Feature request: receiver-oriented phrasing — statements must be true as the user will read them

In long technical sessions, Claude often produces statements that are technically defensible but false as a normal reader understands them. Examples from one evening:

• Reported "Browser X: nothing configured, default settings" — Browser X was not installed at all; only an empty config folder existed, and the model had not even looked inside it (an empty search result was presented as a finding).
• Reported "~260 Mbit/s without VPN" as the line speed — it was a single TCP download; four parallel downloads reached ~440 Mbit/s.
• Described a DNS query as going "via the home connection", which reads as "uses the router's DNS server" — it actually meant "source IP is from the home IPv6 range".

Each time, the user had to detect and correct the misunderstanding. The result: the user must mentally re-check every sentence ("what does it actually mean?"), which destroys trust and is exhausting.

Communication science covers this well, e.g. Schulz von Thun's four-sides model (factual content, self-disclosure, relationship, appeal): every message is received with "four ears", and the sender controls the side he can — precise factual content, no implicit pressure.

Request: make a receiver-perspective check part of the model's behavior — before sending, ask "what will the user believe if he reads this sentence literally, and did I verify exactly that?" For a language model, this is not a nice-to-have; it is the core competence.