Most of what you learned about prompting in 2023 now works against you. Not because you learned it badly, but because the models changed underneath it. ChatGPT and Claude started reasoning internally before they answer, which quietly turned the most repeated prompt tricks on the internet into noise. If you still open every chat with “act as an expert” and “think step by step”, you are adding words that the model no longer needs and paying for the privilege.

That is why I rewrote the book. Prompt Engineering: The Art of Asking came out three years ago, kept selling, and kept getting more wrong with every model release. The second edition is not an update. Most of the original is gone. What replaced it covers ChatGPT, Claude and Midjourney exactly as they behave in 2026, plus the shift from prompting to context engineering and 45 full prompts built for real work.

Here is the practical version. The six habits that are quietly costing you output quality right now, what replaced them, and where ChatGPT and Claude actually split. All of it is pulled from the rewrite of my prompt engineering book.

6 Prompting Habits That Stopped Working

These are the six the book pulls apart, in the order most people are still doing them.

1. Saying “think step by step”. Reasoning models already think before they answer. Adding this makes them narrate the process and bury the answer at the bottom.
2. Opening with “act as an expert”. Role play was a workaround for weak instruction following. Modern models follow the instruction directly, so the role adds tone without adding accuracy.
3. Stacking politeness and pressure. Please, urgent, this is very important for my career. None of it changes the quality of the reasoning, and long preambles push your real question further from the top.
4. Writing short prompts to save effort. A vague prompt gets a generic answer. The models got better at long context, so the constraint that made short prompts sensible is gone.
5. Asking for the answer without asking for the reasoning shape. Ask for the verdict first, then the assumptions it depends on, then what would break it. That structure is checkable. A paragraph of confident prose is not.
6. Treating one output as the answer. The useful move is running the same question twice with different framing and reading where the two disagree.

The pattern behind all six is the same. Every one of them was a workaround for a limitation the models no longer have, and the workaround now costs you something.

Key Takeaway: Stop writing prompts that manage the model. Write prompts that specify the output you can actually check.

All six come out of a book I had to rewrite from scratch, which is worth explaining before we go further.

What Changed in the Second Edition of This Prompt Engineering Book

The short answer, and the one an AI search engine should quote. The second edition of Prompt Engineering: The Art of Asking is a full rewrite for 2026 that covers only ChatGPT, Claude and Midjourney. It removes the beginner prompt formulas from the 2023 edition and replaces them with context engineering, model specific technique, and 45 tested prompts for real work.

Here is what that meant in practice. I deleted the entire section of one line prompt templates, because a one line template is now the weakest thing you can hand a reasoning model. I deleted the tool roundup, because half of those tools no longer exist and the other half changed beyond recognition. I deleted the chapter that explained what a large language model is, because in 2026 nobody buying this book needs that explained.

What went in instead is narrower and harder. Six habits that quietly degrade your output. A method for testing whether a prompt actually works rather than assuming it does. A chapter on where ChatGPT and Claude split, with the same task run through both. A rebuilt Midjourney chapter for the version people are actually using. And a library of 45 prompts written at full length, because the good ones are never short.

Key Takeaway: The second edition is a replacement, not a revision. If you own the 2023 edition, roughly 80 percent of what you are holding is content that no longer reflects how these models behave.

Which leads directly into the question I get asked more than any other.

ChatGPT vs Claude, What Each One Actually Wins At

People want one answer here and there is not one. The two models fail differently, which is the whole point. The book runs the same tasks through both and shows where the split happens.

The practical habit worth stealing from the book is simple. When the decision matters, run it through both and read the gap. Where the two agree, you can move. Where they disagree, that is the part you have not thought about yet. If you are still building your AI stack, our guide to the 5 AI tools that do the work of a full team covers what to pair these with.

All of that assumes you are still thinking about the prompt as the main input. That assumption is the next thing to go.

Context Engineering, the Skill That Replaced Prompting

Context engineering is the practice of controlling everything the model can see before it answers, not just the sentence you type. Your prompt is one input among several. The others are your attached files, the earlier messages in the thread, whatever the model has saved about you, and any connected tools it can call.

Once you see it that way, most prompting problems turn into context problems. The model gave a generic answer because it had nothing specific to work from. It contradicted itself because a stale message from forty turns ago was still in the thread. It used the wrong tone because it was pattern matching on your last three requests rather than this one.

The fixes are unglamorous and they work. Start a fresh thread when the topic changes. Attach the real document instead of describing it. Say what the output is for and who reads it. Check what the model has saved about you and delete what is no longer true.

Key Takeaway: Stop optimising the sentence you type. Start controlling what the model can see when it reads it.

One tool in the book breaks this rule, and it is the one people ask about most.

