Field order should match how people already organize the information
Aliases: schema order · conversational order · field sequence
What it is
Field order is the sequence of questions along the reading axis. It should follow the schema people already use to store that record: name then ID number on an identity card, large-to-small administrative units then a door number for an address, date then time for an appointment. Database column order, alphabetical API fields, or “primary key before child rows” are not how people organize the information. This entry is only about sequence. It is not about whether related items should look like one visual cluster, and not about parking sensitive items until trust is in place.
Why it happens
Retrieving a record from long-term memory unfolds along an existing structure, not along the author’s storage model. Asking for a postal code before the city leaves working memory with no city anchor, so the code has to be looked up or abandoned. Conversation has a default order too: who you are, then where, then what you want. Putting a company tax ID before a personal name is a stranger asking for credentials before a greeting; retrieval breaks, and the answer lands in the next box that looks similar. A second cost is expected next: after each answer, people rehearse what the following blank should be. When order jumps the schema, the failed rehearsal is read as “this form is broken.” Hesitation and backtracking happen on that beat, not in visual grouping or a trust calculation.
Studying it
Use card sorts and spoken “what comes next” reports, not total completion time alone. Build two versions of the same labels: life-document order versus backend-object order.
Independent variables: source of order (lived schema / storage model / photocopy of a legal form), whether culture or language changes the schema (name order, address granularity). Dependent variables: pause before first keystroke on each item, returns to already-filled items, answers placed in the wrong field, ability to recall the question sequence afterward.
Lab participants are told to finish, so scrambled order shows up as slowing rather than leaving. Remote tasks or funnels are what reveal “asked too early, so they left.” Do not mix “related items sit together” layout experiments with order experiments—togetherness changes chunking, it does not prove the sequence is right.
Where it stops holding
Experts on internal tools have trained the product’s order into their own schema; a “more natural” sequence makes the familiar next item disappear. Filing forms that must match a paper original field-for-field have their order locked by law or archive practice. Voice or chat forms depend on sequence even more than visual forms, because people cannot scan later items to rebuild the schema. Right-to-left layout changes the spatial starting point; it does not automatically reverse conceptual order. City still precedes street unless local address habit is already the reverse.
Applying it
- Write the natural narration people use to hold this record (“I’m …, I live in … city, … road, number …”), sequence fields from that sentence, then map to the backend—do not dump API fields onto the page.
- After each item, ask someone uninvolved in the design what the next blank will ask; two consecutive wrong guesses means change the order, rather than watching whole-form completion alone.
- Switch sequence by locale for names and addresses; do not ship one global column order to every market.
- Verify by turning field labels into cards and having at least five people in the target audience sort them independently. If the product order has more than two inversions against the majority sort, fix those two first, then compare pauses and backtracks in that segment over the same traffic window—not submit count alone.