Naming · 5 min read

The dictation test, and why we run it on strangers

Two colleagues laughing together at a desk

If somebody cannot spell it after hearing it once, you will pay for that in every support call for a decade.

The test is simple enough that most teams assume they have already done it. Usually they have not. Reading a name aloud to a colleague who has seen it written down is not a dictation test. Neither is asking a co-founder's partner over dinner. Once somebody has seen the spelling, they cannot unhear it, and the test stops measuring anything.

Why strangers, specifically

Everyone inside the company is contaminated. They have seen the name in the deck, typed it into a registrar search and read it on the brief. They hear the letters because they already know them. Your customers will not have that advantage. The first time most of them meet the name it will be spoken: on a call, in a podcast, across a table at a conference. A stranger is the nearest thing to that listener you can recruit.

How we run it

  1. We model the spelling first, to catch silent letters, doubled consonants and vowels with more than one plausible spelling.
  2. We say the name once, at normal speed, with no spelling and no context about the business.
  3. The listener writes down what they heard, without asking for a repeat.
  4. We run the same test across four English accents.
  5. We compare every written version against the intended spelling, and treat any recurring variant as a failure of the name, not the listener.

The model catches the obvious problems early and cheaply. Only people catch the rest. A name can pass every automated check and still produce three different spellings from three strangers, because real listeners bring accents, habits and expectations that no model fully anticipates. That is why the human round runs on survivors, after the automated screens have done the heavy cutting.

What a failure costs

A name that fails dictation does not fail once. It fails in every support call where the customer has to ask how to spell it, every introduction where somebody adds 'that is with a K', and every email that lands at the spelling you did not buy. We call that last one type-in leakage: traffic and mail going to someone else's address because the obvious spelling of your name is not your name.

Names are read silently in a deck and then spoken aloud for the next twenty years. A name that looks elegant on a slide can be unspellable over a phone line. We have killed favourites because a stranger wrote them down wrong, and we would do it again. Losing a name in the verify stage is cheap. Keeping it costs a decade of explaining.

Share this post

Filed under Naming

All posts

Want these in your inbox?

A short letter, roughly monthly, about naming and the domain aftermarket. No pitching.