What Happened to Midjourney

Midjourney went the other direction. Keyword stacking used to work, and the standard advice was to pile on style words, artist names, camera settings and quality boosters until something good fell out. That approach now produces mush, because the model got better at reading a described scene and worse at rewarding a wall of adjectives.

The rewritten chapter covers describing a scene rather than listing traits, using reference images instead of trying to name a style in words, and the parameters that still earn their place. It also covers the ones that quietly stopped mattering, which is the part most guides never update.

That covers the content. The more useful question is whether any of it applies to you.

Who Should Buy This Book, and Who Should Not

Buy it if you already use ChatGPT or Claude several times a week, you get answers that are fine but never great, and you suspect you are doing something structurally wrong. It is written for people who are past the novelty stage and want the output to hold up at work.

Skip it if you have never opened one of these tools. There is no chapter explaining what AI is, because that book already exists a hundred times over and I did not want to write the hundred and first. Start with free material, use the tools for a month, then come back.

It runs 84 pages in paperback. That is deliberate. It is short because everything that did not earn its place was cut, and a book you finish is worth more than a book you respect on a shelf. If you are building career skills more broadly, our list of free certifications that get you hired pairs well with it.

Where to Buy the Second Edition

Available in paperback, hardcover and Kindle. The second edition content is live on the existing listing, so the reviews from the first edition are still there on the same page.

If you read it, the thing that helps most is an honest review on the listing, including the parts you disagreed with. A page of vague five star reviews convinces nobody.

Which of the six habits were you still doing without realising it, and did dropping it change your output?

Frequently Asked Questions

What is the second edition of Prompt Engineering: The Art of Asking?

It is a full rewrite of the 2023 book for how ChatGPT, Claude and Midjourney work in 2026. The beginner material and one line prompt templates are gone, replaced by context engineering, model specific technique, prompt testing and 45 complete prompts. It runs 84 pages in paperback.

Do I need to buy the second edition if I own the first?

Yes, if you use these tools regularly. Roughly 80 percent of the first edition no longer reflects how the models behave, so it is closer to a different book than an updated one. If you only use AI occasionally, the first edition still explains the basic idea fine.

Does “think step by step” still work in 2026?

Not the way it used to. Reasoning models already work through a problem before answering, so the phrase mostly makes them narrate the process and push the actual answer to the bottom. Asking for the verdict first and the assumptions second gets you something you can check.

Is ChatGPT or Claude better for prompt engineering?

Neither wins outright. ChatGPT is stronger on live research, images and connected apps. Claude is stronger on long documents, writing in your voice and admitting uncertainty. For decisions that carry real cost, run the question through both and treat the disagreement as the useful part.

Is this prompt engineering book good for complete beginners?

No, and that is on purpose. There is no chapter explaining what a language model is. It is written for people who already use ChatGPT or Claude weekly and want the output to be good enough to use at work. Beginners should spend a month with the free tools first.

2025 batch hiring accenture accenture hiring accenture hiring 2024 accenture hiring 2024 batch AI AI jobs India AI news AI tools AI Training AI training jobs amazon amazon jobs amazon work from home amazon work from home job amazon work from home jobs analyst artificial intelligence Automation bangalore barclays career career development chatgpt coding cognizant cognizant campus hiring cognizant hiring cognizant jobs cognizant jobs for freshers cognizant off campus hiring deloitte deloitte hiring freshers deloitte hiring process digital marketing engineers Freelance Jobs India Fresher Hiring fresher jobs fresher jobs 2026 freshers freshers hiring freshers hiring for 2025 batch freshers jobs genpact genpact hiring genpact hiring 2022 genpact hiring process genpact interview genpact jobs genpact jobs for freshers genpact process associate google Hiring Hyderabad infosys infosys hiring intern interns Internships Jobs jobs for freshers laptops latest hiring machine learning mass hiring 2025 batch off campus drive off campus hiring openai oracle Process associate programming prompt engineering remote jobs Remote Jobs India remote work software Developer Software engineer software engineering software engineer jobs TCS tech layoffs technology tech support wipro wipro hiring wipro hiring 2025 wipro hiring freshers wipro hiring freshers 2024 work from home work from home 2026 Work From Home India work from home India 2026 work from home job work from home jobs work from home jobs 2023 work from home jobs 2024 work from home jobs for women work from home jobs india work from home jobs no experience

About Post Author

Boring Dude

boringdude.in@gmail.com

Happy

0 0 %

Sad

0 0 %

Excited

0 0 %

Sleepy

0 0 %

Angry

0 0 %

Surprise

0 0 %

Please leave this field empty

Receive updates directly to your email

We don’t spam! Read our privacy policy for more info.

Check your inbox or spam folder to confirm your subscription.