Most searching now happens on a phone, with one thumb, often quickly. That produces a specific and predictable set of query errors, and a catalogue that expects exact spelling will fail a large share of the people trying to use it.

What thumb typing actually produces

Four error types dominate, and they're consistent enough to design around.

  • Dropped letters. Vowels go first, because they're often unstressed and the word remains guessable.
  • Transpositions. Two letters swap when both thumbs move at once.
  • Doubled letters. A key registers twice on a slow response.
  • Truncation. The query is submitted before it's finished.

None of these are carelessness. They're the physical result of small keys, imprecise contact, and variable response times, and they happen to everyone.

Rate matters too. People type search queries faster than they type messages, because a query feels disposable and nobody proofreads one. That combination — high speed, no review — is why query text is consistently messier than any other text the same person produces.

Deliberate shortening is different

Separately from errors, people shorten queries on purpose. Fewer characters means fewer chances for autocorrect to intervene, and in this field there's an added reason: a partial word attracts less attention on a shared screen.

Autocorrect is the specific villain. It's tuned for ordinary language and reliably mangles anything specialised, replacing a term with a common word that shares a few letters. Typing less, and stopping early, is a rational defence.

This is the same discretion logic that made numeric identifiers popular in the first place, described in our piece on why people share numbers.

What catalogues should do about it

TechniqueFixesCost
Fuzzy matchingDropped and doubled lettersLow
Transposition toleranceSwapped letter pairsLow
Alias tableSpelling variants and abbreviationsManual but small
Prefix matchingTruncated queriesLow
Did-you-meanEverything elseModerate

An alias table is the highest-value item because this field's vocabulary is fragmented across spellings anyway — closed up, spaced, hyphenated, pluralised — the split described in our piece on spelling variants.

Prefix matching handles truncation almost for free, and truncation is more common than any single misspelling.

Live results as you type help more than any of it, because they let a person stop early once they see what they wanted. That turns truncation from a failure into a feature, and it removes most of the pressure to spell anything correctly in the first place.

The suggestion feedback loop

Search suggestions are built from what people already typed, which means errors feed back into the system. A common misspelling appears as a suggestion, more people select it, and it becomes more common still.

That's why misspelled forms persist for years rather than dying out. They're being actively reinforced by the interface meant to help.

Catalogues can break the loop by mapping variants to a single entry internally while still accepting all of them at the front door. The user gets what they wanted; the index stays clean.

What they should not do is silently rewrite the query without saying so. Showing results for a corrected term, with a visible note and a way to insist on the original, keeps the user in control while still doing the helpful thing.

What good search doesn't excuse

Tolerant matching is a convenience layer. It doesn't substitute for real metadata underneath — title, artist, year, format, language, tags — which is what makes results meaningful, as we set out in the piece on metadata.

And it changes nothing about the floor: characters are fictional and unmistakably adult, with no ambiguity about age, no real-person sexual depiction, and no coercion framed as romance. Our standards are in the control room, and connected creation with real account and moderation controls stays on models.

Designing for how people actually type

The mental model to discard is the one where a user types a correct query and the system finds it. Real queries are short, misspelled, half-finished, and typed with one thumb while doing something else.

Designing for that isn't lowering standards — it's matching the interface to the input it actually receives. Fuzzy matching, aliases, and prefix search cover almost everything, and they're all cheap. The alternative is a catalogue that works perfectly for people who type carefully on a keyboard, which is nobody.