Last updated: August 2026

I save far more research papers than I actually read.

A paper appears in PubMed, Google Scholar, ResearchRabbit, or the reference list of another study. The title looks relevant, so I open it.

Then comes the more important question:

Is this paper actually worth spending 30–60 minutes reading carefully?

Researchers rarely have enough time to read every potentially relevant paper from beginning to end.

So before committing to a full read, I screen the paper first.

My goal is not to determine the entire scientific quality of a study in five minutes.

I am simply trying to decide:

Read now, save for later, or skip.

To make that decision, I usually work through seven questions.

My 7-Step Paper Screening Workflow

When I first open a research paper, I usually check it in this order:

Identify the research question
Scan the abstract
Look at the main figures
Check the experimental model
Look for the critical controls
Ask whether the conclusion matches the data
Decide whether the paper deserves a full read

One thing is deliberately missing from the beginning of that list:

I usually do not start by reading the Introduction from beginning to end.

When I am screening a paper, I want to understand its architecture first.

What is the question?

What experiments were done?

What is the central claim?

Only after that do I decide how much time the paper deserves.

Step 1: Identify the Actual Research Question

The title tells me the topic.

The research question tells me whether I care.

Those are not the same thing.

A paper with a title involving:

mitochondria + macrophages + inflammation

may sound relevant if I work on innate immunity.

But the actual question could be:

Does mitochondrial morphology change during inflammation?
Is mitochondrial metabolism required for cytokine production?
Does a mitochondrial protein regulate macrophage differentiation?
Can a mitochondrial pathway be targeted therapeutically?

Those are very different papers.

So I try to reduce the paper to one sentence:

“The authors are asking whether X affects Y under condition Z.”

If I cannot identify the question after looking at the title, abstract, and beginning of the Introduction, I stop and figure that out first.

A paper becomes much easier to evaluate once I understand what the authors are actually trying to demonstrate.

Step 2: Scan the Abstract — But Treat It as a Map

The abstract gives me the authors’ version of the paper.

I usually look for four things:

Background

What problem are they studying?

Knowledge gap

What was not known before this study?

Main finding

What do the authors claim to have discovered?

Significance

Why do they think the finding matters?

The most useful sentence is often some version of:

“Here, we show that…”

That usually contains the central claim.

But I do not treat the abstract as evidence.

The abstract tells me:

what the authors think the experiments show.

The figures help me decide whether I agree.

Step 3: Look at the Main Figures Before Reading the Whole Paper

This is probably the biggest change I made as I became more comfortable reading research papers.

I used to read papers in this order:

Introduction → Results → Discussion

Now, when I am screening a paper, I often move from:

Abstract → Figures → Results

The figures reveal the logic of the paper much faster.

I look at:

Figure titles
Experimental groups
Axes
Sample sizes
Controls
Statistical comparisons
Whether each figure logically leads to the next

I am trying to reconstruct the argument.

A mechanistic paper might roughly look like this:

Figure 1: Is there a phenotype?

↓

Figure 2: What causes it?

↓

Figure 3: Is the proposed factor necessary?

↓

Figure 4: Is it sufficient or mechanistically connected?

↓

Figure 5: Does the mechanism matter in vivo?

Not every paper follows this structure.

But looking at the figures usually tells me whether there is a coherent experimental story worth investigating further.

Step 4: Check the Experimental Model

A paper can ask an interesting question and still use a model that does not answer the question I care about.

So I check:

Cell line or primary cells?
Mouse, fly, organoid, or human samples?
In vitro or in vivo?
Genetic manipulation or pharmacological treatment?
Overexpression or endogenous regulation?
Acute stimulation or chronic disease model?

The model determines what conclusions are reasonable.

For example, a mechanism demonstrated in:

HEK293 cells with strong protein overexpression

may still be biologically interesting.

But if I am trying to understand innate immune signaling in primary macrophages during infection, I may not prioritize that paper.

This does not mean that one model is inherently better than another.

The important question is:

Can this experimental model actually answer the question the paper is asking?

