Check Three Things Before You Read a Single Number
I review charts for a living, and my first 10 seconds on any new chart are mechanical. I look at the axis origin, the unit labels, and the date range. Those three checks catch the majority of deceptive charts before I even read the title. Everything else — color, chart type, annotations — is secondary, because a beautiful chart with a truncated axis still lies.
The axis origin is the first thing I check because it is the cheapest trick to pull. Most charting tools now default to auto-scaling the Y axis, which means Excel or Google Sheets will happily start your bar chart at 95 instead of 0. On a line chart that is sometimes defensible; on a bar chart it is almost always misleading, because bars encode length and we read length as magnitude. A bar that starts at 95 and ends at 103 is drawn as if it were eight times taller than a bar from 0 to 1. That visual ratio is what your brain records, not the label.
The second check is unit and magnitude. Revenue in thousands or millions? Percentages of a total or percentage-point changes? A chart labeled "+18%" could mean 18 percent growth or a jump from 3.2% to 3.8% conversion. Those are different stories with the same headline. The third check is the time window. A 12-month window starting in a trough will look like a recovery no matter what happened before it.
Screenshot the chart and crop it to the plot area only, cutting off the axis labels and title. Then ask a colleague what story it tells. If they describe something different from what the axis labels say, the chart is doing too much work with too little honesty. I use this trick in reviews because it isolates the visual from the caption, which is where the spin usually lives.
Trick 1: The Truncated Y Axis
This is the most common misleading technique in business charts, partly because the tool does it for you. Take a Monthly Revenue table with four rows: January 98, February 99, March 101, April 103 (figures in thousands of dollars). Plot it as a bar chart with the Y axis starting at 0 and the four bars look nearly identical — a flat quarter, essentially no story. Plot the exact same numbers with the Y axis starting at 95 and January looks like a fraction of April. Same data. Two completely different impressions.
Open a blank sheet and enter your data as two columns: Month in A1:A4 (January through April) and Revenue in B1:B4 (98, 99, 101, 103). Select B1:B4, insert a clustered column chart, then right-click the vertical axis, choose Format Axis, and set Minimum to 0. This is your baseline. Screenshot it, because you will compare against it.
Duplicate the chart. On the copy, set the vertical axis Minimum to 95 and Maximum to 105. The bars will now start near the bottom of the plot area and stretch to wildly different heights. April will look roughly five times the height of January. Print both charts and put them side by side on your desk; that side-by-side is the fastest way to train your eye.
Divide 103 by 98 to get 1.051. That is a 5.1% increase over three months. Now measure the drawn heights in the misleading chart: January spans 3 units above the 95 baseline, April spans 8. The drawn ratio is 2.67 to 1, more than 50 times the actual ratio. Any time you catch yourself thinking "that growth looks steep," this is the arithmetic to run.
For a line chart tracking a metric that never approaches zero — say, a server uptime percentage that swings between 99.2 and 99.9 — starting at 99 is reasonable and starting at 0 hides the entire story in a flat line. The rule I use: bars must start at zero, lines may be truncated if the chart is clearly labeled and the metric cannot plausibly be zero. Write that rule on a sticky note.
A common mistake is assuming the truncation is malicious. Most of the time it is the default setting in the tool and nobody touched it. That matters for how you raise it, which I cover later.
Trick 2: The Double Y Axis That Manufactures Correlation
A dual-axis chart plots two series with two different scales on the same plot. In the wrong hands, it is a correlation factory. You put revenue on the left axis and website sessions on the right axis, then adjust the maximum of one axis until the two lines cross in a satisfying way. Slide the right axis maximum from 50,000 to 80,000 and an unrelated pair of series suddenly moves in lockstep.
Real example. Suppose Monthly Revenue is 98, 99, 101, 103 and Support Tickets is 210, 208, 205, 201 (in thousands of dollars and ticket counts respectively). Plot both with auto-scaled axes and support tickets appear to crash while revenue climbs — a nice "our product is improving" narrative. Now override the right axis to a minimum of 190 and the ticket line looks almost flat. Neither chart is wrong arithmetically. Both are wrong rhetorically.
Copy the two-column data, insert two separate line charts stacked vertically with the same X axis labels. Read the two shapes independently. If the relationship survives side-by-side reading, it is probably real. If it only appears when the lines share a canvas, the axes did the work, not the data.
