Honey “probably” eases a child’s cough, and that word is a precise rating

Key takeaways

• Reviews that follow the GRADE rating system signal how sure they are with set wording: “reduces” for high certainty, “probably” for moderate, “may” for low and “very uncertain” for very low.
• A 2018 Cochrane review found that honey probably eases children’s coughs more than a placebo, but the true size of the benefit could be anywhere from slight to large.
• About one in three new drugs approved from 2001 to 2010 later drew a boxed warning, a safety notice or a withdrawal, a median of 4.2 years after approval.
• In five experiments with 5,780 people, stating the uncertainty around a number lowered trust in it only slightly, and mostly when the uncertainty was described in words rather than as a numeric range.
• When evidence is thin, favor choices that are low-risk, reversible and time-limited, and decide in advance what will make you stop or get help.

In April 2018, a team of Cochrane reviewers updated their answer to an old family question: does honey help a child’s cough? They pooled six randomized trials with 899 children, ages 1 to 16¹.

Their summary said honey “probably” eases cough symptoms more than a placebo, a look-alike with no active ingredient².

Many readers would take “probably” as a polite shrug. It isn’t. In reviews that follow a rating system called GRADE, that word marks a specific level of confidence: moderate³. Swap it for “may,” and the reviewers are telling you their confidence is low. Drop the qualifier and say “reduces,” and they are claiming high certainty³.

Health research is full of signals like these. Most of us read right past them, then feel whiplash when advice changes.

This article teaches you to read them. It covers the kinds of uncertainty in health evidence, how researchers express it in numbers and words, why honesty about it costs less trust than people fear, and how to decide well when the truthful answer is “we don’t know yet.”

Not knowing comes in several kinds

GRADE is used by Cochrane, an international network that pools research, and by many groups that write medical guidelines³⁴. It rates the evidence behind each finding and marks it down for specific problems⁴. Those problems double as a map of what can make any health claim uncertain. Our article on systematic reviews explains the full system.

Chance. Small studies bounce around. With few people, a lucky run can look like an effect or hide a real one. GRADE calls this imprecision⁴. Honey’s edge over placebo rests on just two trials with 402 children¹.

Study quality. Flaws in how a study was run can tilt its results. A common one is people knowing which treatment they got. GRADE calls this risk of bias⁴.

Fit to you. Studies test certain people, doses and time frames. If yours differ, the results may not carry over. GRADE calls this indirectness⁴. Most children in the honey trials got honey for a single night, which the reviewers named as a limit¹.

Disagreement. When studies point in different directions, confidence drops. GRADE calls this inconsistency⁴.

Time. Some effects, good or bad, take years to show up. Short trials can’t catch them.

That last kind applies to every new product, natural or not. In 2017, a team led by Nicholas Downing reported in JAMA on all 222 new drugs and biologic medicines the FDA approved from 2001 through 2010⁵. Over a median of nearly 12 years, 71 of them, or 32%, had a major safety event after approval⁵.

Those events were 3 market withdrawals, 61 new boxed warnings (the most prominent warning on a drug label) and 59 formal safety notices. The median gap between approval and the first event was 4.2 years⁵.

Supplements have time gaps of their own. The FDA doesn’t decide whether a supplement works before it goes on sale⁷. NIH’s complementary health center says St. John’s wort appears to help mild to moderate depression, but it’s uncertain whether the benefit holds past 12 weeks⁶.

The NIH Office of Dietary Supplements notes that many supplements haven’t been well tested for safety in pregnancy or breastfeeding⁷.

A confidence interval shows the results the data can’t rule out

A study’s main result is a single best guess, called a point estimate. The confidence interval around it is the range of results that fit the data reasonably well.

Most studies report a 95% confidence interval. Roughly, if you ran the same study many times, about 95% of the ranges built this way would contain the true effect.

Honey makes a good example. The trials rated how often children coughed on a scale from 0 to 6, where lower is better¹. Compared with a placebo, honey lowered that score by 1.62 points on average. The 95% confidence interval ran from 0.22 to 3.02 points¹.

Read that range out loud and the story changes. A drop of 0.22 points would be hard to notice. A drop of 3 points would be hard to miss. The data fit both.

So “honey probably helps” is fair. “Honey cuts coughing in half” would claim more than anyone knows.

Now compare honey with dextromethorphan, a common drugstore cough medicine. The difference was 0.07 points in honey’s favor, with a range from about 1 point better to about 1 point worse¹. The reviewers rated that evidence low certainty and said there may be little or no difference¹².

That doesn’t prove the two are equal. It means the studies were too small to tell them apart.

Three questions decode almost any interval:

• Does the range include “no effect”? For a difference, that’s zero. For a ratio, it’s 1. If no effect sits inside the range, the study can’t rule out that nothing happened.
• Is the whole range worth having? If even the best end is too small to matter to you, the result doesn’t matter much either.
• How wide is it? Wide ranges usually mean few people or few events. Expect them to narrow, and possibly shift, as more studies come in.

