shribe!

Everyone boasts about how AI can write a whole research paper in a few minutes.

But if it’s so good, why hasn’t everyone published five times in Nature and Science last year? (or gotten straight A’s?)

In this video, I’ll show you why.

To do so, I will uncover the three reasons that keep you from being successful with AI-supported writing.

Before we get into the three reasons, let’s look at what triggered this whole discussion.

A task force at Organization Science, one of the leading journals in management research, analysed what happened to submissions after ChatGPT became widely available.

And the first thing they found was simple:

More papers.

A lot more papers.

Since the launch of ChatGPT in November 2022, submissions to Organization Science increased by 42 percent. For comparison, even during COVID, when many academics suddenly had fewer conferences, less travel, and more time at home, submissions increased by only 20 percent.

So at first glance, this looks like a productivity revolution.

AI arrives, and suddenly people produce more research.

But here is the problem.

More output does not automatically mean better output.

And this is where the report gets really interesting.

Because when the authors looked more closely at the submissions, they found that the increase was not evenly distributed across all papers.

The growth came mainly from manuscripts with substantial AI-generated text.

In other words, the journal was not just getting more papers.

It was getting more AI-heavy papers.

And that leads us to the first reason AI-supported writing often fails.

Reason #1: AI makes it easier to produce text, but not necessarily easier to produce quality.

This is the trap many aspiring researchers and students fall into.

When AI gives you a polished paragraph, it feels like progress.

You see clean grammar.

You see academic wording.

You see structure.

You see citations, transitions, and confident sentences.

And your brain thinks: “Great, that saved me so much time.”

But often, something else is happening.

The work is getting longer.

The work is getting smoother.

The work is getting more finished-looking.

But the actual idea underneath may not have improved at all.

And that is dangerous, because good writing is not just about producing sentences.

Good writing is thinking made visible.

It is where you notice that your argument is weak.

It is where you realise that two ideas do not actually connect.

It is where you discover that a paragraph sounds good, but does not really say anything.

If AI does too much of that work for you, you may skip the uncomfortable part where the real improvement happens.

So this first graph already tells us something important:

AI can remove friction from writing.

But sometimes that friction was the part that forced you to think.

---

Now let’s move to the second part of the report.

Because if AI-heavy submissions only increased in volume, you might say:

“Okay, maybe there is more noise in the journal from low-quality submission. But if I use AI, perhaps it will still improve my writing.”

That would be the optimistic version.

You would think that more AI would also cleaner language.

More polished abstracts.

More readable manuscripts.

No more typos.

And to be fair, this is what most of us expect from AI.

But the report found something surprising.

After the launch of ChatGPT, the readability of abstracts submitted to Organization Science actually declined.

Not improved.

Declined.

And this is the second reason AI-supported writing often fails:

Reason #2: AI can make your writing sound better while making it harder to understand.

That sounds strange, but you have probably seen this before.

AI loves sentences that sound like this:

“This study contributes to the ongoing conceptualisation of dynamic organisational transformation through a multidimensional lens.”

It sounds academic.

It sounds serious.

It sounds like something that belongs in a journal.

But after reading it, you have to ask:

What does that actually mean?

And that is the problem.

AI is very good at producing what I would call “academic texture”.

It gives you the surface features of academic writing:

longer words, smoother transitions, more abstract nouns, more confident framing.

But that does not automatically create clarity.

In fact, it can do the opposite.

The report shows that AI-heavy writing tends to use longer words, more complex sentence structures, more jargon, and more nominalisations.

You know, those words we somehow start using in academia and then pretend they are normal human language:

“conceptualisation”

“contextualisation”

“operationalisation”

“problematization”

Individually, these words can be useful.

But when they pile up, the writing becomes heavy.

And this is why AI text can feel so deceptive.

So what happens when you write with AI is that you stop too early.

You accept the paragraph because it sounds professional.

But the reader still has to work too hard to understand what you are trying to say.

And in academic writing, that is a serious problem.

Because your goal is not to sound impressive.

Your goal is to make a difficult idea understandable.

That is the difference between writing that merely looks academic and writing that actually communicates expertise.

---

Now let’s look at the part of the report where the problem becomes visible.

Because up to this point, you might still think:

“Okay, maybe AI writing is harder to read on average. But maybe that is just because academic writing is hard anyway.”

So the authors included a very simple comparison.

They took the title and abstract from a real published paper and compared it with a version generated by ChatGPT.

And this example is fascinating because the AI version is not obviously bad.

At first glance, it sounds quite impressive.

The human-written version is about loss-framed performance incentives.

In simple terms, it studies whether bonuses work better when people receive the money upfront and lose it if they do not meet their targets.

The original abstract explains this quite directly.

It says what the researchers tested.

It says where they tested it.

It says what they found.

And it says why the result matters.

Now compare that to the AI-generated version.

The AI version uses phrases like:

“theoretically incomplete”

“substantial heterogeneity across task dimensions”

“reference-dependent agents engaged in multidimensional tasks”

“inefficient reallocation of effort”

Again, none of this is necessarily wrong.

