CBD levels in flowers: what the certificates say

The survey comes from the catalogue's public API, queried on 16 September 2026. Two calls are enough to reproduce it. The first lists the products with their categories, which isolates the 72 references attached to the flower category. The second fetches, for each reference, the published batch analysis attached to it.

From each analysis, a single line is kept: the one whose symbol is CBD, expressed as a percentage of mass. Three points matter for anyone wanting to redo the calculation.

• The CBD line alone is read. Cannabidiolic acid, CBDA, is neither converted nor added to it. That is the convention applied everywhere else on the site, and mixing two conventions would produce figures that cannot be compared.
• A value showing as 0.0 % is set aside as a data gap rather than counted as a measurement. In this survey, none of the 72 pages is in that position, so the count stays at 72.
• Resins and oils are excluded. The catalogue separately holds 15 hash and extract references and 5 oils, whose levels are not comparable with those of a dried flower, as the article on flower or resin sets out.

The 72 batches carry an issue date between 1 and 27 June 2026. This is therefore a snapshot, not a historical series.

First reading, in bands of two points. Each lower bound is included, each upper bound excluded.

The distribution is clearly bimodal: a first tight group below 8 %, a second between 10 and 12 %, and a trough between the two. That trough is in no way biological, it reflects the make-up of the catalogue, where two families of batches sit side by side.

Second reading, through statistical markers. The right-hand column gives the working, so that every value can be recalculated from the 72 measurements.

The mean is 0.67 point below the median, which reflects the weight of the lower group in the series. That is why the median is the marker to keep: a mean on its own would flatten two distinct populations into a figure describing neither.

For readers who prefer so-called total CBD, that is the CBD line plus CBDA converted by the 0.877 factor, the median of the same survey rises to 11.58 % and the maximum to 15.41 %. The ranking of the references does not change; only the levels go up.

Here is the most useful figure in this survey, and the least flattering. All 72 pages carry a published batch analysis with a usable CBD value. However, only 36 pages carry that level in the field the product page uses to display it. The other 36 show no percentage at all: the value exists in the batch analysis, but it has not reached the page.

The gap is therefore a publishing gap, not an analysis gap. The check confirms it: of the 36 pages that do display a level, all 36 agree with the corresponding batch analysis to within 0.05 point, which is the precision of a one-decimal display. No disagreement at all.

A third figure deserves stating, because it illustrates the subject exactly. Thirty-six product names inherited from the original catalogue still carry a percentage in their wording, and those percentages are worth nothing: compared with the batch measurement, 18 are above and 18 below, with gaps running from 36.62 points below to 9.63 points above. The extreme case is a name announcing 45 % for a batch measured at 8.38 %. The site strips those percentages out of displayed titles for precisely that reason, and shows only the value from the analysis.

The query "30 % CBD flower" exists; matching batches far less so. Three reasons can be read in the figures above.

First, a dried flower is whole plant material: bracts, small leaves, stems. Cannabinoids concentrate in the trichomes, which make up only a fraction of it. Getting past 20 % means removing the rest, which produces a resin or an extract rather than a flower.

Next, the legal threshold constrains the whole plant, not only the final THC. The arrêté du 30 décembre 2021 (French ministerial order) reserves the sector to varieties of Cannabis sativa L. listed in the common catalogue or the French catalogue, with a Δ⁹-THC content no higher than 0.30 %, and applies the same threshold to extracts. Across the 72 batches surveyed, Δ⁹-THC runs from 0.12 % to 0.29 %: no batch reaches the threshold, and the highest point stays 0.01 point below it. That framework is developed in the article on the legal THC level.

Finally, a level announced with no batch number and no analysis commits nobody. A 30 % value seen in a shop is checked like any other: name of the laboratory, date, scope of its accreditation, batch number. The verification method is described in our article on the accreditation of a testing laboratory.

One last point of transparency, because it governs how everything above should be read. The 72 analyses collected here are cannabinoid panels published per batch, measured by high-performance liquid chromatography, with a batch number and an issue date. At the date of the survey, none of the 72 attaches the PDF file issued by the laboratory.

The distinction is not cosmetic, and it is applied internally: a panel with no attached certificate is not announced as a laboratory result. The figures published above therefore describe faithfully what the pages carry, and they should be read for what they are: product reference values, awaiting the corresponding signed document. The checking circuit applied to every batch, from receipt to going on sale, is described on the our laboratory page, and the most frequent reading questions are gathered in the frequently asked questions.

• Phytogrammes, CBD flowers category (the 72 references surveyed)
• Phytogrammes, Gelonade page (highest value in the survey, batch PG-26-9D06)
• Phytogrammes, Triple Cheese page (lowest value in the survey, batch PG-26-116F)
• Phytogrammes, "Comment nous testons chaque lot"
• Légifrance, arrêté du 30 décembre 2021 (article 1, I and III)
• Conseil d'État, decision of 29 December 2022, no. 444887
• ISO/IEC 17025, general requirements for the competence of laboratories
• Cofrac, certificate no. 1-5810 (cannabinoid assay by HPLC-DAD on hemp matrices)