AI Certifications: Which Ones Are Worth Anything

Any ranking of specific certificates goes out of date within a year and rests on claims about what employers value that nobody writing it has surveyed. What survives is a test you can run yourself, and the test is short.

Why this page is not a list

Two reasons, both of which are worth stating plainly rather than apologising for. First, the programmes turn over constantly: curricula are rewritten, vendors launch and retire credentials, and a recommendation made today describes a syllabus that will have changed by the time you enrol. Second, and more importantly, any claim about what “employers value” is survey data. Nobody here has surveyed employers, the answer differs by country, sector and company size, and a confident ranking would be invention dressed as advice.

What can be reasoned about honestly is what a certificate is structurally capable of demonstrating, which does not depend on the market at all. That reasoning gives you a test, and the test works on a certificate that launches next year.

Four questions to ask of any certificate

• 1. What exactly does passing prove? Write the sentence: “A person who holds this can …”. If the sentence is “recall which service does what”, that is a real but narrow claim. If you cannot finish the sentence without using the word “understands”, the credential does not know what it is testing either.
• 2. Who verified it, and could they have failed you? A meaningful pass rate implies a meaningful fail rate. A programme where completion follows attendance is a receipt, not an assessment. Ask what proportion of candidates do not pass; if the answer is not published anywhere, assume it is high.
• 3. Was anything built under observation? A proctored practical exercise is worth more than a multiple-choice exam by a wide margin, because the failure modes of this work are about judgement under uncertainty and multiple choice cannot reach them.
• 4. Is it attached to a product or to a subject? Product certificates prove familiarity with one vendor’s surface, which is genuinely useful if you work in that ecosystem and worthless if you leave it. Subject certificates age better and are rarer. Neither is wrong; they answer different questions and are frequently confused.

A fifth question that is not about the certificate: what would you have done with those forty hours? That is the real comparison, and it is the one the marketing never frames.

Most certificates in this field fail question 3, and it is worth being explicit about why that matters rather than treating it as a technical quibble. An assessment that cannot observe you building something can only test recall, and recall is the part of this work that has been most thoroughly commoditised — it is looked up, and increasingly it is looked up by the same models the certificate is about. A credential that certifies the most easily substituted half of the job is not worthless, but it is being sold as evidence of something it never examined.

Question 2 is the one to insist on hardest, because it is the one with a commercial conflict behind it. A programme earns money from enrolment and reputation from completion, and both point in the same direction: pass people. That is not an accusation of bad faith, it is an observation about incentives, and it is why an unpublished pass rate should be read as an answer rather than as an omission.

What a certificate can prove

Being fair to the format, because the dismissive take is as unearned as the enthusiastic one. Certificates are structurally good at three things:

• Coverage. A syllabus forces you across material you would not have sought out. Self-directed learning has a characteristic hole in exactly the topics you find boring, and a curriculum patches it.
• A deadline. An exam date is a commitment device. For many people this is the entire value and there is nothing wrong with buying it deliberately.
• A shared vocabulary. Particularly for vendor-specific credentials: knowing what a platform calls things makes you effective in a team that uses it, and that is real even if it is not deep.

What they are structurally poor at is the thing this field most needs demonstrated: judgement under uncertainty. An exam has a correct answer. The work does not — it has trade-offs whose resolution depends on what a failure costs, which is why an artefact with a measurement attached carries information a certificate cannot.

Three situations where it is worth it

• Somebody else requires it. Partner status, procurement rules, a consulting engagement, a regulated sector, an employer’s internal framework. Here the certificate is not evidence, it is a key, and the only question is cost. Do it and stop thinking about it.
• You are changing field and have no artefacts yet. A credential is a weak signal, and a weak signal beats no signal when a screener has thirty seconds. Its job is to get you far enough to show the thing you built — which means it is a supplement to a portfolio, never a substitute.
• Your employer pays and you want the structure. If the time is not coming out of your own evenings, the calculation changes entirely. Take the coverage and the deadline; they are worth having.

And one situation where it is reliably not worth it: as a response to feeling behind. A certificate produces a sense of progress that is largely uncoupled from capability, which makes it an unusually effective way to spend a month without changing anything — the same trap the fatigue page describes.

What a certificate is competing with

The honest comparison is not “certificate versus nothing”. It is “certificate versus what else forty hours buys”, and the alternatives are specific:

The last row is the one most people overlook, and it is available to anyone already employed. A change made at work, measured properly and written up, outperforms every other row on this table for almost every purpose — and unlike the others, someone is paying you while you do it.