Fastest-Growing Beverage Flavor Profiles

What this measures

Two things that turn out not to line up.

The first is demand: how many people typed the flavor into Google, per month, in the United States. That is curiosity, not consumption — but people search before they buy, and buyers search before they brief, so it moves earlier than any sales figure you can get.

The second is supply: how many separate companies actually ship a drink with that flavor in its name. Not how many products — how many companies. Six versions from one firm is a company's bet. One version each from twelve firms is a trend.

The interesting result is that the two are close to inversely related.

The clearest thing in the data

Dirty soda. The Utah drive-through format — soda, flavored syrup, cream — and the cleanest shape of any food trend in the set.

dirty soda

Source: Searches: Ahrefs Keywords Explorer API, US, pulled 11 September 2026. Dates: Fortune, 24 July 2025; McDonald's USA press release, 28 April 2026.

Two step changes rather than a curve. January 2024 ran at 6,128 searches; March 2024 hit 49,400 — eightfold in two months. Then a second climb through mid-2025 to a peak of 80,788 in July, settling near 65,000 where it has stayed for a year. A spike that decays was a news cycle. A plateau held for four quarters is a format that found an audience.

Further back than the chart: the term ran at roughly 1,000 searches a month through 2021. It is up about 65-fold.

The growth ranking

Term	Avg searches/mo	YoY	Prior YoY	Volatility
protein soda	10,492	+202%	+718%	0.72
dirty soda	62,303	+63%	+140%	0.38
black sesame latte	2,337	+51%	+20%	0.26
lion's mane coffee	2,157	+35%	+8%	0.36
lychee drink	2,938	+26%	+12%	0.20
prebiotic soda	13,813	+24%	+82%	0.49
pandan latte	1,308	+19%	+5%	0.50
mushroom coffee	221,485	+19%	+9%	0.16
matcha latte	80,416	+18%	+40%	0.20
taro milk tea	26,219	+17%	-3%	0.18
tepache	36,812	+10%	-6%	0.15
brown sugar boba	10,198	+4%	-7%	0.14
yuzu drink	1,115	-6%	+1%	0.12
butterfly pea tea	12,313	-9%	+7%	0.12
mangonada	64,703	-13%	-1%	0.36
chamoy	95,895	-14%	-29%	0.20
horchata	220,117	-19%	+20%	0.29
ube latte	5,973	—	+25%	—
spicy margarita	12,712	—	+17%	—
pistachio latte	6,065	—	+17%	—
calamansi juice	5,767	—	+6%	—

Source: Ahrefs Keywords Explorer API, US, pulled 11 September 2026. Full 25-term set in the repository. — means Ahrefs projected that term's latest twelve months from the year before instead of measuring them, so no growth or volatility is given.

Three things worth reading carefully.

The prior-year column matters more than the headline. Protein soda grew 202% on top of 718%. Both are large and they say opposite things — still growing, growing far less steeply than it was. A term decelerating from 718% to 202% is at a different point in its life than one accelerating from 8% to 35%, which is where lion's mane coffee is. One year of data cannot tell those apart.

Volatility above about 0.40 means read it twice. Protein soda sits at 0.72, pandan latte at 0.50, prebiotic soda at 0.49. In those series a couple of months carry most of the annual total, which usually means a launch or a press cycle rather than sustained interest. Pandan latte moves between 470 and 2,700 searches with no pattern; at that size the growth figure means very little.

Growth rates on small numbers are cheap. Mushroom coffee gets 221,485 searches a month and grew 19%. Elderflower soda gets 140 and grew 18%. The percentages are nearly identical and the facts are not comparable.

Why the comparison is year-on-year

Beverage search is seasonal enough that month-to-month reading is useless. Mangonada is the clearest case.

mangonada

Source: Ahrefs Keywords Explorer API, US, pulled 11 September 2026.

Every year it troughs near 37,000 in December and peaks near 116,000 in May or June. Start from December and mangonada looks like it triples. Start from June and it looks like it collapses. Neither is true — across three years it is roughly flat and slightly down on the latest twelve months.

So every growth figure here compares a full twelve months against the full twelve before it. Whole years cancel the season out. The cost is that the table cannot detect anything that started in the last quarter, which is the right trade for a report meant to be cited.

What got made

Now set that against the shelf. Distinct brand owners shipping each flavor, counted twice — once in USDA's branded catalogue, once in Open Food Facts, independently built by different people using different methods.

