iPhone Dictation Should Not Replace Ordinary Words With Contact Names

August 25, 2026 Human Experience Reform

Human Experience Reform public Evaluation

Stable Evaluation ID: UXR-EVAL-0023

iOS Dictation Contacts-Based Personalization, Contextual Ranking, and User Control

Ordinary speech should not become an unrelated person's name merely because that name exists in Contacts.

Bounded Evaluation: current evidence supports a recurring user-visible pattern and control question. It does not establish one internal pipeline cause or population prevalence.

In plain English

One contributor reports that iPhone keyboard Dictation sometimes replaces ordinary words and phrases with names, surnames, or nicknames drawn from Contacts. Some substitutions later self-correct while similar ones persist, forcing the user to reread, repair, change pronunciation, avoid words, and previously even delete legitimate contacts.

Why this matters

Dictation is often used precisely when the user is not watching every word. An unintended contact name can corrupt meaning, confuse a recipient or AI system, expose an association the user did not intend to reveal, and turn an accessibility or convenience feature into a constant proofreading task.

Evidence and limits

Supported

Repeated first-person reports and private screenshots.
An escaped substitution that was interpreted downstream as intentional content.
Apple documentation stating that Siri and Dictation may use contact names, nicknames, and relationships as context.
Independent Apple Support Community reports describing similar substitutions.

Not established

One exclusive internal cause.
A population prevalence rate.
Identical behavior on every device or iOS version.
Legal liability or deliberate design intent.

Current findings

Positive: later sentence context can sometimes correct an initially wrong contact-name candidate, and Apple publicly discloses that contact information may influence Dictation.

Negative: correction is inconsistent, so the user cannot know which plausible-looking proper names will remain. The resulting vigilance, repair, altered speech, communication confusion, and privacy risk are user-borne.

Conditional finding: context sensitivity means the problem can appear intermittent even when it is recurring. That variability is part of the burden, not proof that every transcription fails.

Reform requested

Keep contact-derived personalization subordinate to spoken language and sentence meaning. Provide a clear control that leaves Dictation enabled while independently excluding Contacts-derived names, nicknames, and relationships. Users should not have to delete contacts, rename people, distort pronunciation, or avoid ordinary words.

Acceptance test

With Contacts containing names or nicknames that sound like ordinary English words, ordinary dictated sentences consistently retain the semantically appropriate words. The user can disable Contacts-derived Dictation personalization independently of Dictation itself. When that control is off, contact data alone does not introduce a proper name into the transcription.

What could change HXR's view?

A current Apple control or documented setting that achieves the requested separation.
Version-specific evidence narrowing or resolving the behavior.
Technical evidence explaining candidate ranking while preserving user privacy.
Current tests showing ordinary-word accuracy and reliable control.

Apple: respond or document reform

An authorized representative may correct facts, explain current controls, supply evidence, propose reform, document implementation, or request verification without admitting fault.

Start a private organization response

Have you encountered this?

Share the spoken phrase, resulting transcription, device/iOS version, whether it later self-corrected, and whether the inserted name existed in Contacts. Redact all real contact names and personal details before public sharing.

Privacy: never publish contact identities, nicknames, relationships, support case numbers, or private message content merely to prove the pattern.

Back to blog