First person invites inferences of capability and blame

What it is

“I booked it for you” and “the order has been submitted” report the same event and license different inferences. First person casts the system as an agent that can act, know, and be blamed: people then ask “why did you do that,” wait for an apology on error, and overestimate what it understands. This is not whether the sentence sounds nice; it is pronouns allocating capability and responsibility. A persona can be stable without “I.” Once “I” is in, capability bounds and the address for blame are written into the grammar together.

Why it happens

Pronouns cue animacy and agency. First person plus an action verb is heard as intention, knowledge, obligation. Listeners treat the interlocutor as a social actor, and on error they look for reasons and someone to apologize — not for a status code. Passive or event report (“living-room light is off”) leaves the agent empty; responsibility sits on a process or the environment, and inferred capability stays narrower.

“I don't know” points at a mind that is ignorant; “that information isn't available” points at missing data. The first invites “then go look it up” or “how can you not know,” recoding the system as a person who will not do their job. On money and medicine, “I already transferred it” also implies that this “I” bears the transfer; the actual actor may be a bank API. First person can raise cooperation on friendly tasks and, in domains that cannot hold responsibility, creates a wrong address for blame.

Studying it

Minimal pairs: the same outcome in “I + verb” versus an event-report sentence. Measure who is taken as responsible, what else the system is expected to be able to do, and whether an apology is demanded after error. After a miss, compare “I got that wrong” with “that song wasn't found”: where blame points, and whether people start arguing reasons with the system. Do not use liking as the dependent: liking can rise while capability inferences rise — those are the two things to split.

In logs, watch whether users switch to “why did you…” after failure. Skills with dense questioning of that form should have their success lines checked for first-person agency.

Where it stops holding

Chinese and Japanese often drop subjects, so “I” leaks less than in English; agency moves into verb choice and mood. Products designed as named companions are supposed to carry some agency; stripping “I” fights the character, but the capability inference remains and has to be aimed in refusals. Children respond more animately to first person. Some accessibility users want a nameable “I.” Screen readers speaking control names do not run this pronoun mechanism.

Applying it

• Decide the speaking position first. Acts the system can complete and be answerable for can use “I turned the light off for you.” Money, medicine, law, and acts that only forward a request use event report: “order submitted,” “living-room light is off.”
• On failure, avoid “I don't like” / “I don't want,” which recode cannot as preference. Prefer “that song wasn't found” or “I can't do that” — and if “I” stays, attach what it can do, so the leftover is not only an unwilling agent.
• Do not use “I already transferred it” for something not yet true in the world, or done by someone else's back office.
• How to check: play a first-person failure to someone who did not write it. If they start reasoning with the system, or ask “then why didn't you…,” the line has allocated responsibility it cannot carry — change the pronoun or switch to event report.

Related

• Same group: M2.10.1 Persona is the joint of wording, sentence length, and forms of address · M2.10.2 Persona fractures first on failure and refusal
• Nearby: M2.05 Persona · M2.04 Error-recovery wording · M4.11 Gender and stereotypes in voice assistants
• Search terms: first-person agency · CASA · blame attribution