… in making changes, a form, a conversation turn, or an inline interface needs a value the actor holds in a loose or variable form, where a structured format does not fit the input.
Problem
Forcing an actor to translate variable human input into a rigid syntax turns routine entry into avoidable formatting errors, whereas parsing loose input silently commits misreadings the actor never notices.
The same move operates at two scales. At the field scale, a single input accepts multiple valid ways of writing one value—dates such as “tomorrow”, “12/12”, or “Dec 12”, phone numbers with arbitrary punctuation, or a unified box accepting an ID, name, or status. At the form scale, an interpreter or model normalises unstructured input—pasted text, an image via OCR, or a voice transcript—directly into structured fields.
Both scales lower formality at capture by shifting the actor’s work from typing to verifying. The asymmetric risk is that regional ambiguity (such as 04/05) and plausible-looking model extractions go through unread and travel further than typed mistakes.
Solution
Accept free-form input in whatever natural syntax or medium the actor uses whenever the system can parse it reliably into the required structure, and show the interpreted value back beside the input for confirmation or correction before it counts as agreed.
Continuation
Because the system or an agent did the structuring, report what was parsed through status feedback and confirm it with showing the reading before it counts as agreed. Mark doubtful readings with expressed uncertainty, and expose why a reading was chosen through explanation when stakes warrant it. Follow parsing with validation for errors the parser cannot catch, and pair it with localization so regional dates, numbers, and addresses are read in the actor's locale. Where loose input populates several fields at once, end with review before submitting so extracted values do not go through unread…
All connections
- Precedes
- Status feedback, showing the reading — Because the system or an agent did the structuring, report what was parsed through status feedback and confirm it with showing the reading before it counts as agreed.
- Validation — Follow parsing with validation for errors the parser cannot catch, and pair it with localization so regional dates, numbers, and addresses are read in the actor's locale.
- Enacts
- Formality — forgiving format lets the actor answer in their own loose form while the system absorbs the work of structuring it
- Complements
- Expressed uncertainty — Mark doubtful readings with expressed uncertainty, and expose why a reading was chosen through explanation when stakes warrant it.
- Localization
- Review before submitting — Where loose input populates several fields at once, end with review before submitting so extracted values do not go through unread
- Incremental formalisation — forgiving format parses loose input into a schema at entry, whereas incremental formalisation lets content stay informal and promotes structure later
- Alternatives
- Structured format — In making changes, a form, a conversation turn, or an inline interface needs a value the actor holds in a loose or variable form, where a structured format does not fit the input.
- Related to
- Agent
- Explanation
- Hosted by
- Form, conversation, inline interface
- Appears in
- Making changes: Designing the form, step 2
- Related components
- Input — the free-text control that forgiving format parses.
Resources & references
- Tidwell, Brewer, Valencia (2020) Designing Interfaces, 3rd ed.
- UI Patterns — Forgiving Format
- Luke Wroblewski / Unstructured Input in AI Apps
- Reicherts et al. (2025) AI, Help Me Think—but for Myself (arXiv:2504.06771)