But it creates a different reading experience.

The report also gives us a number for this.

The original human-written abstract had a Flesch Reading Ease score of 22.55, which placed it in the 87th percentile of the journal’s sample.

The ChatGPT version had a score of minus 2.19, placing it only in the 24th percentile.

In simple terms:

The AI version was not just a little harder to read.

It was dramatically harder to read.

The human version feels like someone is trying to help you understand the argument.

The AI version feels like someone is trying to sound like a journal article.

The best writing gives the reader a clear path.

The weaker version gives the reader a fog machine.

And that leads us to the third reason AI-supported writing often fails:

Reason #3: AI can hide weak thinking behind strong language.

This is maybe the most important point in the entire video.

Because when your writing is messy, you know it needs work.

But when your writing is polished, you may assume the thinking is finished.

So where does all of this leave us?

Should you stop using AI?

Absolutely not.

In fact, even the authors of the report are enthusiastic about AI. They describe it as a scientific superpower that has fundamentally changed how they conduct research and teach. And honestly, I agree.

AI can help you brainstorm ideas.

It can help you summarize information.

It can help you improve clarity.

It can help you identify weaknesses in an argument.

And it can save enormous amounts of time.

The lesson of this video is not that AI is the enemy.

The lesson is that AI is most useful when it supports your thinking, not when it replaces it.

Because if there is one pattern that appears throughout the report, it is this:

AI makes producing text easier.

But producing text was never the ultimate goal.

The goal is producing insight.

The goal is producing understanding.

The goal is producing ideas that are genuinely worth communicating.

And those things are still difficult.

They require judgment.

They require creativity.

They require critical thinking.

And perhaps most importantly, they require struggle.

That may sound strange, but think about the best idea you have ever had.

It probably didn’t appear because a machine instantly generated the answer.

It appeared because you wrestled with a problem.

Because you noticed a contradiction.

Because something didn’t make sense.

Because you spent time thinking.

The uncomfortable truth is that many of the moments that lead to real learning feel inefficient.

They’re the moments where you’re confused.

Where you’re stuck.

Where you’re rewriting the same paragraph for the third time.

But those are often the moments where the real intellectual work happens.

And that is why I believe the biggest risk of AI is not that it will replace human intelligence.

It’s that it will tempt us to bypass the very processes that develop it.

The future probably won’t belong to the people who can generate the most text.

AI can already do that.

The future will belong to the people who can ask better questions.

Think more deeply.

Judge ideas more critically.

And know when the machine is helping—and when it’s merely creating the illusion of progress.

So use AI.

Experiment with it.

Learn how to get the most out of it.

But don’t outsource the part that actually makes your work valuable.

Because in the end, great writing isn’t created by typing words onto a page.

It’s created by the thinking that happens before those words appear.

And that’s something no AI can do for you.

---

AI can generate your sentences. But it still can’t generate your best ideas.

Every week, I read texts from students and PhD candidates that I have to grade or give feedback on. Lately, my evaluations have unfortunately often not been very positive.

Entire paragraphs sound like polished, empty, and impersonal standard output. They contain words and sentence structures I’ve never heard that person use before.

Many students seem to think: as long as it sounds academic. As long as it sounds somehow smart. And what you get is a text that may sound impressive on the surface, but immediately creates this feeling when you read it: no one is writing here with their own voice. This is someone writing like a glossy bot with a reference list.

And that’s extremely frustrating for supervisors, because they read the same types of texts every day. They develop a sense for whether a sentence is actually trying to say something—or just pretending to.

That’s exactly why today I want to show you five typical AI phrases that are really terrible, why they’re so annoying to read, and how you can replace them so your text actually sounds like you again and stands out from the crowd.

---

The real problem is not AI. I use AI all day long. The real problem is poor writing style—or rather, a lack of awareness of how your writing comes across to others.

A supervisor doesn’t automatically hate a text because a tool may have helped. What’s annoying is sterile, polished, interchangeable writing that sounds like zero personal effort. And once you understand that, you’ll immediately write better than most others.

---

#1 A lot of fluff for nothing

Let’s start with the first category: “importance phrases.” Sentences like:

• “It is important to emphasize…”
• “It is crucial to note…”
• “It should be highlighted at this point…”
• Or: “In today’s fast-paced world…”

At first glance, these sound serious—but they have a problem: they announce importance instead of delivering it. They tell you: “Attention, this is important now.” But when you break the sentence down, there’s often nothing there.

Why is this annoying? Because it wastes space. Academic writing is not theatre fog. A good sentence delivers an observation, an interpretation, or an argument—ideally based on a source you’ve actually read.

A bad sentence just creates an atmosphere of importance.

If you write:
“It is important to emphasize that social media influences adolescents,”
a supervisor won’t think: “Wow, how nuanced.”
They’ll think: “Yes, thanks. And now?”

What exactly is your point? What kind of influence? Under what conditions? Why is it relevant to your research question?