Step 5: Look for the Critical Controls

Once I understand the model, I ask:

What alternative explanation would make this result misleading?

Then I look for the experiment that rules it out.

Suppose a paper claims:

Protein X is required for inflammatory signaling.

I might ask:

Was Protein X actually depleted?
Were appropriate control cells used?
Are the cells still viable?
Does the phenotype appear using more than one perturbation?
Is there a rescue experiment?
Could the treatment itself explain the phenotype?
Is the effect specific to the pathway being studied?

No paper has every imaginable control.

That is not what I expect.

I am looking for the controls that matter for the central claim.

If an obvious alternative explanation remains completely untested, I become more cautious.

Step 6: Ask Whether the Conclusion Matches the Data

This is the most important step in my screening process.

A result and an interpretation are not the same thing.

Imagine an experiment shows:

Knocking down Gene X decreases IL-6 production.

The data may support:

Gene X contributes to IL-6 production under these experimental conditions.

But the Discussion might eventually describe Gene X as:

a master regulator of inflammation.

Those statements are not equivalent.

Whenever I see a strong conclusion, I ask:

What did the experiment directly demonstrate?

I find it useful to distinguish several levels of evidence.

Association

X changes when Y changes.

Necessity

Removing X prevents or reduces Y.

Sufficiency

Introducing or activating X produces Y.

Mechanism

The experiments explain how X leads to Y.

Physiological relevance

The mechanism matters in an organism, disease model, or biologically relevant context.

A paper does not need to demonstrate all five.

But I want to know which level the data actually reach.

This prevents me from treating every mechanistic model as equally established.

Step 7: Decide — Read Now, Save for Later, or Skip

After the first screening, I make one of three decisions.

Read Now

I read the full paper when:

The research question is directly relevant
The experimental logic is interesting
The main figures contain evidence I need to understand
I may cite the paper
The mechanism could affect my own research
I need to evaluate the study in detail

At this point, I stop screening and start reading.

Save for Later

I save the paper when:

The topic is relevant but not immediately important
It may become useful background
One technique or figure is interesting
I expect to revisit the topic later

There is nothing wrong with this category.

Not every useful paper needs to be read today.

Skip

I stop reading when:

The research question is not actually relevant
The title looked more relevant than the experiments
The experimental model does not address the question I care about
The paper repeats information I already understand
I only needed one specific fact
Another paper answers my question more directly

Skipping a paper does not mean I think the research is bad.

It simply means:

This paper is not worth my time for my current question.

That distinction saves a surprising amount of time.

A Real Example: Screening a Nature Paper Before Reading It

To make this process more concrete, I applied it to a paper I recently encountered:

Drosophila immune cells transport oxygen through PPO2 protein phase transition

The study was published in Nature in 2024.

The title immediately makes an unusual biological claim:

Immune cells transport oxygen.

That is exactly the kind of paper where I want to see how well the experiments support the headline before committing to a detailed read.

1. What Is the Research Question?

My first attempt at reducing the paper to one question would be:

Do Drosophila blood cells have a previously unrecognized role in oxygen transport and systemic oxygen homeostasis?

More specifically, the paper focuses on crystal cells and the protein PPO2.

The question interests me because it goes beyond the classical idea of an immune cell performing an immune function.

It asks whether a blood-cell population also contributes to whole-organism physiology.

That is enough to keep me reading.

Decision: continue screening.

2. What Does the Abstract Claim?

The paper proposes that crystal cells participate in oxygen handling through changes in the physical state of PPO2.

The overall model roughly connects:

Oxygen availability

↓

PPO2-containing crystal cells

↓

Changes in PPO2 state

↓

Oxygen acquisition or release

↓

Systemic oxygen homeostasis

This is a large claim.

So the next question is obvious:

Do the figures actually connect these steps?

3. What Do I Want to See in the Figures?

Before reading every paragraph, I look for several kinds of evidence.

Is PPO2 linked to oxygen status?

