Prompt engineering course: The official guides and free courses worth your time

Plans, prices and features checked on the official pages on October 23, 2026.

The best prompt engineering course for most people is free: the official guides from Anthropic, OpenAI, Google and Microsoft, plus one short beginner course such as DeepLearning.AI’s ChatGPT Prompt Engineering for Developers. If you want a certificate, Vanderbilt’s course on Coursera is the best-known paid option, and its price is not shown on the page.

Prompt engineering is the craft of writing instructions so a model reliably gives the output you need. OpenAI describes it as combining iteration with testing, because model output is non-deterministic. The sections below cover the official guides, the courses, the certificate question, the core techniques, and where jobs fit in.

What prompt engineering is and who can learn it

The DAIR.AI Prompt Engineering Guide calls it "a relatively new discipline for developing and optimizing prompts" for using language models across many applications. In plain terms: you decide what to say to a model, how to structure it, what examples and context to give, and how to check the result.

Anyone can learn the basics. The Vanderbilt course on Coursera says only basic computer skills are needed, such as using a browser and accessing ChatGPT. The developer-focused courses add basic Python.

The craft sits next to a wider set of skills. Our piece on which skills to focus on instead of prompt engineering covers that debate. Anthropic’s own guide also says prompting is not always the fix: it notes that you can sometimes improve latency and cost more easily by choosing a different model.

The official guides are the best free prompt engineering course

Each major model maker publishes its own guide, and they are kept current in a way static courses are not.

• Anthropic: the prompt engineering overview assumes you already have clear success criteria, a way to test against them, and a first draft prompt. The techniques themselves (clarity, examples, XML structuring, role prompting, thinking, prompt chaining) live in a page called Prompting best practices, which Anthropic calls its living reference. It also links an interactive GitHub tutorial and a lighter Google Sheets version. Anthropic notes the GitHub tutorial was written for Claude 3 models, so follow the best practices page where the two differ.
• OpenAI: its prompt engineering guide covers message roles, structuring prompts, few-shot examples, adding context through retrieval, and operational practice. It advises pinning model snapshots in production and building evals before changing prompts or upgrading models.
• Google: the Gemini API prompting strategies page covers clear instructions, constraints, few-shot examples, context, and response prefixes. It also says Gemini 4 and Gemini 3.6+ ignore custom temperature, topP and topK values.
• Microsoft: Microsoft Learn’s prompt engineering techniques page covers instructions, examples, cues, supporting content and grounding. It warns that its techniques are not recommended for reasoning models such as gpt-5 and the o-series, and that the examples were tested on a base GPT-4 model.

That last point matters for anyone reading older tutorials. Guidance written for one model generation can be wrong for the next.

Courses compared: Length, level and what you get

These are the structured courses whose official pages were checked for this guide. Prices are only given where the provider states them.

Prices checked on October 23, 2026. Free periods and enrollment terms change, so confirm on the provider’s page before signing up.

A few notes on the table. The DeepLearning.AI developer course has 9 video lessons, 7 code examples and 1 graded quiz, and covers summarizing, inferring, transforming and expanding text, plus building a custom chatbot in a Jupyter notebook with the OpenAI API. It is the most efficient first course if you can read basic Python.

The Vanderbilt course, taught by Dr. Jules White, is the non-coder route. Coursera also says it is included with Coursera Plus and that financial aid is available in select programs.

Anthropic’s course is not about prompting as such. The 4D framework it teaches (Delegation, Description, Discernment and Diligence) treats "Description" as the part about telling a model what you want. Anthropic’s companion course on AI capabilities and limitations covers steerability and context limits.

Certifications: What exists and what employers see

No official "prompt engineering license" exists on the pages above. What you can earn are course certificates and badges, and they vary in what they prove.

• The Vanderbilt certificate on Coursera is shareable on LinkedIn or a resume and requires 7 assignments.
• DeepLearning.AI offers an accomplishment with a graded assignment for PRO members. The membership price is not stated on the course page.
• Google’s Grow with Google page lists the Google AI Professional Certificate on Coursera, and free skills badges on Skills.google such as AI Boost Bites. Duration and price for the Coursera certificate are not stated there.
• Anthropic’s Academy page mentions a quiz per course but no certificate.

Treat any of these as evidence that you finished a structured course. They do not replace a portfolio of prompts and evals that show you can get reliable results on a real task.

The techniques every guide teaches

Across the official guides the same core ideas repeat, which is a good sign that they are durable.

1. Be specific. State the task, the audience, the length and the format. Google advises constraints on both input and output, and Microsoft says to leave as little to interpretation as possible.
2. Show examples. Few-shot prompting means including input and output pairs. OpenAI says to use diverse examples so the model picks up the pattern, and Google warns against so many that the model overfits.
3. Give context. Reference text and grounding data make answers specific. Microsoft says supplying grounding data is one of the most effective ways to get reliable answers.
4. Structure the prompt. OpenAI suggests Markdown headers and lists for sections and XML tags to delimit documents or examples. Google recommends consistent delimiters, and Anthropic lists XML structuring among its techniques.
5. Break the task down. Google describes chaining sequential steps so each output feeds the next prompt. Anthropic also names prompt chaining.
6. Give the model an out. Microsoft suggests allowing a reply such as "not found" when the answer is not in the text, to reduce invented answers.
7. Test every change. OpenAI and Anthropic both stress evals. Without a way to measure, you cannot tell whether a new prompt is better.

The DAIR.AI guide goes deeper into named research techniques such as chain-of-thought, self-consistency, ReAct, Reflexion and retrieval-augmented generation (RAG). It also has a context engineering guide, which is the entry point for the prompt engineering versus context engineering question. The short version is that prompting is the wording of the instructions, while context engineering is about everything else the model sees. Our explainer on the AI agent harness shows how that setup around a model affects results.

How to get started in a weekend

1. Pick one model you already use and read its official guide: Anthropic, OpenAI, Google or Microsoft, as above.
2. Take a short course. The 1 hour 40 minute DeepLearning.AI course suits developers, and Anthropic’s AI Fluency course suits everyone else.
3. Choose a real task, such as summarizing meeting notes or classifying support emails, and write a first prompt.
4. Define what a good result looks like and collect 10 to 20 sample inputs. This is the "success criteria and tests" step Anthropic says to do first.
5. Change one thing at a time (instructions, examples, order, format) and compare outputs against your samples.
6. Re-check your prompts when you switch to a new model version, since guidance differs between models.

Limits, model differences and the future of the job

Reasoning models change the advice. OpenAI says reasoning models do best when given goals and trusted with the details, like a senior colleague, while GPT models need precise, step-by-step instructions, like a junior colleague. Microsoft goes further and says its classic techniques, including chain-of-thought, are not recommended for reasoning models.

The official pages do not give salary or job-count figures, so none are stated here. Treat any salary number you see as coming from the site that publishes it and check how it was collected. What the guides do show is where the skill is used in practice: writing system prompts for apps, building evals, and managing context for agents. Those tasks are rarely a standalone title, and more often part of a developer, analyst or product role.

Whether prompt engineering becomes obsolete depends on what you mean. Writing clever phrases matters less as models get better at following plain requests. Describing a task clearly and verifying the result stays useful, which is why the guides keep returning to success criteria and testing. Our look at why people are forgetting how to code makes a related point about skills that outlast tools.

Common questions