Much better:
“Social media influences adolescents particularly when social comparison and the need for validation play a central role in their daily lives (source).”

Now there’s an actual idea—grounded in research. The sentence has substance.

---

#2 “Not only… but also”

“The method is not only efficient, but also flexible and practical.”

The problem isn’t that this structure is grammatically wrong. The problem is that it often sounds like a pre-built AI template. It feels clean, symmetrical, and polished—but when you read it 500 times, it becomes unbearable.

If you write:
“This theory is not only useful for analysis, but also relevant for practical application,”
it sounds academic—but what does that actually mean?

Which analysis? Which application? Why exactly?

Even worse:
“This finding does not mean X, but Y.”
If it doesn’t mean X, then just leave X out.

To avoid this, you can literally prompt your AI tool not to use such structures. It’s that simple.

The sad part? It takes about five minutes to set up a custom GPT so it doesn’t write like everyone else—but like you want it to.

If you want better outputs from AI, you need some basic prompt engineering skills—and you need to understand what good writing looks like. Otherwise, you can’t give the AI proper instructions.

---

#3 The triple list

Another classic: three-part lists.

Sentences like:

• “The topic is relevant, complex, and multifaceted.”
• “The results show opportunities, challenges, and potentials.”
• “Academic work requires structure, discipline, and reflection.”

This structure sounds smooth and complete—which is exactly why AI uses it all the time.

The problem: it often sounds generic. Three abstract terms, usually neutral, so broad they could fit anywhere.

Once? Totally fine.
Every second sentence? Your whole text starts sounding like it came from a generator.

This is another thing you can (and should) explicitly forbid in your AI meta-prompt.

---

#4 Synonyms

This one is tricky—because it looks like good style.

AI loves constantly switching terms to make a text sound “nicer”:

• “students” → learners, university attendees, academic trainees
• “interview” → conversation, survey, data collection
• “motivation” → drive, willingness, engagement, ambition

And this is the death of academic writing.

Why? Because academic writing isn’t about sounding pretty—it’s about conceptual precision.

If you define a term once, it has a fixed function in your paper. It’s a precise tool. And you don’t constantly swap out precise tools just to make a sentence sound more elegant.

Consistency matters more than variation.

Example:
You define “social support” at the beginning.
If you then constantly switch to “help,” “backing,” “support systems,” “assistance,” etc., your text becomes unclear.

These terms are not identical. They might overlap—or they might not. Either way, it creates confusion.

Unlike the other AI phrases (which are just annoying), this is actually wrong academically.

If I read a scientific paper full of unnecessary synonyms, it gets marked down. Significantly. That was true before AI—and it still is.

That’s why someone with experience + AI gets much better results than someone without experience using the same AI.

Better approach:
Define “social support” clearly once—and then stick with it. Even if it feels repetitive.

Academic writing doesn’t need to sound literary. It needs to be clear.

If a term is central, repetition is a strength—not a weakness.

Language models are trained to sound fluent, varied, and elegant.
Academic writing often requires the opposite: repetition, precision, and consistency.

If you let a default model write freely, you’re setting yourself up to fail.

You need to give it context, examples, detailed prompts—and ideally your own structure, even if it’s painful to start without AI.

---

#5 Punctuation overload

The last category isn’t about phrases, but punctuation patterns:
dashes, semicolons, colons.

AI texts often overuse them to extend sentences or make them sound more “elegant.”

None of these are wrong on their own. The problem is frequency.

Some academics who used to love dashes now avoid them—just to not sound like AI. Sad, but true.

I’ve even seen people (e.g., some of my Chinese PhD students) speak the way AI writes—including “not only… but also” structures. Also sad. Also true.

You don’t need to avoid these punctuation marks completely—but use them sparingly.

If every second sentence contains a dash, semicolon, or colon, something is usually off at the sentence level.

Instead of:
“There is one central problem: many students start too late,”
just write:
“The central problem is that many students start too late.”

Instead of:
“The theory is useful; it structures the analysis,”
write:
“The theory is useful because it structures the analysis.”

Suddenly, the text sounds more natural.

And that’s what makes the difference.

Don’t try to sound as polished as possible—write so that someone can clearly follow your actual thinking.

You let the first half of your thesis timeline slip by, you’re buried in literature, ChatGPT keeps spitting out nonsense, and your supervisor has basically given up on you? It’s time for some real thesis tips that actually help.

That’s exactly why this post gives you seven tips for your Bachelor’s thesis that no one else will tell you – but that can actually help you reduce the pressure and finally feel like: “Okay, I’ve got this.”

1. Your Bachelor’s Thesis Is Just a Training Exercise

The biggest misconception students have is believing their Bachelor’s thesis needs to be a groundbreaking scientific contribution. Maybe it’s the term “academic thesis,” or maybe it’s just the unrealistic expectations we place on ourselves.

But here’s the truth: your thesis is, first and foremost, proof that you’re capable of doing academic work. Nothing more, but also nothing less.