I want evidence that changing PPO2 or crystal cells affects physiological markers related to oxygen availability.

Does PPO2 respond to oxygen conditions?

If the proposed phase transition matters, I want evidence connecting oxygen conditions to PPO2 state.

Does disrupting PPO2 change tissue oxygenation?

This would move the story from molecular behavior toward physiology.

Is there rescue evidence?

A rescue experiment can make a causal model substantially stronger.

Does the mechanism matter for the whole animal?

If the paper claims systemic physiological relevance, I want some organism-level phenotype.

If the study only showed that PPO2 looks different under different oxygen conditions, I would be much less interested.

What makes the paper worth reading is the attempt to connect:

protein behavior → cell function → whole-animal physiology

4. Is Drosophila an Appropriate Model?

For this particular question, Drosophila is useful.

The study concerns oxygen distribution in an organism traditionally understood to rely heavily on its tracheal system.

The model also allows researchers to genetically manipulate:

Crystal cells
PPO2
Oxygen-response pathways
Specific tissues and cell populations

That gives the study strong experimental control.

There is an obvious limitation:

A mechanism demonstrated in Drosophila does not automatically mean mammalian immune cells use the same system.

But that does not invalidate the question.

The study only needs a model capable of testing its own biological hypothesis.

A paper does not need to explain every organism. It needs to use a model appropriate for the claim it is making.

5. What Controls Matter Most?

For a claim this unusual, correlation would not be enough.

I would want to know:

What happens when crystal cells are removed?
What happens when PPO2 is disrupted?
Do physiological oxygen-response pathways change?
Can the phenotype be rescued?
Could the phenotype be explained by PPO2’s known immune functions instead?
Does changing oxygen availability affect PPO2 in the predicted direction?

These questions tell me which experiments deserve careful attention when I move from screening to a full read.

6. Does the Final Model Go Beyond the Data?

The proposed model connects crystal cells and PPO2 with systemic oxygen homeostasis.

Rather than asking:

“Do I believe the paper?”

I break that model into smaller claims.

Claim 1

Crystal cells influence oxygen homeostasis.

Claim 2

PPO2 is required for that function.

Claim 3

PPO2 changes physical state in response to relevant conditions.

Claim 4

That change contributes to oxygen acquisition and release.

Claim 5

The mechanism has organism-level consequences.

Now I can ask:

Is there direct experimental support for each claim?

Scientific papers are rarely all-or-nothing.

I might find the first three claims extremely convincing while thinking one part of the final mechanistic model still requires more evidence.

That is a much more useful way to read a paper than simply labeling it “good” or “bad.”

7. Would I Read the Full Paper?

Yes.

Not because it was published in Nature.

The paper passes my screening because it contains several things I value:

A clear biological question
A surprising conceptual finding
Genetic manipulation
Physiological experiments
Mechanistic experiments
Causal testing
A connection between cell biology and organism-level physiology

At this point, the paper has earned a deeper read.

I would now move through the Results figure by figure and decide exactly which conclusions I think are supported.

Journal Prestige Is Not One of My Screening Steps

I obviously notice the journal.

But I try not to use journal prestige as a shortcut for deciding whether a paper deserves my time.

A high-impact paper can be irrelevant to my research question.

A paper in a specialized journal may contain the exact experiment, mechanism, or technique I need.

The question that matters more is:

Does this paper contain information that could change how I understand my research problem?

That is what earns reading time.

I Check the Date — But “Newest” Does Not Always Mean “Most Useful”

Recency matters in rapidly changing areas such as:

Single-cell technologies
CRISPR methods
AI for biology
New therapeutic platforms

But older papers may still be essential when they are:

Foundational studies
Original mechanistic discoveries
Landmark methodological papers
Classic conceptual work

So instead of automatically choosing the newest paper, I ask:

Am I trying to understand where an idea came from, or what the current evidence says?

Those are different searches.