One safety note from the same review: honey isn’t advised for babies under 12 months old¹².

The words that carry the rating

In 2020, the GRADE Working Group published standard wording for summing up what reviews find³. Their argument: unless a summary tells readers how big an effect is and how confident the reviewers are, it can mislead³. They refined draft phrases through workshops and a 2018 survey of 110 people³. The honey summary shows this wording at work².

Here is our guide to those phrases, plus a few common ones that carry no formal rating at all.

If you read…	How sure the reviewers are	What it means for you
“X reduces pain,” with no qualifier	High certainty3	The true effect is very likely close to what was found4
“X probably reduces” or “likely reduces”	Moderate certainty3	Likely close to the truth, though it could turn out quite different4
“X may reduce” or “the evidence suggests”	Low certainty3	A real possibility, not something to count on4
“The evidence is very uncertain”	Very low certainty3	Nobody knows yet; treat confident claims as guesses4
“Little to no difference”	Read the word in front of it3	Studies found no meaningful effect; “may” in front means that could change
“Slightly”	Signals a small but important effect3	Real but modest; ask whether it’s big enough to matter to you
“Is linked to” or “is associated with”	No formal rating	Often an observational study; it shows a pattern, not proof of cause
“Shows” or “proves” in a headline or ad	No formal rating	Look for the review behind it and its certainty grade
“May support” on a supplement label	No formal rating	Legal label wording, not a research finding

Two cautions. First, the code works only where writers follow GRADE, as in Cochrane summaries and many guidelines. In news stories and marketing, “may” can mean anything.

Second, the way “no effect” is worded matters. In the GRADE survey, 35% to 40% of respondents found “does not reduce” unacceptable for a near-zero result, compared with 20% for “little to no difference”³. Commenters objected that “does not” wrongly treats a result near zero as proof that nothing happens³.

Admitting uncertainty costs less trust than people fear

Many experts worry that saying “we’re not sure” will make people tune out. Researchers have put that worry to the test.

In a 2020 paper in PNAS, Anne Marthe van der Bles and colleagues ran five experiments with 5,780 people⁸. These included a repeat study registered in advance and a field test with readers of the BBC News website. People read facts on topics such as immigration and global warming. Some versions gave a single number; others added the uncertainty, either as a numeric range or in words⁸.

Readers noticed the uncertainty. Yet their trust in the numbers, and in the people reporting them, dropped only slightly, and mostly when the uncertainty was put in words rather than numbers⁸. The practical lesson for anyone sharing a figure: a range such as “between 4 and 6” is the more trusted way to admit doubt.

A 2022 study in PNAS Nexus, led by John Kerr and sharing two authors with the first, tested health messages⁹. In April 2021, nearly 3,000 adults in the United Kingdom each read one of three messages about COVID-19 vaccines¹⁰. The two that matter here were a persuasive one, adapted from National Health Service web content, and a balanced one. The balanced message laid out benefits and risks, admitted uncertainties and described the quality of the evidence⁹¹⁰.

The balanced version was trusted just as much⁹. People who started with negative or neutral views of the topic tended to trust balanced messages more⁹. Plans to get vaccinated didn’t differ between groups⁹.

A second experiment, about a proposed nuclear power plant, found a twist. The persuasive message won more support, even though readers rated it less trustworthy⁹.

Persuasion can win agreement while it loses trust. Keep that in mind whenever anyone offers you certainty with no ranges and no caveats, whether it’s a drug ad, a supplement seller or a public health campaign.

How to decide when the answer is “we don’t know yet”

You often can’t wait for perfect evidence. Medicine faces the same problem every day, and one of its habits is worth borrowing.

Doctors often can’t tell whether a cough, sore throat or ear infection will clear up on its own. One answer is the delayed prescription: the patient gets a prescription but is told to wait at least two days and fill it only if symptoms don’t improve.

A 2023 Cochrane review led by Geoffrey Spurling pooled 12 trials with 3,968 children and adults¹¹. About 93% of people given an immediate prescription took antibiotics, compared with about 30% given a delayed one and about 13% given none¹¹. Satisfaction was similar for delayed and immediate prescriptions, and complications didn’t differ meaningfully¹¹.

That is uncertainty handled well: a cautious default, a backstop and a deadline. You can use the same structure for your own choices.

Your situation	A reasonable approach
Mild problem; the remedy is low-risk, cheap and easy to stop	Try it with a start date, an end date and one symptom to track
Mild problem; the remedy has known risks or drug interactions	Ask a pharmacist or clinician first, or wait for better evidence
A problem that is serious, getting worse or not yet diagnosed	Get checked before experimenting; doubt about a remedy is no reason to delay care
Sudden or severe symptoms, such as chest pain, trouble breathing or signs of a stroke	Call 911
Experts or tests disagree about a major decision	Consider a second opinion before you choose

Before any choice made under uncertainty, ask four questions:

1. What happens if I’m wrong? A mistake you can undo is cheap. One you can’t undo is not.
2. What does waiting cost? Sometimes waiting is safe. Sometimes it lets a problem grow.
3. Would more information change my mind? If so, get it first.
4. What will make me stop or seek help? Write it down before you start.