What does that actually mean?

• You need to show that you can develop a meaningful research question.
• You need to research, analyze, and correctly cite relevant literature.
• You need to apply a method in a clear, comprehensible way to answer your question.

What you don’t need to do:

• Develop new theories
• Open up an entirely new field of research
• Or produce world-changing insights

Once you let go of the idea that your thesis has to be perfect, things get a whole lot easier. The goal isn’t to reinvent the wheel, but to work cleanly and transparently.

Pro tip: Write this sentence on a sticky note and slap it on your laptop screen: “Good is good enough.”

2. Your Research Question Is the Key

A lot of people think the topic is the most important thing. They spend weeks hunting for the perfect one.
“I want to do something with social media… or maybe sustainability?”

But let’s be honest: the topic alone won’t get you far. What really matters is the question you ask.

It’s the clearly defined question that turns a general topic into an actual research project. It gives your thesis direction, narrows the scope, and determines what literature you need and which method you should use. If you linger too long on a vague topic, you’ll lose your thread and start thinking in all directions at once—without making any progress.

Example: “Loneliness in remote work” – sure, that’s a topic. But what exactly is it about?

If you ask: “How does working from home affect the sense of belonging among entry-level employees in service agencies?” – then you suddenly have a clear direction. Something you can actually study.

So it’s really worth putting in extra care at this stage. If you can clearly and precisely formulate your question, everything else will become a lot easier. And your research question can be more niche than you might think – as long as it’s important to a specific target group.

3. Your Supervisor Is Not Your Personal Coach

Many students expect supervision to involve regular meetings, constructive feedback, and guidance during difficult phases. The reality? Usually… far less ideal.

Supervisors are usually extremely busy. They’re working on their own research, teaching classes, writing grant proposals – and supervising ten other theses besides yours. That doesn’t mean they don’t want to help. But they often just don’t have the capacity to walk you through every step.

That’s why one of the most underrated thesis tips is this: don’t wait around for help and manage the relationship actively.

Ask specific questions, be well-prepared, and provide clear interim results when needed. Instead of saying “I’m unsure about my outline,” say something like: “I’m considering combining chapters 2 and 3 – what’s your take on that?” These types of questions are easier to answer and increase the chances of getting genuinely helpful feedback.

The better you communicate, the more likely you are to get useful responses.

You need to act as the project manager of your own thesis. And that means: making your own decisions.

4. Writer’s Block Has Causes

Hardly anyone just breezes through their Bachelor’s thesis. Most people hit a phase where nothing works anymore. You open your document, stare at the screen, and have no clue how to even begin. But these blocks have causes – and they can be deciphered.

It’s usually not laziness or a lack of motivation – it’s uncertainty. You don’t know exactly what to write. Or you’re afraid of getting something wrong. Or maybe you’re just overwhelmed by the amount of information.

So what helps? Just start writing with no pressure.

Type out your thoughts, even if they’re still rough. Or start by jotting down a few bullet points outlining the general direction. The main thing is: get something written. Use an AI tool to get feedback – ChatGPT won’t judge you!

Be patient with yourself and remember: writing helps you think. Not everything needs to be perfect from the beginning. A lot of clarity comes only once things are on the page.

5. Rhythm Over Rigid Plans

The most carefully crafted weekly schedule is pointless if it doesn’t fit into your actual life. Many students plan based on an ideal version of themselves: they assume they’ll be equally productive every day, that motivation will always be available on demand, and that the writing process will unfold in a straight line. But reality is different. Some days you’re in the flow – and some days, you’re not. And that’s totally fine.

One of the most practical thesis tips is this: good time management doesn’t mean scheduling every hour of your day. It means creating realistic time slots when you can work – and building in space for setbacks and breaks. A writing day where you only get one page done can be just as valuable as one where you restructure two chapters or sort through literature.

Grab the books Deep Work and The One Thing – or just a summary. That’s really all you need.

6. The Biggest Time Trap? Literature

You want to start your research – and suddenly you’ve found 50 articles, 20 of which sound super relevant, and each one references 10 more. Welcome to the literature jungle.

What many underestimate: It’s not the writing that eats up your time – it’s the endless searching, reading, and sorting. Eventually, you lose track of everything. AI tools won’t help either – they can’t tell a Organization Science article from some dodgy MDPI fluff piece.

The trick isn’t to read everything – it’s knowing what you don’t need to read. A solid research question helps enormously because it automatically narrows down your literature. You only need to read what brings you closer to answering your research question and what’s valued in your academic field. If not – toss it out, even if it’s thematically related.

A reference management program can also help you keep track and insert citations correctly – without spending hours formatting.

7. The Discussion Is Not a Summary

If you’ve made it to the discussion section, you’ve already come a long way. And yet – this is often the hardest part of the whole thesis. Because now it’s no longer just about presenting results. Now you have to interpret and critically reflect on them.

The discussion is where you show that you truly understand the scientific process. This is your chance to connect the dots and derive meaning. You relate your findings to previous studies or theoretical frameworks. You consider possible societal or practical implications.

