Prompt Engineering Course (2026): Best Free and Paid Picks

Best prompt engineering courses compared

These are the prompt engineering courses a non-coder is most likely to find, with what each page says today.

The test further down shows the two things a course has to teach: practice on your own work, and giving the AI context (files, examples, background), not only wording. Here is what each course page promises on both:

Only two of these say plainly that you work on your own material: OpenAI Academy, whose hub says people learn "by practicing on work that matters to them", and the Maven live cohort. Udemy's prompt engineering courses block automated checks, so we couldn't verify their prices or update dates; if you buy there, watch the free preview and read the "last updated" month first.

Free prompt engineering courses worth your time

OpenAI Academy's AI Foundations is the best free prompt engineering course to start with. You pick a task you do every week and practice "giving clear instructions, providing context, reviewing outputs" on it, then pass an assessment with 80% or more for a badge; 58,163 people had joined it on October 25, 2026. Our best ChatGPT courses guide covers the two Academy courses that follow it.

The best-known Coursera prompt engineering course is still Vanderbilt's Prompt Engineering for ChatGPT, with 748,879 learners and 4.8 stars from 8,231 reviews. You can enroll for free, but the certificate needs a paid option, and its lessons date from 2023: it opens with an "Act as a Speech Pathologist" example and drills persona patterns that today's models need less. IBM's Prompt Engineering Basics (about 9 hours, 694,008 learners) works the same way: free to watch, paid certificate.

If you want a free prompt engineering course with a certificate, there are three, each with a catch:

• Cognitive Class, Prompt Engineering for Everyone: IBM's free skills site, 5 hours with 9 labs and a certificate, but its labs use 2023-era persona exercises.
• Simplilearn SkillUp: listed as one hour of video and a free completion certificate, with 90 days of access after you register.
• Microsoft Learn: free Copilot modules with a badge. Microsoft has retired its standalone Copilot prompting course, so prompting is now a 4-minute unit inside its Copilot path.

LinkedIn Learning is also free through US public libraries that offer it: its help center says patrons sign in with a library card, though their names can't appear on the certificates.

Paid prompt engineering courses: what the money buys

coursera.org/specializations/prompting-essentials-google

Prompting Essentials is a separate program from Google AI Essentials, which covers AI more broadly; our Google AI Essentials review compares the two. Two famous names are courses for developers rather than everyday users: DeepLearning.AI's ChatGPT Prompt Engineering for Developers uses Python and the OpenAI API, and AWS Skill Builder's free Foundations of Prompt Engineering is built on Amazon Bedrock for developers and data scientists.

Prompt engineering certification: what you can actually earn

No AI company sells a prompt engineering certification to the public. OpenAI's only certification, OpenAI Certified, runs inside ChatGPT and is open only to ChatGPT Enterprise and Edu workspaces by invitation (OpenAI's help page). Anthropic's Claude Certification Program has four proctored exams, and its registration page says certification "is open to organizations in the Claude Partner Network" (Pearson VUE's Anthropic page).

The closest thing to an exam that anyone can take is Microsoft's AI Business Professional certification, exam AB-730: a 45-minute proctored exam for business users, no coding required, that tests prompting and chats in Microsoft 365 Copilot (Microsoft Learn). It is a real certification, but only for Copilot. Everything else is a certificate of completion:

What a prompt engineering course should teach: we tested it

Most prompt engineering courses teach you to word a request well. We wanted to know how far that gets you on real work, so we gave Claude one ordinary job three ways: draft replies to three customer emails for Fernwood Flowers, a two-person florist we made up.

The emails were chosen so that a prompt alone can't get them right. A couple asks what wedding flowers cost (the items add up to $780, under the shop's $1,200 wedding minimum). Dan's morning delivery arrived at 6:40 pm (policy: refund the $15 delivery fee and the $10 morning slot, plus a 20% store credit). Lena wants a same-day bouquet at 2:30 pm (the cutoff is 1 pm). We scored every reply on 12 checks: for each email, the figures, the policy, the voice (signed "Mara at Fernwood Flowers", no exclamation marks) and a 120-word limit.

Version 1: a vague prompt, no context

We typed Draft replies to the three emails in inbox/. in a folder that held only the three emails. Claude did not make anything up. It said the folder had no price list or policies, saved three drafts with 23 blanks such as [REMEDY] and [PRICE RANGE], signed them [NAME], and wrote 123 to 167 words each. Useful as a starting point, but none of the 12 checks passed.

