CBD dosage: three numbers that cannot all be right

Three numbers govern this question. They come from three credible places, and they are mutually incompatible.

• About 2 mg a day — the provisional safe intake for cannabidiol as a food, published by EFSA in February 2026
• 10 to 25 mg — a typical serving in a product on sale
• 300 mg or more per day — the point at which a 2023 review from the University of Sydney found benefits became clearly evident

Read them in order. The products on the shelf deliver roughly ten times what the European food regulator provisionally endorses, and somewhere between a tenth and a thirtieth of the quantity at which the research starts finding anything.

Everything sold in this category lives in the gap between those two numbers. That is the actual answer to “how much CBD should I take”, and no chart will tell you, because a chart cannot show a gap it is sitting inside.

This page will not tell you what to take. It will give you the arithmetic, which is arithmetic and belongs to you, and the three sources, which are free to read. What to do with them is a conversation with a pharmacist or a doctor, and the last section is about how to make that conversation useful.

The arithmetic you actually need

This part nobody should be vague about, because it is not medical advice — it is multiplication, and being unable to do it is why people misjudge products by a factor of ten.

Percentage to milligrams per millilitre

A percentage on an oil is a weight-in-volume figure. The conversion is one step:

percentage × 10 = milligrams per millilitre

• A 5 % oil is 50 mg/ml
• A 10 % oil is 100 mg/ml
• A 20 % oil is 200 mg/ml

That is the whole rule, and it works because 1 % means one gram per hundred millilitres, which is 1,000 mg per 100 ml, which is 10 mg/ml.

Total milligrams to milligrams per millilitre

Anglophone products usually print a total instead of a percentage. Divide by the bottle volume:

total mg ÷ millilitres = mg/ml

A 10 ml bottle labelled 1,000 mg is 100 mg/ml — the same as a 10 % oil. A 30 ml bottle labelled 1,500 mg is 50 mg/ml, which is a 5 % oil despite the larger headline number.

This is the single most common way buyers are misled, and nobody is even lying. A 1,500 mg bottle genuinely contains 1,500 mg. It is simply weaker per drop than a 1,000 mg bottle a third of its size.

Drops

A standard pipette delivers roughly 0.05 ml per drop, which is about 20 drops per millilitre. So:

mg per drop = mg/ml ÷ 20

A 10 % oil at 100 mg/ml gives about 5 mg per drop. A 5 % oil gives about 2.5 mg.

That figure is an estimate rather than a measurement. Drop size varies with pipette bore, oil viscosity, temperature and how vertically you hold it, and a 20 % difference between drops is ordinary. If precision matters, a capsule is the format that provides it.

Price per milligram

The only honest comparison between two products:

price ÷ total mg = cost per milligram

Run it before comparing anything else. A large bottle with a big number on the front is frequently worse value than a small concentrated one, and the arithmetic takes five seconds.

Total CBD versus CBD

A certificate may report CBD and CBDA on separate lines. CBDA is the acid form the living plant makes, and heating converts it at a known ratio:

Total CBD = CBD + (CBDA × 0.877)

A product quoting the total and one quoting only the CBD line are not comparable. The same 0.877 applies on the THC side, which is why some jurisdictions compute Total THC = delta-9 + (THCA × 0.877).

Three products, worked through

The arithmetic above is easier to trust once you have watched it overturn an obvious-looking comparison. Here are three products of the shape you actually meet, priced plausibly.

The candidates

• A — 30 ml bottle, labelled 1,500 mg, €60
• B — 10 ml bottle, labelled 1,000 mg, €50
• C — jar of 60 gummies, labelled 1,500 mg, €45

At a glance, A and C look identical and both look stronger than B. Neither impression survives contact with a calculator.

Strength

A: 1,500 ÷ 30 = 50 mg/ml, which is a 5 % oil. B: 1,000 ÷ 10 = 100 mg/ml, which is a 10 % oil. C: 1,500 ÷ 60 = 25 mg per gummy.

B is twice the strength of A, despite the smaller number on the front. Per drop, at roughly 20 drops per millilitre, A gives about 2.5 mg and B about 5 mg. Someone switching from A to B without doing this sum doubles what they take while believing they have downgraded.

Value

A: 60 ÷ 1,500 = €0.040 per mg. B: 50 ÷ 1,000 = €0.050 per mg. C: 45 ÷ 1,500 = €0.030 per mg.

