This isn't client work.
Everything here is public. I read MUD\WTR's live ad library the way I would for a brand that hired me, and the receipts — ad IDs and launch dates — are all sitting in the library for anyone to check. No inside data, no claim about their private numbers.
A disciplined account with one live experiment.
MUD\WTR runs a tight operation: one tested block of primary text with many creatives behind it. Two versions of the bullet stack have run side by side since the same day — 4 February 2026 — which makes the pair a live A/B that's now survived months. The only line that really differs is the mushroom bullet:
// both launched 4 Feb 2026 · verbatim from the library
Look at which one they kept scaling.
Every newer launch in the pull — May, then July — uses Variant B, the concrete number. When a brand keeps putting fresh spend behind one version of a live test, that's the version winning. Their own money is saying the specific number beats the adjective.
There's a smaller tell in the same pull that says the same thing: the review count in the ad descriptions climbs from 40K to 50K to 60K while some body copy still reads “40K” — the wrapper keeps getting refreshed, the tested text underneath doesn't. This account respects what's working.
So the lesson is already sitting in their account: numbers beat adjectives. They've just only applied it in one place — the mushroom dose. Variant A is still live, still leading with the adjective.
Take the move that already won, and point it at doubt.
Competitors are hammering MUD\WTR in the same ad library on three things: sediment, “it's basically just chai,” and the subscription trap. Those are the objections costing the sale — and they're being answered, when they're answered at all, with adjectives.
Run the move their own A/B already won — a concrete number instead of an adjective — at the objections, not just the mushroom count. Put a real figure and a demonstration on sediment and on cost-per-cup, in a proof-led creative.
What this read can and can't see.
Public library only. I can see which variant they chose to scale — I can't see the actual performance numbers behind that choice, so “Variant B is winning” is a well-supported inference from where the spend went, not a figure off their dashboard. Video ad scripts aren't captured in a text pull, so this reads primary text and metadata. On the real account, the first thing I'd do is confirm the read against the numbers they can actually see.