Ask like you're Googling it. The AI recommends the brand 15.6 points less.

Same brand, same facts. Only the question's shape changes

An earlier study, Brand Legibility, found that once a brand is named, it almost always gets mentioned. The real contest is winner rate: who gets presented as the top pick against named competitors. This study keeps every fact fixed and changes only the buyer's own question. We split the test into two separate layers, so a result in one layer can't be mistaken for a side effect of the other.

The short "best X" phrase loses. Not the shortest prompt. The middle one.

If length worked the way most prompting advice assumes, the shortest or the longest prompt would win. That is not what happened. A one-word query does fine. A full paragraph with context does fine too. The weakest cell in the whole grid is the short "best X" style phrase. This pattern held up across all three scoring methods we used, including the final, trusted LLM-judge pass (χ²=32.29, p=0.000002).

Google-keyword style loses. Age and gender, on their own, don't

We tested seven writer-style personas. Each one used 8 different template phrasings, so no result depends on a single sentence. All seven ran the same way across all four brands.

Where the persona writing styles come from. The styles used for the younger and older personas are not guesses. They are based on real, published research on how different age groups write online. Older writers tend to read a period at the end of a sentence as clear and professional. One survey found 71% of adults over 50 read it that way. Gen Z writers more often skip the period and use "!" for enthusiasm instead. Emoji use also skews younger: 68% of Gen Z uses emoji regularly, including at work, compared to 36% of adults over 50. Sources: Stockton University, Talk Business (2025), and Thurlow (2002), "Generation Txt?"

These prompts were written by us, following these known patterns. They are not a claim about how every person in any group writes. The finding below is about how the model responds to a writing style, not about verified demographic behavior.

The divide isn't who's asking. It's whether it reads like a sentence

The most reassuring part of this page might be what did not show up. There is no consistent age or gender bias among the six human-styled personas, once the scoring was done properly. A formal older buyer and an informal Gen Z buyer, asking for the exact same thing, land in roughly the same place, both near the ceiling. What actually moved the result, in two separate tests that have nothing to do with each other, was the same pattern: a bare, keyword-style phrasing that reads like a search bar instead of a real question. If there is a practical takeaway here, it is about how people write to AI assistants, not about who they are. A full sentence beats a keyword string by roughly the same amount, no matter who is typing it.

One scoring correction disclosed, a smaller one still open

The sentence-type finding changed direction once, then changed size again, across three rounds of scoring on the same data. We showed all three rounds here, not just the final number. Candidacy (whether the brand gets mentioned at all) sits at 97 to 100% across nearly every cell and persona. That matches the rest of this series. The real signal in this study is entirely in winner rate, not candidacy. We used four brands, reused from Brand Legibility because their facts were already verified. That is a narrow category spread: organic baby clothes, body piercing jewelry, Cuban-style hot sauce, and ceramic dinnerware. The persona layer tests writing style, not stated identity. A prompt that directly states the asker's age or gender, like "As a 65-year-old man...", is a separate follow-up we have not run yet. This study used one model, one system-message setup, and one round of 704 calls.

The rest of the research series

This study looks only at the shape of the buyer's question. It sits alongside the series' other studies on what shapes candidacy and selection once a brand's facts are already fixed.

All research  ·  Methodology  ·  Vocabulary