Function Words Betray You: Pennebaker’s Pronoun Forensics

James Pennebaker started out studying whether writing about trauma improves health, which it appears to do. To analyze what people had written, he and his colleagues built a program that counted words by category. The program was called LIWC, and the results turned out to be more interesting than the original question.

What predicted outcomes was not the content words — not the nouns and verbs describing what had happened. It was the function words: pronouns, articles, prepositions, conjunctions, auxiliary verbs. The small, invisible words that carry no topic at all.

Function words make up under one percent of a person's vocabulary and over half of the words they actually say. They are produced by procedural language systems, largely automatically, and they are almost never monitored. A person constructing a careful account is choosing content words. The function words come along on their own, which is exactly why they leak.

Pennebaker's corpus findings, developed over two decades and set out in The Secret Life of Pronouns, include several that are counterintuitive.

Status shows up in first-person singular use, and in the direction opposite to intuition. High-status people use I less than low-status people, not more. The reason appears to be attentional: the person with lower status attends to themselves and their own performance, and attention drives pronoun use. Pennebaker found this in email exchanges within organizations, in correspondence, and in his own inbox.

Deception is associated with fewer first-person singular pronouns, more negative-emotion words, fewer exclusive words like but, without and except, and a lower level of cognitive complexity. The proposed explanation is distancing — a liar is less inclined to attach themselves to the statement — plus the reduced complexity that comes from constructing rather than recalling. These effects are real and they are small; this is a statistical signal across many words, not a tell in a sentence, and the literature on linguistic deception detection has the same modest effect sizes as everything else in this domain.

Group identity shows up in the ratio of we to I to they, and it changes measurably during the process of joining a group — which is directly relevant to Part 6 of this guide.

That last finding has an offensive application, and it closes the loop with chapter 12. The we language that Cialdini's unity principle identifies as one of the strongest compliance levers can be engineered. A speaker who systematically replaces I with we, and who refers to the audience and themselves as a single entity, is not describing a relationship that exists; they are asserting one, in the channel the audience is least likely to audit.

Which gives the practical read. Do not analyze someone's pronouns in real time — the effects are too small and you will find whatever you look for. Instead, watch for shifts within one speaker across a document or a conversation: the point where I becomes we, where we becomes they, where a person stops attaching themselves to a claim. The shift is the signal, and shifts are visible without software.