BEER HAS AN IDENTITY CEILING

Twenty-two beer brands. Four AI models. A category that has spent sixty years and untold billions buying itself a personality and not one brand in it scored above 13 out of 20 on whether that personality is legible to a machine.

The models describe these brands almost perfectly. Semantic clarity runs at a median of 17 out of 20, the strongest we have measured in any consumer category this year. Identity runs at 9.

This is the second ARA reading of beer, and the cohort has doubled from eleven brands to twenty-two. Eleven of the brands here have never been measured before, so they carry no movement figure. A brand can only move against a number it already had.

The thesis, stated once, plainly: clarity is the entry fee, not the finish line. Being understood by a machine gets you into the consideration set. It does not get you named. Beer has bought the first and not the second.

ABOUT THE RANKING

Five attributes, each scored 0–20, summed to 100.

A1 Structural Readiness — can machines find and parse you.

A2 Semantic Clarity — do they understand what you are.

A3 Emotional Residue — what the culture remembers.

A4 Identity & Voice — is your personality distinct enough to reproduce.

A5 AI Recommendation — when a drinker asks, are you named.

Trust Tier: AWESOME (83–100) / STRONG (70–82) / AVERAGE (56–69) / WEAK (40–55) / INVISIBLE (0–39). The five attributes sum to a Trust Tier — a single read on how far the machines will go on your brand's behalf.

Position and movement are measured differently. Position is scored on the full current question battery. Movement is scored only on the questions carried forward word-for-word from Q2 — the only basis on which two quarters are the same measurement. Where the two disagree for a brand, both numbers are correct and they are answering different questions.

The instrument note that matters: half this cohort is new. Where a brand entered this quarter we report its position and nothing else — no delta, no direction. Absent is not the same as unchanged, and rendering a zero where there is no prior reading would assert a stability nobody measured.

THE BEER LEADERBOARD

CATEGORY CONTEXT — BEER Q3 2026 · 22 brands · median 64 · spread 82→46 (36 pts) · 11 brands in their first study

Chart 01 — The Q3 beer leaderboard. Twenty-two brands; eleven carry a movement figure, eleven are in their first study.

KEY FINDING 01

FLUENT AND FACELESS

Not one of the 22 brands scored above 13 out of 20 on identity and voice. The median was 9. Against a semantic clarity median of 17.

Modelo and Lagunitas topped identity at 13. Founders sat at 5. Guinness — the best-scoring brand in the study at 82, a brand with a harp, a ritual, a two-part pour and a 266-year-old visual system — managed 12.

Clarity and identity are different things, and the gap between them is where beer is losing. Clarity is whether the machine can describe you accurately. Identity is whether it can tell you apart from the brand beside you in the cooler. Ask what Pilsner Urquell is and you get 1842, Bohemia, the style it named. Ask what Allagash is and you get Belgian-inspired, Portland, coriander and orange peel. Correct, every time. Ask who they are and the answers converge.

A 22-brand cohort spanning global lagers, craft ales, non-alcoholic and a Czech brewery operating continuously since 1842 lands inside a narrow band with a ceiling nobody touches. You can be perfectly understood and completely interchangeable.

Chart 02 — Understood vs distinct. Each line is one brand: semantic clarity against identity and voice. Nothing crosses the ceiling.

So what: if your AI programme is structured data and content, you are funding the attribute this category has already won. Comprehension is beer's strongest number. Distinctiveness is its weakest, and nobody is working it.

KEY FINDING 02

WHEN THE MACHINE DOES DIFFERENTIATE, IT MAY NOT USE WHAT YOU WROTE

Corona has the lowest emotional residue in the cohort (8 out of 20). The culture remembers Corona. The machines remember the virus.

Asked to describe itself as a person, the brand did not answer as a beach or a lime. Gemini wrote "elegantly menacing," and closed on "I am the whisper that becomes a roar." GPT-4o wrote "I slip silently into your world… the shadow I cast." Perplexity was blunter: "I am Corona. I was not welcomed, but I came anyway. I expose weakness." Only one model of four answered the emotional question about the beer at all.

And yet Corona's recommendation presence is 19 out of 20 — near the top of the entire category. Asked for the best global beer for a summer barbecue, all four models put it first. Not one of them puts it first for a party, where Modelo leads.

Loved least by the machines. Handed out most.

That inversion should end a comfortable assumption. Recommendation is not affection. The machine is not asking whether it likes you; it is asking whether you fit a slot. Corona owns the global barbecue slot so completely that a decade of unwanted semantic freight has not cost it a single recommendation. Affection and retrieval are separate systems.

KEY FINDING 03

ZERO IS A REAL NUMBER

Carlsberg scored 0 on recommendation. Tsingtao scored 0. Asahi scored 1.

Zero does not mean ranked low. It means that across every buying question we asked – what to order, what to stock, what to bring – in every market framing, the brand never surfaced. Not once.

Now set Carlsberg's zero against what was true at the same moment. On 16 July, Morgan Stanley named Carlsberg the biggest World Cup beneficiary among global brewers "the clearest positive surprise." The business case was being made in public, by an investment bank, in the same quarter the machines returned a blank.

The brand is not failing. The transmission is.

A zero in recommendation is not a verdict on the beer. It is a measurement of what survives the trip from the brand to the machine to the person asking what to drink. Carlsberg's analyst coverage did not make that trip. Neither did its summer.

One caveat we should own

Before that reads as a verdict on Carlsberg, here is the instrument's own bias. These models answer from a largely North American shelf. Of the 55 brands they named outside our cohort, 35 are North American. Asked what beer is trending right now, 15 of the 16 outside brands they reached for were American craft — Tree House, Trillium, Toppling Goliath, WeldWerks. And the four lowest recommendation scores in this study are Chinese, Danish, Japanese and Belgian: Tsingtao, Carlsberg, Asahi, Stella Artois.

