The sound of a name works on people who aren't listening

126 undergraduates read a press release about a new ice cream in 2004. Same product, same paragraph, one difference: some saw it called Frosh and some saw Frish. The Frosh group rated an ice cream nobody had tasted as creamier, richer and smoother — 5.06 against 4.25 on a seven-point index. One vowel, and about eighty cents of perceived butterfat.

Then Eric Yorkston and Geeta Menon did the thing that makes their paper (Journal of Consumer Research 31(1):43–51) worth reading. In another condition, the press release said plainly that the name was a placeholder and would not be the name on the tub. Those readers scored Frish 4.36 and Frosh 4.08. The effect vanished on request.

Move the disclosure to after the name, and it came back: 5.00 against 4.09.

So Study 2 ran the same design on 111 students with one addition. While they read, an experimenter read a list of numbers aloud and they had to count how many times the word five went past. Among the readers who had been told the name was fake, on a hundred-point scale, Frosh scored 65.3 and Frish 50.5.

They knew the name meant nothing. They could not spare the attention it takes to act on knowing.

That is the finding this page is built on, and it points somewhere uncomfortable: a name's sound does its work on the people giving it the least thought. Which is everybody except you.

Every naming meeting is, methodologically, the control group

Look at the condition where the effect disappeared. The name is flagged as provisional. It is flagged at the moment of encounter. The reader has attention to spare. Nothing else is competing for the judgment.

That is a naming meeting, described in full.

The people you are most likely to ask — cofounder, three friends, the group chat — get told "we're thinking of calling it X" before they ever meet the word. You have handed them the disclosure and the attention budget in the same breath, which is exactly the recipe Yorkston and Menon used to switch the effect off.

Meanwhile the person who decides whether your name works is reading it once, in a sidebar, halfway through something else. They are in the impaired-capacity condition, and they have never once been told it is only a test.

Both groups will give you a confident answer. Only one of them is a measurement.

The vowels replicate. The consonants are folklore with a good publicist

The core finding is 97 years old and has held up better than most things that age.

Edward Sapir (Journal of Experimental Psychology 12(3):225–239, 1929) made up pairs of meaningless words and asked people which one named the bigger thing. His own example: mal and mil, both said to mean "table." Which table is larger? His main run put 100 such pairs in front of 500 subjects, and preference for the a-vowel as the larger object came in between about 75 and 96 per cent depending on the pair and the group.

Worth knowing who those 500 people were, since the study is quoted constantly and described almost never. They were mostly pupils at the University of Chicago High School: six eleven-year-olds, 124 fifteen-year-olds, twenty-one university students, seven Chinese speakers, and eight adults. Sapir notes the adults were the high-school English teachers, who filled in the forms at the same time as their own classes, and he discounts their scores for exactly the reason you would expect. Sound symbolism was founded in a high school, and it replicated anyway.

Richard Klink's 2000 experiments (Marketing Letters 11(1):5–20) put the directions on record for brand names specifically. Front vowels — the i in kiss — carry smallness, lightness, thinness, fastness, coldness, bitterness, femininity, weakness and prettiness. Back vowels carry the opposite end: bigger, heavier, slower, duller.

Front vowels bring prettiness and weakness in the same delivery. There is no good vowel, only a direction, and the direction is a package.

Now the part that goes missing whenever a naming agency explains that V is the most alive letter or that B is the most trustworthy. Almost all of this evidence is about vowels. Yorkston and Menon say so themselves in their general discussion: the role of consonants is, in their words, an underresearched area. The vowel claims rest on a century of replication. The consonant claims rest on somebody's house style, and the two get sold in the same sentence at the same price.

What fluency is worth, in dollars, with the receipts attached

The most-quoted number in naming comes from Adam Alter and Daniel Oppenheimer (PNAS 103(24):9369–9372, 2006). Sixteen Princeton undergraduates rated 89 shares that began trading on the NYSE between 1990 and 2004 for how hard each company name would be to pronounce if they had to say it at an awards ceremony. Name complexity predicted worse share performance after one day of trading and after one week.

Then the number everyone repeats: a $1,000 investment in the ten most fluently named shares beat the ten most disfluent by $112 after a day, $118 after a week, $277 after six months and $333 after a year.

The $333 is where the internet reliably goes wrong. It is the one-year figure and it carries a P < 0.10, which is a way of saying the paper did not clear its own bar on that one. You will find it quoted as a one-week return in a great many naming articles, and the trail leads back to Princeton's 2006 press release, which gives the $333 with no time period attached at all. Somebody supplied one. Everybody else copied it.

The rest of the paper is more interesting than the headline anyway. A third study sorted 665 NYSE shares and 116 AMEX shares by whether their three-letter ticker codes were pronounceable, which produced the best sentence in the literature: companies with names like Barnings Incorporated outperform companies with names like Aegeadux Incorporated. The pronounceable basket won $85.35 in the first day, then $42.40 at a week, $37.10 at six months, $20.25 at a year. The authors state that the difference stopped being statistically significant after day one, and the caption under their AMEX chart admits that with 116 shares their chance of detecting the effect at all ran somewhere between 6 and 24 per cent.

A paper that publishes its own statistical power against itself is doing something no naming blog has ever done.

Here is what it actually says. Fluency buys the benefit of the doubt from somebody who has no other information about you, and it stops buying anything the moment information arrives. That is a real asset, priced correctly. On the day you launch, nobody knows anything about you, and the name is doing more work then than it will ever do again.

The sound is the smaller half of the sound

Return to the ice cream, because the setup teaches more than the result does. Before Yorkston and Menon could test whether Frosh felt creamier than Frish, they had to prove the two names meant nothing. Fifty-six participants listed the first three words each name brought to mind. Two coders, blind to the hypothesis, sorted the resulting 167 words as positive, negative or neutral. Frosh and Frish came out matched: 30 positive against 28, ten negative against 13, 43 neutral against 43.

That control is the finding. Two names had to be built with no difference in meaning before a difference in sound could be seen at all.

Your name will not be that pair. It will remind people of a word, a company, a person they knew, a thing they use every day. Those associations arrive first and they are heavier than the vowel. Phonetics is the part that shows up when nothing else does.

Three things worth doing, given all that

Check that the name has one pronunciation rather than two. This is practitioner practice from the agency screens (the Interbrand-derived seven-step process puts it at step three), and it is the one phonetic property with no evidence problem: a name that forks into two pronunciations gets said two ways forever, and no amount of sound symbolism survives being said wrong.

Pick a direction. Front vowels for small, quick and light; back vowels for large, rich and slow. A payments company and a sleep app want opposite ends of the same scale. Asking whether a name "sounds good" has no answer. Asking whether it sounds like the thing you are building has one.

Test it the way it will be met. Once, in passing, on somebody with no stake, without telling them it is a candidate. Ask what they think it does, ask them to spell it back, and then leave them alone. The second you say "we're thinking of calling it X," you have run the control condition and the result is worth what the control was worth: nothing about the name.

Sound gets you the first three seconds. Everything after that is what you put behind it.