Put Revenue in B1:B4 and Tickets in C1:C4, then in D1 enter =CORREL(B1:B4,C1:C4). Revenue rises, tickets fall, so you get a strong negative coefficient, roughly -0.99 on four points. Now change any single value, say the 201 to 190, and watch how violently the coefficient moves. Four data points is not enough to establish anything.
Ask what would physically cause the pattern. That is the difference between correlation and causation. Revenue up and tickets down could mean faster onboarding, or it could mean support hours were cut in March and people gave up asking. The chart cannot distinguish those, and a dual-axis chart actively discourages you from asking.
When someone shows you a dual-axis chart, ask one question: "What happens if we set both axes to auto-scale?" If the relationship vanishes, you have found the answer. In my experience the person who built it often does not know the answer, and that is the useful signal — it tells you whether the analysis is understood by whoever is presenting it.
Trick 3 and 4: 3D Pies and Cherry-Picked Windows
A 3D pie chart distorts proportions because the front slices get more visual area than the back slices, and the tilt makes the near edge of each wedge seem thicker. A slice that is genuinely 22% of the total can look like 30% if it sits at the front. The same applies to exploded 3D pies, where the detached slice is tilted toward the viewer. I recommend refusing 3D pies entirely; a bar chart sorted by value answers the comparison question better and cannot lie about area.
Selective time windows are subtler and therefore more effective. If Monthly Revenue fell from 140 two years ago down to 88 in mid-2023 and has since recovered to 103, a chart starting in July 2023 shows only 88, 94, 98, 99, 101, 103 — a clean 18-month growth story with a 17% gain. The chart is accurate. The conclusion "we are growing" is technically true and practically misleading, because the business is still below where it was three years earlier.
If a chart shows a trend, ask for the same metric going back to at least the previous peak or the previous trough. Add the earlier rows to the Monthly Revenue table, chart the full series, and look for whether the recent rise is a recovery or a new high. If the presenter cannot produce the longer series, treat the short one as unverified.
Monthly comparisons are noisy because of seasonality. If April 2026 is 103 and March 2026 was 101, that is a 2% gain month over month, but April 2026 against April 2025 might be a 9% decline. Always check year over year for anything with a seasonal pattern: retail, education, travel, energy.
A bar chart of one series with no comparison group is a common offender. "Our conversion rate is 4.1%" tells you nothing without the industry benchmark or last quarter's number. Before you accept a single-bar chart, ask what the baseline is. No baseline, no conclusion.
Trick 5: Cumulative Totals That Hide a Slowdown
A cumulative chart — running total over time — always slopes upward as long as the monthly number is positive. That makes it an excellent tool for hiding a decelerating business. Use the Monthly Revenue series 88, 94, 98, 99, 101, 103 (thousands of dollars). The cumulative line goes 88, 182, 280, 379, 480, 583. It rises every single month with no dip. But the monthly increments are 88, 94, 98, 99, 101, 103 — the growth rate is falling from 6.8% to 2.0%.
If you are handed a running total column, rebuild the monthly values by subtracting each row from the one above. In Excel, with cumulative totals in B2:B7, enter =B3-B2 in C3 and fill down. Then chart column C. This one transformation exposes more hidden slowdowns than any other technique.
In D3 enter =(C3-C2)/C2 and format as a percentage. On the series above you get 6.8%, 4.3%, 1.0%, 2.0%, 2.0%. That column, not the cumulative line, is what tells you whether momentum is building or dying. I put this column next to every cumulative chart I build.
Boards love cumulative charts for annual targets because they always look encouraging. Check the monthly pace against the pace needed to hit the annual target. If the company needs 120 per month to reach the goal and is delivering 103, the cumulative line is reassuring and the reality is not.
The Eight-Point Checklist for Any Chart You Are Handed
Run this list in order on every chart that finds its way into a decision. It takes about two minutes once you have done it a few times, and it catches almost everything.
Is it zero? If not, and the chart uses bars, mentally rescale before drawing any conclusion. For line charts, note the truncation and ask whether the metric can plausibly be zero. If the answer is no, the truncation is fine but should be labeled.
Thousands, millions, percent, percentage points, per capita? A chart showing "revenue in $M" where the reader assumes dollars is off by a factor of a million. Check the axis title and the footnote. If there is no unit label, that alone is a red flag.
Where did the numbers come from — the CRM, a survey of 40 people, a vendor report? Source determines how much weight the chart deserves. A survey of 40 respondents plotted with two decimal places implies precision the data does not have.
Compared to what? Last year, the control group, the competitor, the industry average? A single number on a chart is a fact without a reference point. Ask for the comparison before you form an opinion.
