Data & BI

Verified against ChatGPT · 2026-08-13

Find out if a trend in your numbers is real or just noise before you tell anyone about it

Analyzes a described time-series trend for whether it's a genuine signal worth acting on or normal noise/seasonality dressed up as a story, before it gets repeated in a meeting as fact.

ChatGPT (GPT-5.1)4 fillable variables

The prompt

Ready to copy — highlighted parts are example details you can swap.

Act as a skeptical senior data analyst asked to sanity-check a trend before it gets repeated as a finding in a meeting. Your default assumption should be that most short-term movement in business metrics is noise, not signal — make me convince you otherwise with the data given, don't confirm the trend just because it was described to you as one.

THE TREND AS DESCRIBED
"Our website traffic has been climbing for the last three weeks, seems like the new blog strategy is working."

UNDERLYING DATA POINTS
Weekly sessions: 12,400 / 12,900 / 13,600 over the last 3 weeks, versus a typical weekly range of 11,000-14,500 over the prior 6 months.

KNOWN SEASONAL OR CYCLICAL PATTERNS
Traffic historically climbs every year from late summer into September as back-to-school search interest picks up.

WHAT WOULD CHANGE BASED ON THIS TREND BEING REAL
Marketing wants to double the content budget for next quarter if the blog strategy is confirmed as the driver.

Work through, in order:
1. Restate the trend as a specific, falsifiable claim ("metric X moved from A to B over period Y"), not the vaguer narrative version it was described as.
2. Check it against the known seasonal or cyclical patterns given — if this movement is plausibly explained by a pattern that recurs every year/quarter/week regardless of anything else changing, say so and state how confident you are in that alternative explanation.
3. Assess whether the number of data points and the size of the move are sufficient to distinguish a real shift from normal volatility in this specific metric — a two-data-point trend is not a trend, and a metric that's naturally volatile needs a bigger move to mean something than a naturally stable one.
4. If there's a plausible confound (something else that changed around the same time that could explain the movement instead of, or in addition to, whatever's being credited), name it explicitly.
5. Give a verdict: likely real signal, likely noise/seasonality, or genuinely insufficient data to say — and be willing to land on the third option rather than forcing a confident answer either way.

WHAT NOT TO DO
Do not treat the fact that a trend was described to you as evidence that it's real — evaluate it as skeptically as if you found the raw numbers yourself with no accompanying narrative.

OUTPUT FORMAT
1. Restated falsifiable claim
2. Seasonal/cyclical check and confidence in that alternative explanation
3. Signal-vs-noise assessment given data volume and this metric's normal volatility
4. Plausible confounds
5. Verdict, with the confidence level explicitly stated, and what additional data (if any) would resolve remaining uncertainty

Customize

Optional — swap in your own details for the highlighted parts above.

Why this works

Explicitly instructing the model to default to skepticism and require the data to convince it, rather than accepting the trend as described, directly counters a well-documented conversational tendency: a model handed a confident narrative ("seems like the new blog strategy is working") will, by default, tend to agree with and elaborate on the framing it was given rather than independently interrogate it, since agreeing with a plausible-sounding premise is the lower-friction continuation; stating the skeptical prior upfront overrides that default and forces the analysis to start from the null hypothesis that this is noise. Requiring the trend to be restated as a specific falsifiable claim before anything else matters because vague narrative framing ("traffic has been climbing") smuggles in a direction and a cause simultaneously, while a restated claim (sessions moved from A to B over three weeks) can actually be checked against normal volatility for that metric, separating the measurement from the story being told about it. The explicit instruction to check data volume and volatility against this specific metric, rather than trends in general, matters because three weeks of an inherently noisy metric like weekly web traffic is a very different evidentiary bar than three weeks of a naturally stable metric like headcount — a model not pushed to make this comparison metric-specific will apply a generic "three data points isn't much" caveat without actually engaging with whether three weeks is enough for this particular metric's normal swing range. Naming the decision stakes explicitly (a budget doubling riding on this) raises the bar the analysis is implicitly being held to, and permitting "insufficient data to say" as a legitimate final verdict — rather than forcing a binary real/fake conclusion — is what allows the model to actually land there instead of defaulting to whichever answer sounds more decisive.

What you get back

Restated claim: weekly sessions rose from 12,400 to 13,600 (about 9.7%) over three consecutive weeks. Seasonal check: this window overlaps with the known late-summer-into-September traffic climb, which is a strong alternative explanation independent of any blog strategy change — moderate-to-high confidence this pattern alone could account for most of the movement. Signal-vs-noise: 13,600 sits within the prior 6-month normal range (11,000-14,500), so this move has not yet exceeded typical volatility for this metric. Verdict: insufficient data to attribute this to the blog strategy specifically — recommend comparing this year's late-summer climb rate against last year's same-period climb rate before committing to a budget decision based on this trend.

Verified against

ChatGPT GPT-5.1 · 2026-08-13

Changelog

  • 2026-08-13 Initial publish, verified against ChatGPT GPT-5.1.

Need this built into your business?

If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.

EXPLORE WHAT SCULT BUILDS
All Data & BI prompts

Check your AI visibility

One URL in, a 0–100 score and the exact fixes out.

RUN THE CHECK

Browse all the tools

15 tools across six categories
13 of them never send your data anywhere

Free · No signup · No trial clock

SEE THE DIRECTORY