project a learner submits is scored against a written rubric, every resubmission is kept, and the record is a trajectory rather than a single number.
We do not yet publish outcome data on either. We will when there is enough of it to publish honestly. Until then, the answer to the question in the title is the one we started with: a certification is worth something in three situations, nothing in two, and it is never a substitute for watching someone work.