Term	Searches/mo	Brands (USDA)	Brands (OFF)
mushroom coffee	221,485	4	10
horchata	220,117	39	21
chamoy	95,895	5	0
matcha latte	80,416	134	105
mangonada	64,703	0	0
dirty soda	62,303	0	1
tepache	36,812	3	3
taro milk tea	26,219	5	2
prebiotic soda	13,813	10	15
butterfly pea tea	12,313	0	1
spicy margarita	12,712	6	2
protein soda	10,492	0	2
cherry limeade	6,772	61	37
pistachio latte	6,065	6	7
lychee drink	2,938	45	21
yuzu drink	1,115	21	15
tamarind soda	1,029	24	17
guava soda	321	22	9
elderflower soda	140	35	13

Source: Searches: Ahrefs, US, September 2026. Brands: USDA FoodData Central Branded Foods; Open Food Facts bulk export. Both pulled 11 September 2026.

Read the top and the bottom of that table together. Mangonada gets 64,703 searches a month and no US brand ships one. Elderflower gets 140 and thirty-five brands ship it. Dirty soda gets 62,303 and the closest thing to a packaged product is a Coffee Mate creamer flavored to taste like one.

The two catalogues agree at a Spearman rank correlation of 0.808 across all twenty-five terms. Every disagreement sits at the floor — nought against one or two products, which is noise. That matters: this is not a quirk of one dataset. Two separately assembled catalogues say the same thing.

Why the inversion happens

Because search cannot tell a product from a preparation, and the shelf can.

Dirty soda, mangonada and protein soda are things somebody makes for you at a counter. Elderflower, guava and tamarind are things somebody puts in a bottle. Both generate searches. Only one generates a UPC.

So a high search number with an empty shelf is not an unserved market. It is usually a signal that the demand has gone somewhere a grocery catalogue cannot see.

Where dirty soda actually went

To McDonald's.

The chain began testing dirty sodas at more than 500 US locations in September 2025. On 6 May 2026 it launched six specialty drinks nationally, among them Dirty Dr Pepper — Dr Pepper layered with vanilla and cold foam. The press release says the drinks "are here to stay", which is the language of a permanent menu line rather than a limited run.

Worth noticing: all three of the launch's "crafted sodas" are dirty sodas by construction — a soda base, a flavored syrup, a cream or foam finish. Orange Dream is Hi-C Orange Lavaburst with vanilla and cold foam. Sprite Berry Blast is Sprite with blue raspberry syrup and cold foam. Only one of the three carries the word "dirty" on the menu board.

Against the search series, the lead times are:

Date	Event	Lag from inflection
March 2024	Search inflects — 6,128/mo in January to 49,400 in March	—
July 2025	Search peaks at 80,788/mo	16 months
September 2025	McDonald’s begins testing at 500+ locations	18 months
6 May 2026	National launch, including Dirty Dr Pepper	26 months

Source: Search: Ahrefs, US. Test date: Fortune, 24 July 2025. Launch: McDonald's USA press release, 28 April 2026.

One case is not a rule, and the honest reading is narrow: for one format, in one chain, consumer search led the test by eighteen months and the national rollout by twenty-six. Whether that interval generalises is exactly what a longer series would answer, and we do not have one yet.

Why there is no launch chart

We tried to build one. It cannot be done from free data, and the reason is worth recording because it is not obvious from the outside.

Every free food dataset we examined is an archive that stopped:

• USDA FoodData Central all but stopped publishing branded records after 2023. A query for beverage terms returns 2,953 records for 2023 and 54 for 2024. Counting products by quarter would plot USDA's ingestion schedule.
• Open Food Facts is alive and growing, but its US beverage records arrive in irregular bulk waves — roughly 12,000 added in 2017, 14,000 in 2020, 13,000 in 2025, and a few hundred to a couple of thousand in the years between. Also an ingestion pattern, not a market.
• MenuStat, the public chain-restaurant menu database, ends in 2018 and its site no longer resolves.

None of that stops the cross-section above, which asks whether a product exists rather than when it appeared. But it does rule out a launch time series built from the past.

So we are accumulating one forward instead. The catalogues are snapshotted on a fixed cadence, and each snapshot records the exact term definitions it was measured under, so a later change to the tracked list cannot silently corrupt the series. The first snapshot is dated 11 September 2026. Four quarters from now this section gets a chart instead of an explanation.

What is in the tracked set

Twenty-five terms, split between flavors (yuzu, calamansi, ube, pandan, black sesame, lychee, tepache, chamoy, horchata, mangonada, pistachio, taro, elderflower, guava, tamarind, matcha, butterfly pea, brown sugar boba, cherry limeade, spicy margarita) and formats (dirty soda, protein soda, prebiotic soda, mushroom coffee, lion's mane coffee).

The list is deliberately small and deliberately versioned. Adding a term changes what the report measures, so the set lives in a file with its own history rather than being reassembled each quarter.

One limitation to keep in view: both supply catalogues cover packaged grocery products. Neither can see a restaurant menu. Given that the single clearest finding here is a flavor that went to foodservice instead of retail, that gap is the most important thing missing from this report.