Now the ordering inverts. C is the cheapest per milligram and B the most expensive, which is the opposite of what the shelf price suggests — B is the lowest-priced item and the worst value.

What the sums do not settle

C wins on cost per milligram and loses on everything else that matters. A gummy is swallowed, so it sits at the bottom of the bioavailability table, and it usually arrives with sugar and sometimes with other actives.

A and B are oils, which can be held sublingually for two to three times the absorption. So the cheapest milligrams are the ones least likely to reach you, and cost per milligram is necessary rather than sufficient.

The habit worth forming

Do the three sums before reading a single review. Strength per millilitre, milligrams per serving, cost per milligram. It takes under a minute, it is the same arithmetic every time, and it is immune to marketing in a way that opinions are not.

The variables a chart cannot see

Suppose someone did build an honest dosage chart. Here is what it would have to know about you, and why no grid on a product blog can.

Body composition, not just weight

Cannabidiol is highly lipophilic — it distributes into fat and returns slowly. Two people at the same weight with different body composition have different distribution volumes and different timelines, and weight alone does not capture it.

This is also why the first few days are not representative: with repeated administration, levels build until the tissue reservoir reaches equilibrium.

Liver function

Metabolism happens in the liver. Anything that changes hepatic function — a condition, another medicine, age — changes the outcome of the same quantity, and it is the organ already flagged as a concern in EFSA’s assessment.

Genetics, specifically CYP2C19

CYP2C19 is one of the enzymes cannabidiol interacts with, and it is one of the best-documented examples of genetic variation in drug metabolism. Poor-metaboliser and rapid-metaboliser phenotypes both exist, in a minority of people in every population and in a substantially larger minority in some.

That means two people can take the identical product and end up with meaningfully different concentrations, for reasons neither of them has any way of knowing without testing.

Everything else you take

The interaction runs in both directions. Cannabidiol can raise concentrations of medicines cleared by CYP3A4 and CYP2C19, and medicines that induce or inhibit those enzymes can change what happens to the cannabidiol.

A chart that ignores concurrent medication is ignoring the single largest determinant of risk for the population most likely to consult it.

The product itself

Every calculation assumes the label is right. Where testing has found content ranging from none to far more than advertised, the input to the arithmetic is the weakest link in it.

Which is the actual argument

None of this means quantity is unknowable. It means it is personal, and that a number printed for a stranger is a number produced without any of the six inputs above. A pharmacist with your medication list in front of them has five of the six, which is why that conversation is worth more than any page.

Where the numbers on those charts come from

Search this topic and the results include a carousel of dosage charts. They share a format: body weight down one axis, “mild / moderate / severe” across the other, milligrams in the cells.

There is no source

Follow a chart back and it cites another chart, or nothing. There is no regulatory body, no clinical trial and no pharmacopoeia behind those grids. They are, as far as anyone can establish, invented — and then copied, which is how they acquired the appearance of consensus.

Why the body-weight axis is misleading

Weight-based dosing is real in medicine, and it is how the authorised cannabidiol medicine is calculated. But applying that logic to a supplement implies the underlying relationship has been characterised in the relevant population, and for consumer quantities it has not.

The axis borrows the look of pharmacology to lend authority to a number that has none.

Why “mild / moderate / severe” is doing something worse

That axis quietly implies a condition, and a condition implies a claim. A grid whose columns are severity levels is making a medical statement without writing a medical sentence, which is a large part of why the format is so popular.

What to do when you see one

Ask the question that resolves every claim on this site: what is the source, and at what quantity was it shown? If a chart cannot answer either, it is decoration, and it is decoration on a question where being wrong costs money.

What the Sydney review actually found

The most directly relevant paper is a 2023 review in Clinical and Translational Science from the Lambert Initiative for Cannabinoid Therapeutics at the University of Sydney. It exists to answer exactly the question this page is about.

The question it set out to answer

The authors note that there is sufficient clinical trial evidence for high oral quantities — in the region of 10 to 50 mg per kilogram — in intractable childhood epilepsies. Their question was whether low-dose products, the ones widely sold as nutraceuticals, confer any benefit at all.

How they defined the scope

Interventional studies in adults, oral cannabidiol at or below 400 mg per day, any health condition, measuring efficacy or safety and tolerability. Products with THC content above 2.0 % were excluded, so the results are about cannabidiol rather than about cannabis.

That exclusion matters, because it removes the confounding that weakens most of this literature.

The threshold

Their finding, in plain terms: benefits became more clearly evident at 300 mg per day and above.