Ask yourself: What do my results mean? How do they align with the existing literature – and where do they differ? What can we infer from them?

And yes – you’re allowed to point out uncertainties and limitations. Where might your method have been too narrow? What influencing factors couldn’t you account for? What could have been done differently?

The discussion is not the place for summaries – it’s the place for independent thinking. This is where your thesis starts to become genuinely interesting. Unfortunately, this is also the point where most students stop putting in effort.

One of the most overlooked thesis tips? Make the discussion the heart of your thesis, and you’ll go from borderline pass to top-grade candidate.

Internal and external validity, these terms pop up all the time in research methods courses, but what do they actually mean?

If you’ve ever wondered how to tell whether a study is truly solid or just looks good on paper, you’re in the right place.

In this article, we’re cutting through the jargon and getting to the heart of what internal and external validity really mean, clearly explained, with examples that actually make sense.

So whether you’re working on a research project or just trying to wrap your head around these concepts for an exam, let’s break it down together.

What was validity again?

Exactly—it’s one of the three quality criteria used to assess the rigor of scientific studies.

Validity answers the question:

How well does a method measure what it is supposed to measure?

Or put differently: Can I answer my research question using this approach?

In another article, I’ve already given a general overview of the three quality criteria: objectivity, reliability, and validity.

But to clarify what validity means, here’s a simple example:

Imagine you step on the scale every morning to check your body weight. However, the scale is off by 2 kg every time. It provides consistent readings, so it’s reliable. But because it doesn’t show your true weight, it lacks validity.

The question of validity mainly comes up in quantitative research. Within this context, the characteristics of validity have been refined over time. In general, we distinguish between internal and external validity.

#1 Internal validity (quantitative research)

Internal validity refers to the degree to which researchers can draw causal conclusions from the results of a particular empirical study.

The most cited work on this is the classic “Experimental and Quasi-Experimental Designs for Research” by Campbell and Stanley (1963). Since experiments are considered the gold standard for testing causal relationships, internal validity is especially important in this context.

In an experiment, we test the causal link between an independent variable and a dependent variable. Internal validity is achieved when valid causal relationships can be derived from the results.

But how do we determine that?

Essentially, by ruling things out. Are there any other factors that might have influenced the dependent variable—factors that could explain the change even without manipulating the independent variable? In other words, are there any confounding variables that blur the causal relationship?

Let’s take an example:

In an experiment, the temperature inside a car is adjusted. People sitting in the car are asked how comfortable they feel at each temperature. Temperature is the independent variable. The reported comfort level is the dependent variable.

If you want to measure the effect of temperature on comfort, you better make sure the seat massage function is turned off! Otherwise, people might feel more comfortable—not because of the heat, but because of the massage. That doesn’t mean temperature has no effect, but it does make it harder to separate the influence of heat from the massage.

So, internal validity is all about minimizing the influence of potential confounding variables (Hussy et al., 2010).

#2 External validity (quantitative research)

Once internal validity is ensured, we can think about external validity. This refers to how well the results can be transferred to other contexts—such as different people, places, or points in time.

More casually, external validity is about whether the findings can be generalized to other situations (Hussy et al., 2010).

The term isn’t only used in experimental research. It’s also relevant in survey research, where generalizability matters too.

So how do we establish external validity?

Again, we use a kind of elimination process to identify and reduce threats to external validity.

One such threat is sampling bias—whether the sample is skewed in some way. This might happen, for example, if participants didn’t volunteer freely, or if certain conditions make the sample unrepresentative of the broader population.

#3 Internal and external validity in qualitative research

In quantitative research, internal validity has top priority. Only once that’s established do we look at external validity.

In qualitative research, it’s the opposite—or more accurately, internal validity doesn’t apply. Qualitative research isn’t about testing causal relationships, so internal validity isn’t relevant.

External validity, however, can play a role. Generalizing to a broader population might be important, especially if the sample is randomly selected.

For example, if you’re conducting a content analysis and select a sample of 100 videos from a population of 1,000 YouTube videos, you might ask whether the findings generalize to the other 900. But this kind of generalization isn’t usually the goal of qualitative studies.

Where external validity does matter is in generalizing to other contexts and situations. That is a goal of qualitative research. Just think about Grounded Theory: the aim is to generalize the findings of an empirical study by developing useful theoretical concepts and identifying their relationships (Hussy et al., 2010).

Learn how to use ChatGPT for Academic Writing – if you’ve ever used AI for academic writing, you’re probably familiar with the dilemma: where’s the line between smart support and academic misconduct?

Plagiarism can have serious consequences – ranging from poor grades to having your entire paper invalidated. But does that mean you shouldn’t use ChatGPT at all? Not at all! In fact, there are many smart and ethical ways to integrate AI into your academic writing process without falling into the plagiarism trap. And that’s exactly what this post is about!

Today, I’ll walk you through seven ways to use ChatGPT effectively and ethically in your academic writing.

