How to Write a Thesis Statement That Survives Revision: The 3-Question Test

Most thesis statements that come back marked "needs more work" fail one of three diagnostic questions. Not big philosophical questions. The three questions a tired supervisor with a stack of fifteen drafts is silently asking as they read your opening paragraph.

Run your thesis through these before you submit. If it fails any of them, you already know what the margin note is going to say.

The 3 questions

1. Could a smart classmate, reading the same texts, reasonably disagree with this? If no, you've written a fact or a summary, not a thesis.

2. What does the main verb commit to? Explores, examines, and looks at commit to nothing. Strong theses use verbs that name a claim — causes, undermines, depends on, fails to account for, requires.

3. What does the second half of the sentence add that the first half didn't? If the part after the comma just restates the topic in different words, your thesis is hiding a tautology behind a transition.

That's the test. The rest of this post is what each question actually catches, with real drafts running through them.

Question 1: Arguability — "Could a smart classmate disagree?"

This is the UNC Writing Center's "So what?" test reframed as a question you can answer in five seconds. UNC's version is excellent but slightly abstract — students read it, nod, then keep writing the same un-arguable thesis. "Could a smart classmate disagree?" gives you a concrete person to imagine pushing back on the sentence.

Here is the kind of claim it catches:

Social media affects how teenagers communicate.

Could a smart classmate disagree? No. Anyone who has read any paper on the subject agrees that social media affects teenage communication. The only interesting question is how. Stating that it does is a summary of the field, not a thesis.

Try this version instead:

Among US teenagers aged 13–17, the shift from public Instagram feeds to private "close friends" stories has reduced cross-group communication more than it has changed any single platform's user count.

A smart classmate could disagree with that. They could argue the shift hasn't been as large as the headlines suggest. They could argue group fragmentation predated the close-friends feature and is driven by something else. The thesis is now contestable, which is the only state in which it can be defended.

The arguability test is also the fastest filter for AI-assisted drafts. We've reviewed dozens of student introductions written with ChatGPT's help over the past eighteen months, and the failure mode is consistent. The model produces a fluent paragraph that ends on an unarguable sentence about how the topic is "complex" or "multifaceted." Both words signal failure on this test. If your thesis says the topic is complex, the next move is to say what is complex about it, specifically enough that someone could be wrong.

Question 2: What does the verb actually commit to?

Most weak theses have a soft verb in the middle holding the whole sentence up. The fix is usually a single-word swap.

Soft verbs that commit to nothing:

• explores, examines, looks at, considers, addresses, investigates, discusses, analyzes, focuses on

Verbs that commit to a claim:

• causes, prevents, undermines, depends on, requires, predicts, contradicts, replaces, fails to account for, distorts, sustains

Watch the same sentence flip from soft to strong with one substitution.

• Soft: "This paper examines the relationship between caffeine intake and exam performance."

• Strong: "Caffeine intake above 400 mg in the 12 hours before an exam predicts worse short-term recall, despite improving subjective alertness."

The soft version is a description of the paper's contents. The strong version is a claim about the world that the paper has to defend. Supervisors read for the second kind. When they write "where's the argument?" in the margin, what they mean is: I cannot find a verb that points at one.

A rule we've found useful in our own editing: if the thesis verb can be replaced with "is about," the thesis isn't a thesis yet.

Question 3: What does the second half add?

This is the question that catches the most cleverly disguised weak thesis. The structure looks fine. The sentence is long enough to feel like a thesis. But when you read it carefully, the part after the comma is the same idea as the part before it, rephrased.

Example:

Climate change is a major problem facing the world today, and it is something that needs to be addressed urgently by governments and individuals alike.

Two clauses joined by and. Read them separately:

• Clause 1: Climate change is a major problem.

• Clause 2: It needs to be addressed urgently.

Clause 2 is a restatement, not an addition. "Is a major problem" and "needs to be addressed urgently" mean the same thing in context. You could delete either clause and lose nothing.

The fix is to make the second clause carry a non-obvious claim that depends on the first:

Because every additional year of delay roughly doubles the cost of equivalent emissions cuts, the most defensible climate policy is one that prioritizes near-term reductions over politically easier long-horizon targets.

