Map an Article's Claims Against the Evidence It Gives

Break a persuasive article into its claims, match each to the evidence actually offered and find the logical gaps before you share or cite it.

At a glance

Best for: Students, journalists, analysts and anyone about to cite, share or act on an opinion piece, report or LinkedIn post that sounds convincing.
Tested on: Claude · Opus 5.5
You fill in: article
You get: Main claim Giving every student an AI tutor raises attainment, so every school should do it by next September. Claim map Claim Evidence in the text… (full result below)

research

writing

education

decision-making

Prompt

Map the argument in this article so I can see what's actually supported. Work only from the text below. Don't bring in outside facts unless you label them "outside knowledge".

Article: {{article}}

Output:

1. Main claim: the single thing the author most wants me to believe, in one sentence in your own words.

1. Claim map: a table with columns # | Claim (paraphrased) | Evidence given in the text (quote of 15 words or fewer, or "none") | Evidence type (data, study, anecdote, expert opinion, analogy, none) | Support (strong / partial / weak / none).

1. Logical gaps: where a conclusion goes further than its evidence (e.g. correlation read as cause, one example generalised). Name the gap and the claim number.

1. What's missing: the one piece of evidence that would most strengthen or sink the main claim.

1. One-line verdict: how much of the main claim survives if you keep only the strongly supported rows.

Stay neutral: judge the support, not whether you agree with the conclusion.

Fill in

{{article}}: The full text of the article or passage to analysee.g. Schools that let students use AI tutors are seeing results that should end the debate...

Example input

article: Schools that let students use AI tutors are seeing results that should end the debate. At Harbour Academy, maths grades rose 14% in the year after it introduced an AI tutor for homework. A 2025 survey found 71% of teachers say AI saves them time. Students love it too: "I finally get fractions," one Year 8 pupil told me. Critics say AI encourages cheating, but cheating has always existed, and calculators didn't destroy arithmetic. The evidence is clear: AI tutors raise attainment, and schools that hold back are failing their pupils. Every school should give every student an AI tutor by next September.

1.5k0

Generated Outputs (1)

3 hours ago

Claude

Opus 5.5

3 hours ago

Generated Output

1. Main claim
Giving every student an AI tutor raises attainment, so every school should do it by next September.

2. Claim map

| # | Claim | Evidence in the text | Type | Support |
|---|---|---|---|---|
| 1 | AI tutors improved maths results at one school | "maths grades rose 14% in the year after" | data (one school) | partial |
| 2 | AI saves teachers time | "71% of teachers say AI saves them time" | survey | partial |
| 3 | Students like AI tutors | "I finally get fractions" | anecdote | weak |
| 4 | Cheating worries are overblown | "calculators didn't destroy arithmetic" | analogy | weak |
| 5 | AI tutors raise attainment in general | "The evidence is clear" | none | none |
| 6 | Schools that hold back are failing pupils | none | none | none |
| 7 | Every school should adopt by next September | none | none | none |

3. Logical gaps

• Claim 1 to 5: one example generalised. A single school's results are used to support a claim about all schools.

• Claim 1: cause assumed. Grades rose "in the year after", but there's no comparison group and no mention of other changes, such as new staff, a different exam or a different cohort.

• Claim 1: ambiguous number. "Rose 14%" could mean 14 percentage points or a 14% relative increase, and we aren't told which grades.

• Claim 2 to 5: evidence on a different question. Saving teachers time is not the same as raising pupil attainment.

• Claim 4: weak analogy. A calculator does one narrow task; an AI tutor can produce whole answers, so the cheating risk isn't comparable.

• Claim 7: the timeline has no support. Nothing addresses cost, training or safeguarding for a national rollout within a year.

4. What's missing
A comparison of similar schools or classes, with and without AI tutors, over the same period and measured on the same assessments. That would support or sink claim 1, and the whole piece rests on claim 1.

5. Verdict
No row is strongly supported, so what survives is "one school saw better maths grades after introducing an AI tutor, for reasons not yet shown". That is far short of "every school should".

Why it works

"Work only from the text" stops the model filling holes in the argument with its own knowledge, which is exactly what hides weak reasoning. The evidence type column makes the pattern visible at a glance: one anecdote, one survey on a different question, one analogy. The short quote limit anchors each row to what the author actually wrote. Asking for named logical gaps tied to claim numbers turns a vague "this seems weak" into specific points you can raise. The final line tells you how much is left standing.

When not to use it

It judges support within the text, not whether claims are true; a well-evidenced article can still rest on a bad study. Pair it with a fact-check for that. It's less useful for long academic papers, where methods sections matter more than rhetoric; read those with a study-appraisal checklist instead.

Comments (0)

No comments yet. Be the first to share your thoughts!