At 300 to 400 mg there was evidence relating to reduced anxiety and to anti-addiction effects in drug-dependent individuals. Effects on insomnia, neurological disorders and chronic pain were described as more marginal and less consistent.

Below 300 mg per day, the authors call for larger and more robust trials before anything can be confirmed.

What that means for a 25 mg gummy

A product delivering 25 mg is at roughly one twelfth of the level where this review found the evidence became clear. Twelve gummies a day would reach it, at which point you would be consuming twelve times the sugar, twelve times the price, and a quantity of cannabidiol that no food regulator in Europe currently endorses.

The safety observation inside the same paper

One useful detail from the same review: increasing from 60 to 400 mg per day did not appear to be associated with an increased frequency of adverse effects. That is a genuinely reassuring finding about tolerability in that range, and it should be reported alongside the efficacy finding rather than left out of it.

What EFSA’s 2 mg is, and what it is not

The number

On 9 February 2026, EFSA published a provisional safe intake level for cannabidiol as a novel food: 0.0275 mg per kilogram of body weight per day, which is about 2 mg a day for a 70 kg adult. It applies to highly purified cannabidiol of at least 98 % purity.

How it was derived

By applying an uncertainty factor of 400 to a starting point drawn from the available data. That is a very large factor, and it is large because the data are incomplete — the endpoints of concern being liver, gastrointestinal tract, endocrine, nervous and reproductive systems, mostly from animal studies.

Why that matters for how you read it

A figure derived through a 400-fold safety margin is a caution level, not a measured harm threshold. Nobody has demonstrated that 3 mg is dangerous. EFSA is saying that the data do not support asserting any higher number is safe, which is a different statement and a more honest one.

Who it does not cover

The level explicitly does not apply to people under 25, to pregnancy or breastfeeding, or to anyone taking medication. For those groups EFSA states the picture is unresolved — which together covers a very large share of the people buying these products.

The one authorised medicine, and why it does not transfer

What the file says

The European assessment file for Epidyolex describes starting at 2.5 mg per kilogram of body weight twice daily, rising after a week to 5 mg/kg twice daily, with a ceiling of 10 mg/kg twice daily for Lennox-Gastaut and Dravet syndromes, and 12.5 mg/kg twice daily for tuberous sclerosis complex.

For a 70 kg adult, the top of that range is somewhere between 1,400 and 1,750 mg a day.

Why you cannot scale from it

Three reasons, each sufficient on its own.

It is a different population — children with severe refractory epilepsy, not adults with poor sleep. It is a different risk calculation, in which liver enzyme elevation is an acceptable cost against uncontrolled seizures. And it comes with monitoring: bloodwork, a neurologist, and a plan for what happens if the numbers move.

None of those three apply to someone buying a bottle online.

The gap in one line

The quantity with a regulator’s signature behind it is roughly 700 to 900 times the quantity the same institution’s food arm calls provisionally safe. Both are correct, because a medicine is judged on benefit against risk and a food is judged on safety alone.

The label is not the dose

Even a chart with a legitimate source would be wrong, because milligrams swallowed and milligrams circulating are not the same quantity.

Oral bioavailability

Everything swallowed passes through the liver before reaching general circulation, and most of the cannabidiol does not survive it. Oral bioavailability is commonly cited at around 6 %.

On a 25 mg capsule, that is roughly 1.5 mg actually circulating. The number on the jar is a statement about manufacturing, not about you.

The routes compared

Two products with identical labels, one swallowed and one held under the tongue, differ by a factor of two or three in what reaches you.

The variable nobody prints

Taking oral cannabidiol with a high-fat meal has been shown to increase absorption several-fold, with figures around four to five times reported.

Combine that with the route difference and the same labelled quantity can vary by an order of magnitude depending on how and when it is taken. A chart specifying milligrams to the nearest five is claiming a precision the pharmacology does not permit.

What follows practically

Consistency beats precision. Whatever you do, doing it the same way each time — same route, same relationship to meals, same time of day — removes the largest source of variation and is the only way to learn anything from your own experience.

“Start low and go slow” is advice about ignorance

What the phrase actually encodes

It appears on nearly every page about this, and it is sound. But it is worth understanding what it is: a strategy for acting when the dose-response relationship is unknown.

Nobody says start low and go slow about paracetamol, because the relationship there is characterised. The advice is standard here precisely because it is not.

