Back to index
OtherReference site, a weekend to read the core sections properly·Free (paid partner courses sold separately)

Prompt Engineering Guide

4.3

If you have read anything credible about prompting in the last three years, some of it traces back here. It is a well maintained, paper backed reference that covers the actual techniques rather than the usual list of magic phrases. It is a reference and not a course, and that matters for how you use it.

What We Liked

  • Comprehensive, free, and organised technique by technique instead of by vibe
  • Almost every technique cites the paper it came from, so you can go from a paragraph to the source in one click
  • The Prompt Hub gives you working examples grouped by task type rather than abstract advice
  • Genuinely maintained, so newer material on agents, function calling and RAG sits alongside the older basics
  • Open on GitHub and translated into a long list of languages, which means mistakes get caught and fixed

What Could Be Better

  • It reads like a wiki, so there is no learning path, no exercises and nothing checking whether you understood
  • Depth is uneven, and some techniques get a short paragraph and a link where they deserve a walkthrough
  • The academic framing means a chunk of what is listed are research curiosities you will never reach for
  • Paid academy courses are promoted inside the free pages, which blurs the line between reference and funnel
  • Some example prompts clearly date from earlier model generations and are less relevant to current models

Detailed review

The Prompt Engineering Guide has earned its position as the default reference for this topic, and I say that as someone generally suspicious of anything with the words prompt engineering in the title. What makes it credible is the sourcing. When it describes chain of thought prompting, self consistency, tree of thoughts or ReAct, it names the paper and links it. That single editorial decision separates it from the enormous volume of prompting content that presents someone's personal habit as a discovered law.

You can check the claims, which means you can also work out which ones matter. The structure moves sensibly from prompt elements and general design tips into specific techniques, then into applications like RAG, function calling, fine tuning and agents. The Prompt Hub is the part I use most often, because it groups worked examples by task type, so if you need a classification prompt or a reasoning prompt you can start from something that already functions rather than a blank box. The research and papers section is a decent reading list in its own right if you want to go deeper.

Now the honest limitations. This is a reference site and it behaves like one. There is no sequence, no exercises, no feedback and no forcing function, so it is very easy to read the whole thing, feel informed, and be no better at prompting than you were before. Prompting is a skill you acquire by writing a prompt that fails, working out why it failed, and changing one thing at a time.

The guide can tell you which lever exists. It cannot make you pull it. Depth is also inconsistent. The core techniques are covered well, while some of the more exotic entries get a summary paragraph and a citation, which is fine for a reference and frustrating if you arrived hoping to implement it.

The academic bias cuts both ways too. It keeps the quality high, but a working practitioner will find that a large share of the catalogued techniques are things you would never actually use in production, where the winning approach is usually a clear instruction, good examples, and structured output rather than an elaborate reasoning scaffold. The commercial layer is worth flagging plainly. The guide is free and always has been, and there are now paid courses from an academy partner promoted through the site with discount codes.

That is a reasonable way to fund a free resource, and I would still rather they said so more clearly on the pages where it appears. Finally, prompting content ages faster than almost anything else in this field, because model behaviour changes underneath it. Some examples here are visibly built for older models, and a few techniques that mattered a great deal when models were weaker at reasoning matter much less now. The guide is better than most at keeping up, and it is not immune.

My 4.3 reflects a resource that is free, honest, well sourced and continually useful as a lookup. It loses points because it is a reference dressed in the language of a course, and anyone who treats it as a curriculum will finish it knowing a lot of names and very little craft.

[ final ]

The verdict.

Bookmark it, do not try to study it front to back. As a lookup for what a technique actually is and where it came from, nothing free is better. As a way to get good at prompting, you need a project to apply it to, because reading about few shot prompting teaches you almost nothing compared to using it badly and fixing it.