Now the second clause does work. It rules out the alternative (long-horizon targets) and explains why the first clause's framing leads to the policy claim. The thesis picks a fight with a real opposing view.

UCLA Writing Programs frames the same idea as "What, How, So What?" — the what is your topic, the how is your method or evidence, the so what is the part that justifies the sentence's existence. Question 3 is the so what in a form you can apply in ten seconds without pulling out a worksheet.

A worked revision: three drafts

Here is a real-shape example of how a thesis gets pulled through the three questions. Topic: a paper for an undergraduate literature course on Frankenstein.

Draft 1.

Mary Shelley's Frankenstein explores the dangers of unchecked scientific ambition.

Question 1 (arguable?): No. Every reader of Frankenstein agrees the novel is about that. Fail.

Draft 2 tries to add specificity.

Draft 2.

Mary Shelley's Frankenstein explores the dangers of unchecked scientific ambition through the character of Victor Frankenstein, who creates a monster and faces tragic consequences.

Question 1: Still no. The added detail is plot, not argument. Question 2 (verb commits to a claim?): "Explores" is still soft. Fail.

Draft 3.

Frankenstein's monster commits his first murder only after Victor refuses to acknowledge him, which positions Shelley's novel less as a warning against creating life than as a warning against abandoning what we create.

Question 1: Could a smart classmate disagree? Yes — they could argue the monster's violence is innate, or that Shelley's primary target is hubris in creation itself rather than the abandonment that follows. Pass.

Question 2: "Positions" is a real claim-verb here, and "less as X than as Y" carries a contestable judgment. Pass.

Question 3: Does the second clause add? Yes — it rules out the conventional reading (warning against creation) and substitutes a different one (warning against abandonment). The two halves work together. Pass.

Draft 3 is defensible. It picks a fight, picks a specific fight, and uses the second half of the sentence to sharpen the first. A supervisor reading this knows what the paper is going to argue.

What supervisors actually mean

When a draft comes back with comments on the thesis, the language tends to repeat. Here is the rough translation:

• "Too broad" → Failed Question 1. The claim is general enough that no one could disagree.

• "Where's the argument?" → Failed Question 2. The verb doesn't point at a claim.

• "Restates the prompt" → Failed Question 3. The two halves of the sentence are the same idea twice.

• "Unclear stance" → Usually Question 1 again, sometimes Question 2. The thesis describes a topic without choosing a position on it.

• "Needs to be more specific" → Often Question 2. Vague verbs invite vague claims.

The mistake students make is treating these as five different problems. They are mostly three.

Where AI tools help, and where they cause this exact problem

AI writing assistants are good at producing grammatically clean thesis statements that read like a thesis statement. They are not good at making the underlying claim arguable. The default failure mode of a one-shot prompt like "write me a thesis on X" is a sentence that reads smooth, has a verb, and would not survive Question 1.

The way we've seen students use these tools well is to invert the workflow. Instead of asking the model to write the thesis, ask it to attack the thesis you've already drafted. Paste your sentence and ask: "What is the strongest counterargument to this claim?" If the model can't generate a serious counterargument, the thesis is failing Question 1, and you should know that before your supervisor does.

This works because counterargument generation is a task language models are reasonably calibrated on, while original argument generation against your specific evidence usually isn't. In posts we've worked through with students, this loop has turned a soft draft into a defensible one in three or four iterations — faster than waiting two weeks for the next supervisor meeting.

One more pass, after you've written the paper

After you've drafted the body of the paper, run the three questions one more time. Almost every paper drifts during writing. The evidence you end up using rarely matches the evidence you planned to use, and the claim that's actually supported is rarely the claim you set out to support. If the body of your paper now argues something slightly different from your thesis, rewrite the thesis to match the paper, not the other way around. The thesis is supposed to summarize the argument you actually made, not the one you intended to make on day one.

If the thesis after the rewrite still passes all three questions, you're done. If it doesn't, the gap between your claim and your evidence is the next thing to look at, and that's a different problem from how the sentence is worded.