Spell It Your Way: What Failed Searches Taught the Omnibox

In The Omnibox we promised recall-first search: type 2:255, a surah name or a phrase you remember, and land on the verse. We said that whether you type Kahf or Al-Kahf, it reliably resolves your intent.

Then we read the log of searches that found nothing. It showed us where that promise broke.

What people actually typed

Very few failed searches were vague. Almost all of them were people who knew exactly what they wanted and spelled it their own way:

• surai al qhamar and 'alrom': a surah name, typed by ear.
• ali 'imran • ayah 96 and شعرا ۱۴۶: a surah and a verse number, the second in Persian digits.
• ألا بذكر الله تطمئن القلوب: a verse typed from memory in everyday Arabic spelling.
• mercifull and Muhamad: an extra letter, a missing one.

Every one of these is a recall search. The person was not exploring. They were being failed by spelling.

One word, two spellings

The Arabic case is the most interesting. The Qur'an text we use is in Uthmani script, the spelling of the printed mushaf. People type the spelling they use every day. The same word looks different in each:

ٱلْقُرْءَانَ (Uthmani) and القرآن (everyday)

The letters differ, in the alif forms and the hamza, before you even reach the vowel marks. So a search that compared text letter by letter found nothing, even for a verse pasted from a well-known source.

Search now compares the two at the level of their shared skeleton: the core letters, without vowel marks or the alif and hamza forms that vary between spellings. Both words above become the same string, so ألا بذكر الله تطمئن القلوب finds Ar-Ra'd 13:28. A citation pasted after the verse, such as "(الأنعام: 97)", is ignored.

It isn't complete. Uthmani spelling sometimes writes a waw where everyday Arabic writes an alif (الصلوة and الصلاة), and those still don't match.

Forgive the query, never the text

The rest follows the same rule. A surah name within a letter or so of the real one opens that surah. A name followed by a number opens that verse. "Prophet" now also finds "Prophets", and "Muhammad" finds the translation's "Muḥammad".

If a search still finds nothing, we try once more with each word spelled the way the translation spells it. When that works, the page says so: No results for "Muhamad", followed by the results for "muhammad". You see what we changed.

That transparency is the line we hold. Search may adjust your spelling, but it never adjusts the Qur'an. The Arabic and the translation are still shown exactly as they come from the source, as described in Chain of Custody. We make the search forgiving so that the text never has to be.

Some searches should still return nothing. A beautiful supplication that isn't a verse of the Qur'an is one of them. Saying so honestly is better than guessing.