A Short Screening Checklist

If I need to screen a paper quickly, I reduce the entire process to five questions:

What is the actual research question?
Which experiment provides the strongest evidence for the main claim?
Is the experimental model appropriate?
Do the controls rule out the most obvious alternative explanations?
What will I know after reading this paper that I do not know now?

The fifth question is often the deciding one.

If I cannot explain what I expect to gain from a full read, I probably do not need to read the paper yet.

After a Paper Passes the Screening

Once I decide that a paper deserves a full read, I change how I approach it.

Now the goal is no longer:

“Is this worth reading?”

It becomes:

“Exactly what do these experiments demonstrate?”

At that point, I go through the Results and figures carefully, inspect the relevant methods, and take notes on the experiments that matter.

A simple format I like is:

Question → Experiment → Result → Interpretation → My concern

For example:

Question: Is Gene X required for macrophage activation?

Experiment: Knock out Gene X in primary macrophages and stimulate the cells.

Result: Cytokine production decreases.

Interpretation: Gene X contributes to the inflammatory response.

My concern: Does deleting Gene X affect cell viability?

This keeps the data separate from both the authors’ interpretation and my own.

Where AI Fits Into This Workflow

AI can make paper screening faster.

I may use an AI tool to:

Explain unfamiliar terminology
Clarify a complicated method
Help identify the paper’s central question
Locate a concept within a long paper
Compare the basic claims of several papers

NotebookLM can be especially useful once I have collected several related studies.

But there is one question I do not want to outsource completely:

“Is this evidence convincing?”

That requires looking at the experiments.

My rule is:

Use AI to navigate the paper. Use the data to judge the paper.

Three Common Mistakes

1. Reading Every Paper From Beginning to End

This sounds rigorous.

Usually it is just inefficient.

Different papers deserve different levels of attention.

A paper that is central to my project may require hours.

Another may only be worth reading for one figure.

2. Treating the Abstract as Evidence

The abstract is the authors’ compressed version of the story.

It is useful for orientation.

But when the claim matters, I want to see the experiment behind it.

3. Confusing Statistical Significance With Biological Importance

A statistically significant difference is not automatically a biologically important one.

I still ask:

How large is the effect?
Is it reproducible?
Does it matter in a relevant biological context?
Does it support the conclusion being made?

The P value is part of the evidence, not the entire interpretation.

Key Takeaways

I do not read every paper I save.
I first identify the research question and central claim.
I scan the main figures before committing to a full read.
I check whether the experimental model fits the question.
I look for the controls that matter for the central conclusion.
I separate association, necessity, sufficiency, mechanism, and physiological relevance.
I make a simple decision: Read now, save for later, or skip.
I do not use journal prestige as a substitute for evaluating relevance.
AI can help me navigate papers, but I still judge the experimental evidence myself.

And before committing to a detailed read, I keep coming back to one question:

What will I know after reading this paper that I do not know now?

Frequently Asked Questions

Should I read every research paper from beginning to end?

No.

The amount of attention a paper deserves depends on why you opened it.

Some papers deserve a detailed figure-by-figure read.

Others may only be useful for one experiment, method, or background point.

Should I read the abstract or the figures first?

I usually scan the abstract first to understand the authors’ claim and then move quickly to the figures.

The abstract tells me the proposed story.

The figures show me the evidence behind it.

How quickly can I decide whether a paper is worth reading?

For an initial screen, I am not trying to understand everything.

I only need enough information to identify the research question, inspect the main experimental logic, check the model and critical controls, and decide whether a deeper read is justified.

Some papers can be ruled in or out very quickly.

Others require more inspection before the decision becomes clear.

Can AI tell me whether a research paper is worth reading?

AI can help identify the research question, summarize terminology, and navigate a paper.

But I would still inspect the main figures, experimental model, and controls myself before deciding whether the evidence is relevant or convincing.

Final Thoughts

The hardest part of literature reading is not understanding every paper.

It is deciding which papers deserve your limited attention.

I used to treat saving a paper almost like creating an obligation to read it.

Now I treat literature reading as a filtering process.

First:

Is this relevant?

Then:

What is the actual question?