What "evidence-based" actually has to mean now
Every nutrition app, supplement brand, and Instagram account in 2026 calls itself evidence-based. The term has lost most of its meaning. Here's what I actually evaluate when I use the phrase, and what I think you should ask of anyone using it about your food.
I’ve been a registered dietitian for fifteen years and the phrase “evidence-based nutrition” has done some heavy lifting in that time. In 2010 it meant something specific. In 2026 it means almost nothing. Every supplement, every protein-blend startup, every macro app calls itself evidence-based. Some of them are. Most of them are using the phrase the way “all-natural” was used in the 90s.
Since this app’s value proposition leans on the word — Macroline talks a lot about provenance and verified data — I’ve been thinking about what I actually mean when I use it for my own patients. If you’re a person trying to evaluate whether a tool, a coach, or a recommendation is evidence-based, here’s what I’d want you to look for.
The four levels
When I evaluate a recommendation against “evidence-based,” I’m asking which of four levels it sits at. The phrase by itself doesn’t tell you which level — and it gets misused most when people apply it at the wrong level.
Level 1: Mechanism. A claim is consistent with established physiology. Example: “Protein helps satiety.” True at the mechanism level — protein triggers GLP-1 and CCK release, which slow gastric emptying and signal satiety to the brain. This is the easiest level to clear, and the level most “evidence-based” claims sit at. It says nothing about effect size or whether the intervention works in real life.
Level 2: Acute studies. A short-term, controlled study showed an effect. Example: “Adding 30g protein to breakfast reduced lunch calorie intake by 15% in a controlled feeding trial.” Now we have a measurable effect, but only in the very narrow conditions of that study. People often jump from this to broad claims, which they shouldn’t.
Level 3: Long-term randomized trials. Multi-month interventions with controls, in free-living humans, measured against meaningful outcomes (weight, lean mass, cardiovascular markers, mortality, etc.). Example: SURMOUNT-1 for tirzepatide. STEP-1 for semaglutide. The PREDIMED trial for the Mediterranean diet. Few interventions have this level of evidence. The ones that do are usually pharmaceuticals or large-cohort dietary patterns, not specific foods or supplements.
Level 4: Replicated in multiple populations and contexts. Findings hold across age, sex, ethnic background, fitness level, geography. Example: “Daily steps correlate with all-cause mortality.” Replicated dozens of times globally. This is the gold standard, and almost nothing in food-specific recommendations clears it cleanly.
Most “evidence-based” content sits at Level 1 or 2 and gets discussed as if it’s at Level 3 or 4. That’s the slippage that’s emptied the term.
What I evaluate before using the phrase
When a tool, coach, or recommendation describes itself as evidence-based, I check the following before I use the phrase about it myself:
1. Does the claim cite specific studies, or just “research shows”?
“Research shows” is meaningless. “Two RCTs in metabolically healthy adults found…” is meaningful. The first formulation has been the dominant pattern across nutrition content for thirty years and is functionally a tell that the writer didn’t read the studies.
2. Does the claim match the level of evidence to the level of confidence?
A confident “X causes Y” claim from a Level 1 mechanism finding is overconfident. A hedged “consistent with what we know about Y” claim from a Level 3 finding is underconfident. Evidence-based isn’t a single switch; it’s a calibration of how strongly you state things.
3. Does the claim acknowledge contrary findings?
Most popular nutrition claims have contrary findings somewhere in the literature. Studies disagree. The honest version of evidence-based notes the disagreement. The marketing version of evidence-based pretends the literature is uniform. If a nutrition source has never told you “the evidence is mixed on X” about anything, they’re not actually working from the evidence.
4. Does the source change positions when new evidence emerges?
This is the one most people don’t think to check. Real evidence-based practice updates. The fat/cholesterol guidance shifted multiple times between 1990 and 2020 as the literature evolved. The protein recommendations shifted upward as anabolic resistance research matured. Sources that haven’t updated anything in five years aren’t evidence-based; they’re picking a defensible position and freezing.
5. Does the source have a financial incentive that biases the claim?
Not disqualifying — most clinicians have some financial relationship to the recommendations they make. But disclosed and considered, vs. hidden and unexamined. A protein company saying “high protein is essential” is making a true claim with a non-coincidental financial alignment. That doesn’t make the claim wrong; it makes the claim worth a second source from someone without the financial alignment.
How this applies to nutrition apps specifically
Apps that call themselves evidence-based have several specific patterns I look at:
Where do their food values come from? USDA’s FoodData Central is the gold standard for unbranded foods in the U.S. Manufacturer label data is the standard for branded products. OpenFoodFacts is community-sourced and variable quality. “We have over a million foods” tells you nothing about where the values came from.
This is exactly why Macroline’s tier badge system is interesting to me — it makes the source explicit on every food you log, which is a level of transparency I wish more apps had. “Authoritative” and “Estimated” are different epistemic objects, and the user should know which one they’re looking at.
Are computed totals from snapshot data or live data? When you logged that meal three months ago and the food entry has since been corrected, did your old log update or stay the same? Snapshot semantics (the log preserves the values as of the time you logged) are evidence-based; mutating semantics (the log updates retroactively when the database changes) are not. Macroline’s snapshot-on-log approach is correct here.
Does the app have a clear answer on what they don’t know? “We don’t track sodium accurately for restaurant meals because the data isn’t reliable” is more evidence-based than “all our nutrition values are accurate.” The honest version of an app acknowledges its data limits. The marketing version doesn’t.
Are the AI-generated values flagged differently from the verified ones? With AI-driven food entry getting common, the difference between “we measured this in a USDA lab” and “Claude estimated this from the menu description” is enormous. If the app conflates them, that’s not evidence-based; it’s marketing-clean.
What patients should ask their dietitian
If you’re working with a clinician and you want to know whether they’re evidence-based:
- “What’s the strongest study you’d cite for this recommendation?” — Forces a level 3 reference, not a “research shows” wave.
- “What would change your mind?” — Tests whether they have a falsifiable framework.
- “Are there populations where this advice doesn’t apply?” — Tests whether they understand effect-size variance across groups.
- “What do you think is overhyped right now?” — Anyone who’s truly evidence-based will have a list. The list is the marker.
Note these questions don’t require the clinician to be right. They require the clinician to have engaged with the evidence honestly. There’s a difference, and it’s the actual signal.
What I tell my own patients
When a patient asks if a particular practice or supplement is “evidence-based,” I usually answer with a level qualifier and a confidence range, not a yes/no.
- “Creatine for resistance-training adults: Level 3, well-replicated, modest but real effect on strength gains.”
- “Berberine for blood sugar in non-diabetic adults: Level 2, suggestive, limited replication, I don’t recommend it routinely.”
- “Apple cider vinegar for weight loss: Level 1 mechanism, no consistent clinical effect, widely overhyped.”
This is the version of “evidence-based” I think is actually useful. It’s slower and more cumbersome than yes/no. It’s also honest.
The word has been overused to the point of partial uselessness. The way back is specificity — which level the claim sits at, where the data came from, whether the source updates as new evidence emerges. If a tool, a coach, or a recommendation can’t articulate those things, “evidence-based” in their hands is marketing copy. When it can, it’s the thing the term was originally supposed to describe.
Macroline is not medical advice. The judgments above reflect my professional opinion as a registered dietitian and may not apply to your specific clinical situation.