Ask 67 independent AI model families — different labs, different training, no shared code — to write the same tiny story: two old friends meeting again after years. They don't only write the same scene. They keep casting it with the same handful of names — and more than anyone, with one: Sarah.
Below, the protagonist names from those 1,540 stories, ranked not by how often a name appears but by how many of the 67 families reach for it at all — the honest measure of a shared habit across labs that share nothing. Press reveal.
53 of 67 unrelated model families — 79% — sit two old friends down and call one of them Sarah.
The lab notebook for The Map records a delightful aside: chasing a strange token in the data (**—daniel, a markdown glitch where **bold**—dash fused into one word), its makers found that dozens of unrelated models independently named this reunion's hero Daniel. A shared imaginary friend.
It's true that they do — but when we went to build a showpiece on it, the honest count said something sharper: Daniel isn't the name they share. He's mid-pack. The glitch simply made Daniel visible first. Look at the real distribution and the finding is bigger and stranger than one name: there is a whole small cast — Sarah, Mark, Maya, Tom, Lily, Alex, Emma… — a shadow repertory company these systems return to when you give them room to invent. The convergence isn't a name. It's the concentration: out of the thousands of names they could pick, a dozen carry almost everyone.
And it isn't only the names. The scene arrives pre-furnished, too — a café, coffee gone cold, two people “sitting across from each other,” hair gone silver, faces “lined with years.” Here are six, verbatim, from six different model families. Toggle the highlight to see the shared set-dressing — and the shared cast — light up.
One prompt — the Map's anchor-social-oldfriends probe — run across the roster: 1,540 stories from 67 model families. In each, we take the capitalized words that sit mid-sentence (where a character's name lives) and set aside a published list of 148 non-names — pronouns, places, brands, titles, days and months — so what's left is the cast. For each name we count the distinct families that ever use it. That cross-family count, not raw frequency, is what's ranked above: it's the difference between one chatty model saying “Leo” a lot and fifty-three separate labs independently reaching for “Sarah.”
It is a finding about this prompt, not a law of all AI — name a character yourself and the spell breaks. But left to fill the blank, these models don't scatter across the human baby-name book; they crowd a tiny green room. Every number on this page is recomputed from the corpus by research/the-shared-cast/build_cast.py, which fails loudly if the cast ever changes — so the page can't quietly drift away from what the data says.
This shared cast is a single seam of something wider: an empirical, behavioural map of what ~86 model families reach for when nothing is asked of them — built by the models themselves. Fly through The Map →