Why it is still the right approach

Given genuine uncertainty, beginning at the bottom and changing one thing at a time is exactly what a careful person should do. The phrase is not wrong. It is simply not the reassurance it is usually presented as.

The part usually left out

Going slow only produces information if you also record something. Changing quantity every few days while relying on memory produces a feeling, not a finding. A number written down weekly — a validated scale, a sleep time, a pain score — is what turns the approach into an experiment.

And a defined stopping point

Decide in advance what would count as failure: a quantity you will not exceed, and a period after which you stop. Without one, “go slow” becomes an open-ended escalation with a sunk cost attached, which is how people end up spending a great deal on an answer they never get.

What is in the bottle changes the arithmetic

Content that does not match the label

A 2024 review in The Journal of Pain examined products sold in North America and Europe and found cannabidiol content ranging from none to much more than advertised.

Every calculation on this page assumes the label is accurate. Where it is not, the arithmetic is exact and the input is wrong, which is the worst combination — it produces confident numbers that mean nothing.

The batch certificate is the only fix

A certificate of analysis from an independent laboratory, with a batch code that matches the bottle in your hand, is the only way to know what you are calculating with. A generic certificate on a website proves the company once tested something.

What else to read on the certificate

The cannabinoid profile — CBD, CBDA, THC, THCA, CBN, CBG — and the contaminant panels for heavy metals, pesticides, residual solvents and microbials. Hemp accumulates heavy metals from soil, which makes that panel more relevant here than for most plants.

What we will not tell you, and why

The rule this site is built on

We do not publish a quantity for anyone to take. That is not caution about liability — it is a constraint checked automatically every time this site is built, and a page that breaks it does not publish.

The reason

Telling a stranger how much of a substance to consume, without knowing their weight, their liver, their prescriptions or their pregnancy status, is practising medicine on someone you have never met. That it is common does not make it defensible.

There is also the specific issue: cannabidiol inhibits CYP3A4 and CYP2C19, and the people most drawn to it are disproportionately already taking something that shares those pathways. A generic number is at its most dangerous for exactly the reader most likely to follow it.

What we will do instead

Give you the arithmetic, so you can convert any label into a comparable figure. Name the three official numbers and where they come from. Explain what changes how much reaches you. And be explicit that the gap between the evidence and the shelf is the real finding, rather than papering over it with a grid.

Making the conversation useful

Who to ask

A pharmacist, for the interaction question — it is what they are trained for, it costs nothing, and it takes two minutes. A doctor, if you are managing a condition or taking anything prescribed.

What to bring

• The product label, or a photograph of it, including the supplement facts panel
• The batch certificate, if one exists
• Your full medication list, including anything over the counter and any supplements
• The specific question, which is usually “does this interact with what I take?” rather than “does this work?”

The question that gets a real answer

“Is there an interaction between this and my medication?” is answerable, quickly, from a database a pharmacist already has open. “How much should I take?” often is not, and asking it first tends to end the conversation before the useful part.

The short version

The arithmetic

Percentage × 10 = mg/ml. Total mg ÷ ml = mg/ml. mg/ml ÷ 20 ≈ mg per drop. Price ÷ total mg = cost per milligram. Total CBD = CBD + (CBDA × 0.877).

The three numbers

EFSA’s provisional safe intake is about 2 mg a day for an adult. Products sell 10 to 25 mg. The Sydney review found benefits became clearly evident at 300 mg a day and above.

What that gap means

Products are sold at roughly ten times what the food regulator provisionally endorses, and at a fraction of where the evidence begins. That is a structural squeeze rather than a scandal, and knowing about it is worth more than any chart.

And what no page can give you

A number that accounts for your weight, your liver, your prescriptions and what you had for lunch. Anyone printing one has decided that looking authoritative matters more than being right.

Why we wrote a dosage page with no dose in it

There is an obvious commercial objection to this page: the reader came for a number, and a competitor will hand them one. That is true, and we have decided it does not change anything.

A number invented to satisfy an expectation is worse than no number, because it is acted upon. The reader who leaves here with the conversion rules, the three official figures and one question to ask a pharmacist is better equipped than the reader who leaves a competitor with a cell from a grid — and they are equipped permanently, because the arithmetic works on every product they will ever pick up.

If that costs us the visit, it costs us the visit. It is the same reasoning that keeps clinical evidence and purchase links on separate pages here, enforced by the build rather than by good intentions, and it is the only version of this page we would be willing to sign.