1. Identifying Research Gaps and Formulating Hypotheses

You’ve done your research, but it feels like everything has already been said about your topic? ChatGPT can help you uncover unanswered research questions or methodological weaknesses in existing studies. This technique is used by many researchers to develop innovative research ideas – a real advantage if you’re writing your Bachelor’s or Master’s thesis.

ChatGPT is well-known for its ability to generate creative ideas. According to a study by Lee and Chung (2024), ChatGPT outperforms traditional tech-based brainstorming methods – like Googling – by a wide margin. This is because the AI can combine seemingly unrelated concepts in novel ways, sparking new insights.

Another trick: Use ChatGPT as a peer reviewer for your ideas. Ask it to evaluate your research question from the perspective of a critical academic. This can give you valuable insights on where to refine your approach.

Once you’ve got your research question, you can also use ChatGPT to support the hypothesis-building phase. Rather than asking it to generate a ready-made hypothesis, use it to explore relevant variables or test different hypothesis formats. For example, if your topic is the impact of remote work on productivity, you might ask: “What factors influence productivity when working from home?” or “Are there studies showing a positive correlation between remote work and increased productivity? What were the methodological, theoretical, or conceptual limitations of those studies?”

Try to think a little outside the box when writing your prompts – instead of just asking it to generate this or that.

2. Conducting a More Targeted Literature Search

Many students spend hours on Google Scholar or other academic databases without a clear system or search strategy. This typically leads to two major problems: First, relevant studies are often missed because the search terms aren’t well chosen. Second, the overwhelming number of search results can make it difficult to identify the most important and high-quality sources.

Instead of just typing keywords into a search bar, ask ChatGPT for support: Request relevant key terms or alternative expressions that are commonly used in academic literature. For instance, the term “digital learning” might also appear in studies as “e-learning” or “computer-assisted learning”. By asking ChatGPT for these synonyms, you’re less likely to miss key studies.

To narrow down the flood of results, ask ChatGPT which journals are the most reputable in your field. This will help you target your search and find higher-quality sources. Professors want to see papers that cite studies from the journals and conferences they themselves read and publish in. The quality of your references matters more than the topic match. Imagine you find a study that fits your research topic perfectly. If it’s from an obscure journal, it’s practically irrelevant to your academic field. It sounds harsh, but that’s the reality. Your reference list will be judged based on the quality of the sources it includes. So always make sure your argumentation is grounded in the most respected research.

However, having ChatGPT search for studies for you? That’s something you should seriously reconsider. In a study published in the prestigious PNAS journal, Lehr et al. (2024) tested ChatGPT’s performance across various research-related tasks. Literature searching was where the AI performed the worst. Interestingly, it did quite well when it came to advising on research ethics – which is kind of funny, if you think about it.

3. Improving Your Methodology

Choosing the right research method is one of the biggest challenges in academic writing. Instead of being unsure whether to go with a literature-based approach, or a qualitative or quantitative design, you can ask ChatGPT: “Which empirical method is best suited to answer my research question?” This can help you weigh the pros and cons of different methods more effectively.

Ideally, you choose the method that best fits your research question. But in reality, it’s often the other way around. Your supervisor tells you, “Please conduct interviews.” And now the method is fixed – your task becomes finding a research question that fits the method.

Operationalizing your concepts is also essential. ChatGPT can help you identify suitable items or scales that have already been used in previous studies. That way, you ensure your variables are clearly defined and measurable.

You can also use the AI to critically reflect on your methodology. Ask: “What methodological weaknesses might arise in my study?” This will help you identify potential limitations early on – and address them proactively.

If you’re planning an empirical study, ChatGPT can also assist in selecting appropriate statistical techniques or methods of analysis. This ensures that your data is analyzed systematically and with solid reasoning. However, you should not let ChatGPT analyze your data directly – the risk of errors is still too high. Instead, perform your statistical tests yourself and use ChatGPT to review your approach. According to Lehr et al. (2024), ChatGPT is quite good at spotting when you’re not following statistical best practices.

4. Researching Theories

Theories are the backbone of any academic paper. ChatGPT can give you an initial orientation by suggesting theories that might be relevant to your topic. Important: AI is not an academic source! Use ChatGPT as a starting point to explore theoretical frameworks, and once you’ve identified one that fits, switch to scholarly literature for further research.

A clever strategy is to ask ChatGPT to compare different theories, for example: “Compare Theory X with Theory Y – what are their strengths and weaknesses in relation to Topic X?” This kind of overview can make the decision-making process during your “theory casting” much easier.

You can also ask ChatGPT to name specific applications or research fields where a given theory is frequently used. This helps you assess how relevant a theory might be for your project — or, conversely, highlight that a theory has already been thoroughly examined in a specific area.

Another smart approach is to ask the AI about theoretical developments or critical debates: “How has Theory X evolved over the past ten years?” or “What are the main criticisms of Theory Y?” This gives you not just a general overview, but also insight into which aspects of a theory are currently debated or evolving.