Remember, too, that a review’s result is an average across many people. Some do better and some do worse. Our companion article on asking questions before trying a remedy shows how to test a choice fairly on yourself.

Honest uncertainty fits natural-health values

Many people come to natural health after being told something with total confidence that later proved wrong. That experience can push in two directions. One is distrust of every expert. The other is a sharper ear for how sure anyone has a right to be.

The second serves you better, and it fits values many natural-health readers already hold: careful observation, patience with the body’s pace and humility about what anyone can control.

It also cuts both ways. A healer who never says “I don’t know” deserves the same scrutiny as a drug ad that shows only benefits. The person who tells you how sure they are, and why, is treating you as a partner in the decision.

What the evidence shows

This table uses our library’s four-level scale, which is broader than GRADE. Many consistent trials can rate “strong” here even when GRADE reviewers marked the evidence down a step for design flaws.

Claim	Evidence rating	What it rests on
Honey eases children’s cough more than a placebo over a night or two	Promising but early	Two trials with 402 children; moderate certainty; wide range of possible effect1
Honey works about as well as dextromethorphan for children’s cough	Not enough evidence	Low certainty; the range includes real differences either way1
Delayed prescriptions cut antibiotic use without more complications	Strong evidence	12 randomized trials with 3,968 people; use fell from about 93% to about 30%; GRADE certainty moderate11
Serious safety problems can surface years after a drug is approved	Strong evidence	All 222 new drugs and biologics approved from 2001 to 2010, tracked for a median of nearly 12 years5
Stating uncertainty as a range costs little public trust	Promising but early	Five experiments, mostly using news-style facts rather than health choices8
Balanced health messages are trusted as much as persuasive ones	Promising but early	Two experiments in the United Kingdom9

Questions readers ask

The bottom line

“Probably,” “may” and “very uncertain” aren’t weasel words when a careful reviewer uses them. They’re ratings, and the confidence interval shows the range behind each rating. Reading both tells you how much weight a finding can bear.

When the honest answer is “we don’t know yet,” you can still decide well. Choose options you can reverse, set a time limit and a stop rule, and save your boldest moves for evidence that has earned them.

Keep exploring: questions before taking a supplement and benefit-risk assessment.

Related reading

• How systematic reviews work, and why cranberries failed one and passed the next
• A 22% risk cut can mean helping 2 or 3 people in 1,000
• Every health choice trades benefit for risk, including doing nothing
• When a second opinion changes something, and how to get one
• Feeling better after trying something isn’t a test, but you can run one

Sources

1. Oduwole O, Udoh EE, Oyo-Ita A, Meremikwu MM. “Honey for Acute Cough in Children.” Cochrane Database of Systematic Reviews, 2018 (CD007094.pub5). cochranelibrary.com
2. Cochrane. “Honey for Acute Cough in Children” (plain-language summary). Cochrane, April 10, 2018. cochrane.org
3. Santesso N, Glenton C, Dahm P, et al. “GRADE Guidelines 26: Informative Statements to Communicate the Findings of Systematic Reviews of Interventions.” Journal of Clinical Epidemiology, 2020;119:126-135 (accepted manuscript). pubmed.ncbi.nlm.nih.gov
4. Sambunjak D, Santesso N, Cumpston M, Brennan S. “Assessing the Certainty of Evidence.” Cochrane Training, 2021. resources.cochrane.org
5. Downing NS, Shah ND, Aminawung JA, et al. “Postmarket Safety Events Among Novel Therapeutics Approved by the US Food and Drug Administration Between 2001 and 2010.” JAMA, 2017;317(18):1854-1863 (summary by PCOM Library). library.pcom.edu
6. National Center for Complementary and Integrative Health. “St. John’s Wort.” NCCIH, updated May 2025. nccih.nih.gov
7. NIH Office of Dietary Supplements. “Dietary Supplements: What You Need to Know.” ODS, updated January 4, 2023. ods.od.nih.gov
8. van der Bles AM, van der Linden S, Freeman ALJ, Spiegelhalter DJ. “The Effects of Communicating Uncertainty on Public Trust in Facts and Numbers.” Proceedings of the National Academy of Sciences, 2020;117(14):7672-7683. research.rug.nl
9. Kerr JR, Schneider CR, Freeman ALJ, Marteau T, van der Linden S. “Transparent Communication of Evidence Does Not Undermine Public Trust in Evidence.” PNAS Nexus, 2022;1:pgac280. repository.cam.ac.uk
10. The Communication Initiative. Summary of “Transparent Communication of Evidence Does Not Undermine Public Trust in Evidence.” Comminit, 2022.
11. Spurling GKP, Dooley L, Clark J, Askew DA. “Immediate Versus Delayed Versus No Antibiotics for Respiratory Infections.” Cochrane Database of Systematic Reviews, 2023 (CD004417.pub6). pubmed.ncbi.nlm.nih.gov