Why you shouldn't blindly trust a single THC number

When a package says '24% THC', it looks like a fixed property of the plant – as if you were stating its weight. In fact, it's the result of a measurement that can turn out differently depending on the lab, method, and calibration. Several independent studies show that these differences are substantial in practice – not just theoretical.

This doesn't mean COAs are worthless. It means a single number from a single lab should be treated with healthy skepticism, especially when it comes out unusually high.

Researchers bought 23 flower samples from 10 dispensaries in Colorado and re-analyzed them with HPLC at a single reference lab. Result: the actually measured THC content was on average 23–36% lower than what was originally stated on the label – specifically around 14,98% instead of the stated 20–24%.

78% of the samples were even below the label's own declared minimum value. The researchers also found a link between price and stated potency, suggesting that some producers deliberately choose labs that deliver higher values – more on that shortly.

Important for context: this is a single study with a relatively small sample from a single US state. It doesn't prove a widespread problem in every market, but it shows that the problem is real and documented at this scale.

Another study (Jikomes & Zoorob, 2018, Scientific Reports) had access to the complete traceability dataset from Washington State – not just a few samples, but the entire market. For the same flower chemotype, they found a median THC value between 17,7% and 23,2%, depending on which lab performed the testing.

This isn't variation due to different plant material – it's the same chemotype. The difference arises systematically from the respective lab.

Particularly revealing is a test with distillate: a single, homogeneous sample was sent to five licensed Californian labs. Measured via HPLC, the results ranged from 77,83% to 94,46% THC – a difference of more than 16 percentage points for exactly the same material.

For comparison, the same sample was also measured using a spectroscopic method – there, the spread was noticeably narrower. This suggests that the large range was more due to differences between labs and methods than to the material itself.

Part of the explanation is purely economic: higher THC values often sell for higher prices. When a provider can choose between several labs and one of them tends to deliver higher values, an incentive arises to commission exactly that lab – a behavior known in the industry as 'lab shopping'.

This does not necessarily mean that a lab is deliberately cheating. It can also be due to looser internal standards, poorer calibration, or simply less rigorous methodology. The result for you as a buyer, however, is the same: a number that is higher than it should be.

A ring test (also called an interlaboratory comparison) works like this: an identical sample is sent to several labs without them knowing that it is a test. The results are then compared. Large deviations show where a lab needs to improve.

The 'Emerald Test' is the established proficiency-testing program of the cannabis industry, conducted twice a year for years – it tests potency, pesticides, heavy metals, and residual solvents. Labs that want to hold ISO 17025 accreditation (see 'Reading a COA correctly') must participate in such programs – this is not an optional extra, but part of what accreditation actually means.

The problem is now also on the radar at the government level: the US NIST runs CannaQAP (documented, among other places, in NIST IR 8519, 2024), a program that anonymously sends identical samples to labs across the country to measure the real variance in results. This shows: it is not a footnote, but a recognized, actively investigated problem.

Not every deviation is a red flag. Colorado, for example, regularly allows a tolerance of up to 15% for potency labeling – an edible with a declared 10 mg may officially contain between 8,5 and 11,5 mg without being considered a violation. A certain degree of measurement uncertainty is normal and accounted for.

The deviations described above – 23–36% on average, in some cases more than 16 percentage points between labs for identical material – are, however, clearly above this regularly tolerated variation. That is the difference between 'normal measurement uncertainty' and a structural problem.