When does it start and why? If the start date is not a natural boundary like a fiscal year or a product launch, ask who chose it. Extend the window backwards mentally and ask whether the story survives.
Two Y scales on one plot is the strongest correlation trap. Note the maximum of each axis and ask what happens if both auto-scale. If you cannot get an answer, treat the apparent relationship as unproven.
"Conversion rate 50%" from 4 visitors is noise, not news. Find the denominator. If the chart shows percentages, ask for the underlying counts. Small denominators produce wild percentages that mean nothing.
Buckets like "under 30," "30 to 45," and "over 45" can be drawn to hide or create an effect. Try moving one boundary and see whether the conclusion flips. If it does, the finding is about the buckets, not the data.
How to Question a Suspicious Chart Without Being the Problem
You will be in meetings where a chart is wrong and the room is nodding. The failure mode is not missing the problem; it is raising it in a way that makes people defensive and gets you labeled difficult. The fix is to ask about the chart's construction rather than accusing anyone of deception. Most people did not build the chart themselves and are happy to explain what they know.
These four phrasings work. Use them verbatim if you freeze up.
This works on truncated axes because it is a neutral request about presentation, not a challenge to the conclusion. If the answer is "the effect goes away," you have made the point without saying the word misleading. If the answer is "the effect holds," you have improved confidence in the finding, which is also a win.
Framing it as context-seeking rather than skepticism gets you the longer series almost every time. Presenters usually have the data and simply chose a window. Once the wider view is on screen, let the room draw its own conclusion.
Source questions are almost always welcome, because they let the presenter demonstrate rigor. And on the occasions when the answer is "a vendor whitepaper" or "a survey of 40 people," you have surfaced the limitation without attacking anyone.
My favorite question because it works on any chart, misleading or honest. It forces the room to name the evidence that would falsify the claim. If nobody can name anything, you are looking at a belief, not an analysis. Ask it late in the discussion, after the conclusion has been stated.
Raise chart problems in the meeting if the decision is being made in the meeting; raise them privately first if the presenter is a peer who will present again later. A peer who gets publicly corrected in front of their director will remember the messenger, not the message. I send a two-line message before the session: "Worth double-checking the Y axis starts at 95 — happy to help rebuild it." That has never once gone badly.
Five Mistakes Readers Make Even on Honest Charts
Some errors are yours, not the chart's. These are the five I see most in analyst and manager reviews.
The title says "customer satisfaction improved" and you accept it. The footnote says the survey response rate fell from 40% to 12%, which means the happiest customers are now the only ones answering. Always read the footnote before the title sticks in your memory.
Revenue up, tickets down, therefore onboarding improved. That is a hypothesis, not a finding. You need a mechanism or an experiment. If the chart is a before-and-after with no control group, you cannot separate your change from the season, the campaign, or the economy.
A 100% improvement from 1 to 2 units is not a trend. Percentages hide denominators, and small denominators produce dramatic percentages. If a chart is heavy on percentage change, ask for the raw counts before you use the number in a decision.
A forecast of "$1,247,382" looks more credible than "about $1.2M," and it is not. The extra digits are noise dressed as rigor. Round aggressively when you present, and keep the raw figures only in the appendix.
Smoothed lines and rolling averages lag. A 7-day average will keep rising for days after the underlying metric actually turned. If the chart hides the raw daily points, plot them yourself before you act on the direction.
Practice on Real Charts This Week
Reading charts is a trained eye, not a talent. The fastest way to build it is to rebuild the misleading version yourself, which is why the axis exercise earlier has you plot both charts. Once you have personally dragged an axis minimum to 95 and watched two nearly identical bars become a dramatic slope, you will never un-see it in someone else's deck.
Start with three charts this week: one from your own team's reporting, one from a news article, and one from a vendor's sales deck. Run the eight-point checklist on each and write down the answers. If chart selection is where the gap is, the Excel chart selection guide covers when each chart type actually earns its place, and the data storytelling guide covers how to present a finding honestly without stripping out the persuasive parts. If you are building a portfolio, the visualization portfolio guide shows how reviewers judge whether your charts can be trusted.
The last thing to internalize: an honest chart is not a weaker chart. Truncating an axis does not make your growth look better to a competent reader; it makes your chart look untrustworthy the moment someone checks. The analysts whose numbers get believed over years are the ones who set the axis to zero, label the units, show the full window, and point out the limitation before anyone asks.
