Interpret P-Values, Effect Sizes and Confidence Intervals in Plain English

Paste your statistical output and get a plain-English reading of each number, what it supports, what it does not prove, and the misreadings to avoid.

You getPlain-English summary, table and what this result supports

Prompt

Prompt
You are a statistician who explains results to non-specialists accurately and without jargon.

I will paste statistical output (software output, a results table, or a paragraph from a paper) right after this prompt. If I have not said what was studied, who was measured, or the sample size, ask me up to 3 short numbered questions in one message and wait for my answers.

Rules:
1. Explain every p-value, effect size and confidence interval in one or two plain sentences, using the actual numbers from my output.
2. Keep "statistically significant" separate from "practically important." State the effect size in real-world units where possible.
3. Describe p-values correctly. Never call one the probability that the result is true or due to chance.
4. Flag misreadings the output invites, such as significance treated as importance, "not significant" treated as "no effect," correlation treated as causation, or many tests run without correction.
5. Do not add numbers that are not in my output. If something you need is missing, say so.

Output format:
- Plain-English summary (3 sentences max)
- Table: Statistic | Value | What it means | What it does NOT mean
- What this result supports
- What it does not prove
- Watch out for (bulleted misreadings)
- Questions to ask the analyst or authors

End with one line naming the biggest assumption or missing information that limits this interpretation.

Example output

At a glanceKey statistics from the sleep app A/B test412 users, two-sample t-test on nightly sleep minutes
Mean difference
6.2 min
Feature group slept more on average
p-value
0.03
Not a 3% chance of a fluke
Cohen's d
0.21
A small effect; groups overlap heavily
95% CI
0.6 to 11.8 min
Low end is near zero

Example for: a product manager reading an A/B test of a sleep app's wind-down feature, 412 users, two-sample t-test on nightly sleep minutes.

Plain-English summary

Users with the wind-down feature slept on average 6.2 minutes more per night than users without it. A difference this large would be fairly unusual if the feature truly had no effect (p = 0.03). The true difference could plausibly be anywhere from about half a minute to almost 12 minutes, so the effect is real but small.

What each number means

StatisticValueWhat it meansWhat it does NOT mean
Mean difference6.2 minFeature group slept 6.2 min more on averageThat every user gained 6 minutes
p-value0.03If the feature had zero effect, a gap this big or bigger would show up about 3% of the timeA 3% chance the result is a fluke, or a 97% chance the feature works
Cohen's d0.21A small effect: the two groups overlap heavilyThat the effect is meaningful for users
95% CI0.6 to 11.8 minThe range of true effects consistent with this dataA 95% chance the true value sits in this exact range

What this result supports

  • The feature probably increases sleep a little for this kind of user.
  • The effect is unlikely to be zero or negative, given the interval excludes zero.

What it does not prove

  • That users feel more rested or sleep better quality, since only minutes were measured.
  • That the gain is worth the build cost; 0.6 minutes is consistent with the data.
  • That it works for groups not in the test.

Watch out for

  • Calling this a "significant improvement" in marketing. Significant here means detectable, not large.
  • Ignoring the wide interval. The low end is near zero.
  • If other metrics were also tested, one p = 0.03 among many may be chance.

Questions to ask the analyst

  1. How many outcomes were tested, and was there a correction for multiple comparisons?
  2. Were users randomly assigned?
  3. How was sleep measured: phone sensor or self-report?
  4. What gain did the team decide in advance would be worth shipping?

Biggest limitation: the output does not say whether assignment was random, so a causal claim rests on that assumption.

Written by AI from this prompt. Your results will vary.

Tips and common edits

Why this works

  • The 'What it does NOT mean' column forces the model to correct the most common misreading of each number, not just define it.
  • Separating significance from practical importance stops a tiny but detectable effect being reported as a big win.
  • Banning added numbers keeps the explanation tied to your actual output instead of invented context.

Use this when a report, paper or software printout hands you numbers you need to understand. Have the output ready to paste into ChatGPT, Claude or Gemini, plus a line on what was studied.

Common edits

  • Ask for a two-sentence version you can drop into a slide or email to your manager.
  • Add 'compare this with a second result I will paste' to interpret two studies side by side.

Make it repeatable

Keep using this prompt without pasting it in every time.

Prompt details

By
Prompts-GPT editors
Published
How prompts are written and checked

Did this prompt work?

Keep exploring

More in Research