Version 2: the kind of prompt a course teaches

Next, a fresh folder and the prompt a good course would have you write: who you are, who reads the reply, the tone, a length limit, the sign-off, and an instruction to mark missing facts instead of guessing.

You are helping me, the owner of [BUSINESS], reply to customers. Draft a reply to each message below. Readers are [WHO THEY ARE]. Keep each reply [TONE], under [NUMBER] words, no exclamation marks, signed "[NAME]". Use only facts I give you: if a price, fee or policy is missing, write [CHECK: what is missing] instead of guessing. Show the drafts here and do not send anything.
[PASTE THE MESSAGES]

It worked as far as a prompt can. All three replies were signed correctly, stayed between 77 and 107 words, and had no exclamation marks. But every price, fee and policy came back as a [CHECK] marker, nine of them, because the prompt couldn't supply what the shop charges or how it handles a late delivery. Six of 12 checks.

Version 3: the same short prompt, with context

Last, the one-line prompt from version 1 again, in a folder that also held the four context items above, 398 words in all. Claude read them first, then got every figure and policy right: the $780 total against the $1,200 minimum, the free consultation and the 30% deposit for the wedding, the $25 refund and 20% credit for Dan, and "past the 1 pm cutoff, $65 plus $15 tomorrow" for Lena. It left two [CHECK] notes for things no file covered: whether June 12 is free, and how the refund gets back to Dan.

We ran each version three times to make sure this wasn't luck:

Two lessons for anyone shopping for a course. First, today's models rarely invent your prices when they don't know them; they leave blanks. So a course that spends hours on tricks against made-up answers is solving yesterday's problem. Second, wording took us from 0 to 6 checks and context took us from 6 to 12. A practical course has to teach both.

The syllabus checklist that follows

Tick these off against a course's lesson list before you pay:

Free prompting guides from OpenAI, Anthropic and Google

Before you pay for a course, read what the three companies themselves recommend. Each publishes a free prompting guide, kept up to date for its newest models, and they agree on more than they differ:

These guides are references, not courses: no exercises, no feedback, no certificate. A course earns its price when it makes you apply them to your own work and checks the result.

How to check a prompt engineering course before you pay

Prompting advice has changed with the models, so an old course can teach habits that now backfire. Watch a free preview lesson and look for these signs:

• GPT-4o in the model picker: it left ChatGPT on February 13, 2026 (OpenAI's retirement notice), so those screens are older than that.
• A lesson that builds a custom GPT: personal plans can no longer create new ones, and custom GPTs retire on December 11, 2026 (OpenAI's FAQ).
• Rules shouted in capital letters: Anthropic's guide says recent Claude models respond more strongly to instructions, so "CRITICAL: you MUST" wording should be softened, not taught as a trick.
• Role-play and tag templates for every request: Anthropic's prompting post for 2026 calls heavy role prompting and XML tags less necessary with modern models.
• One prompt in an empty chat, lesson after lesson: no files, projects or saved instructions means the course skips the half that took our test from 6 to 12.
• A job title as the product: "become a prompt engineer" promises are a sales line, not a syllabus.

To test a specific course, paste its lesson list into the assistant you use:

I'm choosing a prompt engineering course. Here is its lesson list: [PASTE THE LESSON LIST]. It was last updated in [MONTH YEAR].
Facts as of October 2026: GPT-4o left ChatGPT on February 13, 2026, and custom GPTs retire on December 11, 2026. Anthropic and Google now advise softer wording instead of capital-letter rules, examples over long descriptions, and long documents before the question. Assistants keep standing instructions and files in projects, custom instructions or files such as CLAUDE.md.
List the lessons that are likely outdated, the lessons that still matter, whether I will practice giving the AI my own files and instructions, and 3 questions to ask the seller before I pay. I use [ASSISTANT] for [WHAT I DO AT WORK].

For warning signs that apply to any paid course, see our guide on whether AI courses are worth it.

What to do next

This week, pick one task you repeat, such as answering one kind of customer email, and run our three-version test on it with the scoring sheet in this guide's bonus. If a good prompt alone gets you most of the way, a free course and the official guides above are enough. If the version with your files wins by a wide margin, as it did for us, choose a course that teaches standing instructions, files and examples. Our Claude Code course, in early access, teaches that with CLAUDE.md files and skills on real business tasks, and our guide on how to learn AI has a 30-day plan to practice either way.