Additionally, ChatGPT can help you draw interdisciplinary connections: “Are there overlaps between Theory X from psychology and Theory Y from sociology?” This can help you uncover new perspectives for your research and possibly create original theoretical links in your paper.

5. Strengthening Your Argument

A strong academic paper relies on clear argumentation and thoughtful discussion. And this is where ChatGPT can help in a somewhat unconventional way. Imagine you’ve constructed an argument that seems perfectly logical and convincing to you — but how would a critical reviewer assess it?

Most students overlook potential weaknesses in their arguments. But you can ask ChatGPT to deliberately deconstruct your reasoning. For example, ask: “What arguments could be used to contradict my thesis?” or “How might a critic challenge my reasoning?”

This turns ChatGPT into a “devil’s advocate,” giving you a fresh perspective on your work. It also allows you to proactively incorporate counterarguments — making your argumentation more robust and resistant to criticism.

6. Ensuring Argumentative Consistency

Once you’ve found strong arguments, it’s important to maintain consistency throughout your paper. Especially in longer texts, contradictions or unclear transitions can easily sneak in.

ChatGPT can help identify these inconsistencies if you ask it to systematically analyze your argument and point out any logical breaks. Try prompts like: “Identify inconsistencies or flawed reasoning in my text.”

Another common issue in academic writing is a lack of clear logical structure. Transitions between arguments might be missing, or there may be logical jumps that are hard for the reader to follow. ChatGPT can help by checking your paper for coherence and suggesting ways to improve the flow between points. A simple prompt would be: “Do you notice any logical leaps or missing links between my arguments?”

For smoother transitions, try: “How can I improve the argumentative connection between Section A and Section B?”

7. Improving Academic Writing Style

Another challenge many students face is mastering the academic writing style, and with ChatGPT, many simply ask the tool to rephrase their text to make it sound “better.”

But here’s what most don’t realize: ChatGPT can also help you improve your academic writing more deliberately. Instead of just asking it to rephrase your text, have it analyze and critique your writing style. For instance, you could ask: “What common writing mistakes do students in my field make?” or “Is my writing style appropriate for an academic paper?”

Keep in mind that ChatGPT has been trained on thousands of academic articles. So what it considers “academic” often reflects phrasing that frequently appears in published studies. This has led to a trend where professors joke about texts that overuse words like “crucial” or “key.” The widespread use of ChatGPT has caused an inflation of these terms. Don’t fall into this trap — and don’t let the AI drain the originality from your writing style.

One especially useful feature is that you can ask ChatGPT to adapt your style to a specific journal. If you’re planning to submit your paper to a respected journal, you can ask the AI to revise your text to match the style typically used in that publication. This helps you understand what academic writing looks like across different disciplines.

Final Thoughts – ChatGPT for Academic Writing

ChatGPT can make academic writing easier — but it’s no substitute for your own analysis and critical thinking. If used wisely, it can help you build better structures, sharpen your research question, compare theories, strengthen your arguments, and refine your writing style.

But don’t use ChatGPT to generate entire sections of your paper or accept its arguments uncritically. And above all: don’t use AI-generated citations! ChatGPT often invents sources — and that can lead to serious consequences. If you’re using an AI tool for literature searches, make sure it provides a link to the original source.

Never rely blindly on AI. Always verify academic sources yourself and treat ChatGPT as a research assistant — not a replacement for you as a researcher.

References

Lee, B.C., Chung, J.(. An empirical investigation of the impact of ChatGPT on creativity. Nat Hum Behav 8, 1906–1914 (2024). https://doi.org/10.1038/s41562-024-01953-1 

Lehr, S. A., Caliskan, A., Liyanage, S., & Banaji, M. R. (2024). ChatGPT as research scientist: probing GPT’s capabilities as a research librarian, research ethicist, data generator, and data predictor. Proceedings of the National Academy of Sciences, 121(35), 1-9. https://doi.org/10.1073/pnas.2404328121 

Grounded Theory sounds like a interesting approach, until you try to apply it. Suddenly, you’re drowning in concepts like open coding, categories, and constant comparison. Where do you even start?

If open coding has you scratching your head, don’t worry, you’re in the right place. This guide breaks it down in a way that’s easy to follow, so you can confidently use this method in your research.

What Makes Grounded Theory Different?

Open coding is a part of the data analysis process in the Grounded Theory approach. If you need a refresher on the basics of Grounded Theory, check out the introductory tutorial first.

Typically, when conducting a study, you might create an interview guide, conduct a dozen expert interviews, and then analyze them using content analysis.

In Grounded Theory, things work a bit differently. Here, data collection and data analysis can alternate. You first conduct a few interviews, transcribe them, and then step back to conduct a preliminary analysis. You apply techniques such as open coding (which we will discuss in detail shortly) and develop your initial categories.

Based on these findings, you return to the field and conduct the next round of interviews. This time, your questions build on what you discovered in your first analysis. This allows you to probe more precisely, dive deeper, and identify insights that help refine your emerging mini-theory.

