How to choose an AI course

Why the rankings all look the same

Search for the best AI course and page one is numbered listicles: platforms ranking their own inventory, affiliate pages ranking whatever pays, and agencies ranking whatever leads to their services. The prices are often stale or plainly wrong. One Australian ranking page lists Google's course, which Google sells as a US$49 monthly subscription, as a flat several hundred dollars.

None of them ask the only question that sorts the market: do you run the business, or do you work in a technical role? Almost everything on those lists is built for the second reader. This guide is for the first.

The course types, with real prices

Prices below were checked against the sellers' own pages in August 2026. They move often, which is itself a lesson about this market.

University prices are GST-inclusive as published. Read the refund clause before paying anything; consumer-guarantee law applies to paid training in Australia and New Zealand.

Named options, at August 2026 prices

A snapshot of verifiable options across the tiers, priced from the sellers' own pages in August 2026. Treat it as a map of the market, not a ranking.

Every price above was checked on the provider's own page. Two university course listings changed or vanished while this guide was being researched, which tells you how quickly to re-check before enrolling.

The completion problem nobody prices in

The research on self-paced online courses is brutal. A study in the journal Science tracked 12.7 million enrolments across six years of MIT and Harvard online courses and found that 52% of people who register never open the courseware at all, and the share coming back for a second year fell from 38% to 7% as the novelty wore off.

That is not an argument against learning online. It is an argument for formats with a date and other people in them. A one-day workshop you attend beats a subscription you abandon, whatever the syllabus says, and no self-paced platform we checked publishes its own completion rates.

Six questions that sort good from grift

• Does it teach using AI, or building it?

  Courses about Python and model training answer a different question from courses about getting real work done with AI tools. Owners and managers nearly always want the second and get sold the first.

• When was the content last rebuilt?

  Not last edited, rebuilt. Microsoft retired its most-recommended beginner AI exam in mid-2026 and replaced it outright. A curriculum older than a year in this field is a history course.

• Who teaches it, and what have they shipped?

  A named teacher with verifiable work beats a faceless funnel. The internet is full of $497 courses on building AI businesses from people who have never built one.

• Is there a date, or just access?

  The completion evidence says a scheduled cohort or workshop is worth more than an open-ended library. Access is what they sell; attendance is what you need.

• Are outcomes shown, or implied?

  Named alumni, real projects, reviews somewhere the seller cannot edit. If the only proof is enrolment numbers, the product is the enrolment.

• What does it really cost to finish?

  Subscriptions price by the month, which means slow learners pay more. Flat-fee courses with GST-inclusive prices are the most honest tier of this market.

Or should you just use the tools instead?

The most common advice from people who learned this recently is that two weeks of deliberate daily use, applied to your own work, teaches more than most courses. For a lot of owners that is correct, and it costs nothing. The honest case for a course is narrower: you want structure and a deadline, you want current practice curated by someone who does the work, or you want your questions answered live rather than guessed at.

And if the real goal is to appear across it rather than to be across it, save the money. Nobody you want to impress is checking for certificates.