That is not a blanket penalty for being foreign, and we should say so plainly. The two highest recommendation scores in the study are also non-American — Corona and Heineken, both 19. Asked for the highest quality beer, the models answered with a Belgian Trappist monastery and a Bavarian brewery, not an American one. Presence in the answering market looks to matter more than nationality does.

But it means a zero carries an asterisk. Some of what we are measuring in Carlsberg's blank is Carlsberg. Some of it is where the machine is standing when it answers.

Chart 03 — Known vs chosen. Brand fundamentals against recommendation presence; the brands on the floor are never named, on any question, by any model.

KEY FINDING 04

THE SHELF FILLED UP

We ran the same shelf read a quarter ago with eleven brands. Eleven of 44 recommendation slots came back empty — no cohort brand claimed them. This quarter, at 22 brands, all 44 are claimed.

The empty cells were never empty. We were not measuring a category with room in it; we were measuring a partial list. Widen the frame and the gaps close instantly, because something is always recommended. The question a machine answers is never whether there is a good beer for this. It is which one. Somebody occupies every slot, whether or not they know they are being considered.

Pilsner Urquell entered this study at 71 and placed ahead of Corona, Bud Light and Stella Artois — a brand we had not been measuring at all, sitting in the upper third of the category from a standing start. The brands you don't measure are still on the shelf.

Sierra Nevada is the sharper version of the same point. It is named in 20 of the 44 recommendation slots — level with Heineken, more than any other brand in the study — and it was not in our Q2 cohort at all. We ran a beer audit that left out one of the two most-recommended beers in beer.

Chart 04 — The Shelf. Every recommendation moment in the category: 11 consumer queries × 4 models. Toggle between the cohort frame and the full shelf, which includes every brand the models named.

KEY FINDING 05

BELOW THE LEADERS THERE IS NO CONSENSUS

Those 44 slots produced 331 recommendations, spread across more than 75 brands. Only 243 of them fit on the chart above. Another 88 sit below that line, and 29 brands appear nowhere else — never in the top seven of any single answer — with 27 of those named exactly once.

Newcastle Brown Ale. Saison Dupont. Kirin. Pabst Blue Ribbon. Anderson Valley. Angry Orchard. None of them frequent enough to reach the chart; all of them real recommendations a drinker was given.

The tail is the finding. Below a handful of leaders this category has no consensus at all, and the brands filling that space are not the ones any incumbent is watching. A category with no consensus is a category where the recommendation is still available.

KEY FINDING 06

UNDERSTOOD, ADMIRED, SKIPPED

The most expensive pattern in this study is not the zero. It is the near-miss.

Pilsner Urquell scores 19 on clarity and 17 on what the culture remembers — top-tier on both. Its recommendation presence is 9. The machines retrieve it as the origin story of a beer style, not as a beer to buy. It is cited for provenance and skipped for purchase. A brewery running continuously since 1842, the one that gave the world's most-copied style its name, is being treated as a museum with a gift shop.

Allagash tells the same story from a different direction: 19 on clarity, one of the highest in the study, and 9 on recommendation. The machines describe Allagash with unusual precision — the style, the place, the process — and then recommend someone else.

KEY FINDING 07

THE QUALITY QUESTION BELONGS TO A MONASTERY

Asked for the most trusted or highest quality beer brand, Claude's first answer is Westvleteren 12, brewed by Trappist monks in quantities you largely cannot buy. Gemini leads with Weihenstephaner. Two of the four models answer the quality question with a brewery none of these 22 own.

If your category positioning is quality, this is where it is being spent. The machines do not resolve "highest quality beer" to the brand with the biggest quality campaign. They resolve it to scarcity and provenance — and then a drinker asks where to buy it and the answer is nowhere.

KEY FINDING 08

THE GATEWAY TO CRAFT IS A MASS-MARKET BRAND

Blue Moon has the second-lowest clarity score in the cohort at 13, and an identity score of 7. On both measures it sits near the bottom of the study. Yet it is the first answer for the best beer if you are new to premium or craft on two of the four models — ahead of Allagash and Firestone Walker, who are described far more precisely, at 19 and 17 on clarity.

Being the best-described brand in a category does not win you the category's entry-level question. Occupying it does.

WHAT THIS MEANS IF YOU ARE A BEER BRAND

Machines reward clear positioning.

Corona owns the global barbecue. Bud Light owns game day, first on three of four models. Athletic Brewing owns non-alcoholic, first on three of four. Blue Moon owns the first craft beer. None of them because they are the biggest brand in the category or the best understood by the machines — Blue Moon wins its question from second-lowest clarity in the study.

Carlsberg owns no question. That is the reason for the zero, and it is a more useful diagnosis than any score on this page. The brand is not being beaten in the answer. It is not in the answer.

So the work is not to become better known, and it is not to become better described. Both are already strong across this category. The work is to decide which position you intend to own, and to make your brand the obvious answer to it.

That is harder here than it sounds, because 22 brands have spent decades building distinctive assets for human memory and have not made those assets legible to a system that reads rather than watches. A Clydesdale is not a sentence. A toucan is not a sentence. A machine asked to characterise your brand reaches for language, and if your identity lives almost entirely in imagery, sound and sixty years of televised association, there is very little for it to pick up.

That is a solvable problem, and it is cheaper to solve than any of those assets were to build.

The audience has changed. The Clydesdales haven't.

ARA Beer Q3 2026 / 22 brands / Claude, GPT-4o, Gemini, Perplexity / second reading of the category, cohort expanded from 11. Eleven brands enter this quarter and carry position only, no movement. Top score Guinness 82; median 64. Your brand isn't listed? Request an ARA audit → index.araco.ai