Prompts for artificial intelligence to analyze academic articles

I have been experimenting with these prompts with some success. I use GPT-4 and upload a PDF of an academic article.

Usually it returns the information in the correct format, but sometimes it adds bold formatting to the Dataview inline fields or converts them from kebab case to sentence case. I don’t know how to make it not do this.

To optimize the information you get, it’s better to use Prompt 1 and then Prompt 2, due to context window. Doing so will, for example, return more detail in the fields outlined in Prompt 2 than if these were combined into one big prompt.

Let me know if you have suggestions for improvements or questions.

Prompt 1 Context and Methods

Use the following template to critically analyze the attached research article. For the "Methodology" and "Research Context and Problem" sections, use only two sentences per bullet point.

# Topics

Topics should be prefixed with hashtags and formatted using kebab case. Provide tags for the field of research, the methodology used, and any other important related concepts.

# Research Context and Problem

- context:: Describe the context in which the research is taking place, e.g., unanswered questions, world events, or other related contextual matters.
- problem:: Describe the problem the research attempts to solve.

# Future Research Opportunities

What future research opportunities exist as a result of this research?

# Methodology

- research-design:: Describe the overall framework and type of research (e.g., experimental, correlational, qualitative).
- participants:: Details about the demographic and relevant characteristics of the study participants, including how they were selected and any inclusion or exclusion criteria.
- sampling-method:: Explanation of how participants were chosen, including specific sampling strategies (e.g., random, convenience, stratified sampling).
- data-collection:: Description of the tools and techniques used to gather data (e.g., surveys, interviews, observations, instruments).
- materials:: Details about any materials used during the study, such as questionnaires, tests, or technology.
- procedure:: Step-by-step description of what was done during the study, including any experimental manipulations, interventions, or sequences of events.
- variables:: Specification of independent, dependent, and control variables, as well as any covariates or confounding factors.
- data-analysis:: Description of the statistical or thematic analysis methods used to process and analyze the data.
- tools:: Details on any software or specific tools used for data collection, management, or analysis.

## Limitations

Identify the limitations of the research according to the following, but only for those types of limitations which are relevant to the article.

### Methodological Limitations

Are there issues with the study's design, procedures, or specific methodologies used?

### Theoretical Limitations

Are there issues related to the conceptual framework or theoretical assumptions of the study?

### Practical Limitations

Are there constraints that arise from logistical or procedural challenges during the study?

### Data Limitations

Are there issues related to the data used in the study?

### Ethical Limitations

Are there any concerns related to the ethical aspects of the research?

### External Validity

Are there any limitations concerning the extent to which the study's findings can be generalized? 

### Internal Validity

Are there any issues that affect the credibility of the study's findings?

Prompt 2: Fodder for Atomic Notes

Identify the top five major findings of this research. Use the following template to format your description.

# Results

## Title of Key Finding

- claim:: A topical statement about the finding phrased as a claim.
- evidence:: Evidence supporting the claim. 
- explanation:: Explanation of how the evidence supports the claim. 
- academic-contributions:: Describe any contributions to theory, or other to the field of study, or other contributions to academic knowledge.
- methodological-contributions: Describe how the methodology used in this research advances a certain kind of methodology, if at all.
- practical-contributions:: Describe how the research could contribute to shaping policy, practice, society, or the world in general.

Related

6 Likes

I’ve been experimenting with similar prompts myself and have encountered some of the same formatting quirks.

One thing that’s helped me minimize these issues is being very explicit in the prompt about maintaining the exact format without adding any extra styling, like bold or case changes.

For example, you could try adding something like, “Please ensure that all Dataview inline fields remain in kebab case without any bold formatting.”

One suggestion I have is to experiment with breaking down some of the more complex fields in Prompt 2 even further.

For instance, instead of just asking for “evidence” under each key finding, you could ask GPT-4 to provide a brief summary of the evidence first, followed by specific examples or quotes. This might help in extracting more granular insights, especially for more nuanced articles.

Thanks for your reply. As it happens, I’ve stopped using this. I wasn’t learning as much or as well as when I just read the articles manually. I thought this would be a supplement to my learning and I guess in some ways it has helped me to catch one or two things I may have missed, but it’s not as game-changing as reading the ol’ fashioned way.

But thanks again for your comment! I hope the prompt or your version of it is bringing you value.

(As a student) I’m currently researching the use of AI to improve student outcomes and have come to a similar initial conclusion - common usage of AI tooling degrades popular retention methods like physical note taking or reading.

I’ve explored a few varieties of this: supplementing notes, extracting key terms with citations and producing glossaries, and creating practice questions. It always comes back to, given 30-60 minutes of free time, an individual can produce something of higher quality while learning throughout the process.

Still I think there’s a lot more to explore. Like automated categorization methods such as Zettelkasten, improvements to semantic search, and better privacy through on-device LLMs.