Imagine a notice announcing that complaints have doubled. That sounds alarming. Now imagine learning that over two equally long reporting periods, they increased from one to two while the number of customers grew from twenty to two hundred. The original statement remains true. The story you are likely to tell yourself about it changes considerably.
This is an invented example, but the reading problem is real: accuracy at the sentence level does not guarantee a fair picture. A number can be correct and incomplete. A quotation can be genuine and unrepresentative. A before-and-after photograph can be authentic while leaving out the other things that changed between the two pictures.
You do not need to become a full-time investigator every time you read the news. A few habits can help you recognize when a striking claim needs more context before you repeat it.
Ask, “Compared with what?”
A price is high compared with another price. A result is unusual relative to some expectation. An increase needs a starting point and a period of time. When these are missing, you are being asked to evaluate a change without seeing its reference.
Return to the complaint example. The total rose, but complaints per customer fell from five percent to one percent, assuming the same counting method and comparable groups. Neither number alone tells you everything. Perhaps one of the new complaints was extremely serious. Perhaps the method for recording complaints changed. The correct response is not to replace one triumphant number with another. It is to ask which measure fits the question you are trying to answer.
The Centers for Disease Control and Prevention's introductory epidemiology materials distinguish counts, proportions, ratios, and rates. The distinction is useful well beyond public health: twenty incidents and twenty incidents per thousand opportunities describe different things. Before reacting to a count, look for the population or exposure behind it.
Find out who was included
“The average customer saved an hour” raises several questions. Which customers? Everyone who tried the service, or only those who continued using it? Was the hour measured, estimated, or reported from memory? Were people doing the same task before and after?
These questions do not prove the claim is wrong. They define what the claim can support. A survey of satisfied users can tell you something about satisfied users. It is a weaker basis for predicting what everyone will experience. A demonstration with prepared materials can show a product's capability without showing the amount of preparation required for normal use.
Look for who is absent. People who stopped responding, abandoned the tool, or could not access the service may matter to the conclusion. Sometimes the missing group is unavoidable. A responsible explanation should at least make the limitation visible.
Follow the link backward
When a claim matters, try to find the original material: the report, dataset, study, announcement, or full interview. A useful first check is whether the headline's strong wording appears in the source. “May help under these conditions” and “solves the problem” are different promises, even when they refer to the same work.
Check the date and the relevant passage. An old report can remain valuable, but it should not be mistaken for an update. A source may discuss a narrower place, group, or activity than the article describing it. If every link leads to another summary, you have learned something about how far you are from the evidence.
You do not have to read every appendix. Find the definitions, the method, and the limitations before treating the main number as portable. A statistic travels badly when its conditions have been left at the departure gate.
Separate what happened from why
Suppose a town introduces a new program and a useful outcome improves. The timing is worth noticing. It does not, by itself, establish how much of the improvement the program caused. Other conditions may have changed too. Perhaps the earlier year was unusual. Perhaps the people participating differed from those who did not.
In everyday reading, a helpful distinction is between “this happened after that” and “this happened because of that.” Evidence can sometimes support the stronger conclusion, but the argument has to do more than place two events beside each other. When the source itself is cautious about causation, a confident summary deserves a second look.
This works in the other direction as well. Asking about alternative explanations is not a license to dismiss every inconvenient result. Apply the same curiosity to claims you enjoy. Your favorite conclusion should not receive the easiest exam.
Decide how much checking the decision deserves
You cannot verify the entire internet before lunch. Match the effort to the consequences. An amusing, low-stakes observation may need little attention. A claim you intend to publish, use to make an important decision, or send to thousands of people deserves more.
If you cannot establish the context, you can say so. “The report says this for the group it studied; I don't know whether it applies here” is a useful statement. It preserves what you know without manufacturing the rest. Choosing not to share an irresistible headline is also an available editorial decision, even for someone whose entire editorial staff consists of one thumb.
The aim is to make your confidence fit the evidence, rather than to become permanently suspicious. A good reader can appreciate an interesting fact while leaving enough space around it for the facts that have not yet arrived.
Further reading
CDC, Principles of Epidemiology, Lesson 3: Measures of Risk — definitions and worked examples of ratios, proportions, and rates. https://archive.cdc.gov/www_cdc_gov/csels/dsepd/ss1978/lesson3/section1.html