This process, known as theoretical sampling, is the key characteristic that differentiates Grounded Theory from other qualitative research approaches.

Now, let’s get back to open coding.

Open Coding in Grounded Theory

According to Strauss and Corbin (1998), open coding is the first of three phases in the qualitative data analysis process for Grounded Theory:

1. Open Coding
2. Axial Coding
3. Selective Coding

Open coding is the first step in engaging with your data—whether it consists of interview transcripts, social media posts, notes, or other written reflections (referred to as memos in Grounded Theory terminology).

To make this process as easy as possible for you, here are five key goals to keep in mind when conducting open coding. After explaining all the steps, we will go through an example from an interview excerpt to illustrate how to apply open coding in practice.

Step 1: Identify Your Categories

Open coding is essentially the embodiment of inductive category formation, meaning you approach the data with a completely open mind, without considering existing theories.

Some scholars recommend incorporating existing theories at a later stage of your analysis—comparing the categories and relationships you have developed with what has already been established.

For open coding, however, the key is to go through your data without any preconceptions, marking similar content with the same category. In other words, everything that belongs together conceptually should be grouped into the same category.

By the way, the terms “code” and “category” mean the same thing. Coding is simply the process of categorizing.

But how exactly do you form a category?

Step 2: Abstraction Instead of Description

Rather than merely summarizing the content and using the interviewee’s wording, you should abstract the content. For example, consider this sentence from an interview transcript:

“I usually use TikTok to get the latest news.”

Instead of assigning the code “news” to this sentence, a more abstract code like “information acquisition” would be more appropriate.

According to Strauss & Corbin, categories can also have different dimensions. For instance, the category “information acquisition” could include the dimension “frequency”—how often does this information acquisition occur?

Another dimension could be “content”, referring to the type of information being acquired. In this example, it’s news, but in another case, it could be the latest fashion trends.

Dimensions of a category can also be treated as subcategories—a concept you may already be familiar with from qualitative content analysis. If you use software like MAXQDA, creating subcategories can be especially helpful.

Step 3: Constant Comparison

This goal revolves around the following questions:

• Are codes being assigned consistently?
• Are the same criteria being applied to categorize dimensions?

You can achieve consistency by continuously cross-checking previously coded text excerpts to ensure that similar meanings are being coded the same way, or adjusting if differentiation is needed.

At this stage, memos play an important role again. Here, you can jot down all your ideas as you code. What is the broader context? What could be a general explanation for what you are reading? Memos serve as a playground where you draft your initial theory.

Step 4: Achieving Saturation in Open Coding

You’ve probably heard that you should stop conducting interviews when no new information emerges. This principle is particularly associated with Grounded Theory literature.

In this context, the term theoretical saturation means that no new categories, variations, or relationships between categories emerge from your data.

However, there are two challenges with saturation:

1. The exact point of saturation is difficult to determine objectively.
   Researchers disagree on when and how saturation is reached, and numerous articles (e.g., Saunders et al., 2018) discuss this topic.
2. In a thesis, you don’t have time for 30 interviews.
   If you are working on a thesis, your time is limited. No one expects you to conduct interviews indefinitely until you reach this elusive saturation point. In this case, it is perfectly acceptable—after consulting with your advisor—to set a target number of interviews and stick to it.

For a fully comprehensive Grounded Theory study, the standard 12–15 interviews in a master’s thesis are usually insufficient. Therefore, it is essential to clearly define in your methodology section whether Grounded Theory is being used as a holistic approach or merely as a methodological toolkit without aiming for a fully developed Grounded Theory study.

Alternatively, you could collaborate with someone else and write a joint thesis. This way, you could impress your evaluators with 20–30 interviews and a more comprehensive Grounded Theory study.

For a PhD dissertation, the expectations are different. Here, you must address theoretical saturation and theoretical sampling, ensuring that the Grounded Theory approach is fully implemented from start to finish.

Example of Open Coding (Grounded Theory)

Let’s take this example from a fictional interview transcript:

“Wow, when I first put on the VR headset, it felt like I was in another world. I was totally surprised by how quickly you can connect with others in this Metaverse. I was in a virtual seminar room and later spoke with the professor. She was from Canada and super friendly and open to my ideas.”

The first step is to identify the key W-questions (Who? Where? What? How? Why?).

• Where? Metaverse
• Who? Professors and students
• What? Virtual seminars
• How? Through a VR headset in a virtual seminar room

The most interesting statement here is about quick social connection—which the interviewee has already highlighted by expressing surprise. You could code this part as “social connection”.

“Wow, when I first put on the VR headset, it felt like I was in another world. I was totally surprised by how quickly you can connect with others in this Metaverse. I was in a virtual seminar room and later spoke with the professor. She was from Canada and super friendly and open to my ideas.”

In later interviews, you would then explore how and why this interaction happens so easily. Could it be because users are represented by avatars, which lowers the barrier to approaching others?

Maybe. Maybe not.

Only Grounded Theory can reveal the answer.