Finding a Name in PDFs When the Spelling Keeps Changing

Last updated August 27, 2026  ·  6 min read  ·  Windows 10 & 11

You search a document set for Mohammed and find nothing. The documents say Mohamed, Muhammad and Mohammad. The person is there in all of them; your search was simply correct in a way the documents were not.

Exact search assumes the document agrees with you about spelling. For names, references and technical terms, that assumption fails constantly — and it fails silently, returning an empty list that looks exactly like an absence.

Where the variation comes from

• Transliteration. Names from non-Latin scripts have several standard romanisations, all correct, none matching each other.
• Typing errors. Anything keyed by hand carries a small, steady rate of error — and one wrong character defeats exact matching completely.
• OCR. Scanned documents produce characteristic confusions: rn read as m, l as 1, 0 as O.
• Formatting differences. Surname first, initials only, middle names present or absent, hyphens gained or lost.
• Reference numbers. Invoice and case numbers written with and without prefixes, spaces, leading zeroes or dashes.

In a set of documents written by many people over several years, all five appear at once.

What fuzzy matching does

Fuzzy search matches text that is close to your query rather than identical to it. The measure is Levenshtein distance: the number of single-character insertions, deletions or substitutions needed to turn one string into another. Mohamed and Mohammed are one edit apart; a search tolerating one or two edits finds both.

PFinder implements this across a whole set of loaded documents, alongside synonym expansion for cases where the wording rather than the spelling differs. It is free, runs on Windows, and searches the documents on your own PC.

Searching for something you cannot spell exactly

The core technique is counter-intuitive: search less of the word, not more of it.

1. Write down the variants you already know

   List the spellings you have seen — in correspondence, in the file name, in an earlier document. This tells you how much variation the set actually contains before you tune anything.

2. Search a short stable fragment, not the full name

   Variation clusters at the start and end of names, so search the middle: 'hamm' rather than 'Mohammed'. A shorter fragment gives the search fewer characters to disagree about.

3. Enable fuzzy matching at a strict setting

   Start with a small edit tolerance. This catches single typos and one-character transliteration differences without letting unrelated words in.

4. Loosen one step at a time

   Increase the tolerance gradually and watch what appears. Stop as soon as clearly unrelated matches show up — that is one step past useful.

5. Use the second phrase to anchor the search

   Combine the name with a distinctive second term using AND — a company, a place, a reference number — so a loose name search stays focused on relevant documents.

6. Export the results before narrowing further

   Export the HTML report at the widest useful setting, so you have a record of everything the search surfaced before you started filtering it down.

Choosing how fuzzy to be

Fuzziness is a trade, and both ends of it fail. Too strict and you are back to exact matching, missing the variants you enabled it for. Too loose and unrelated words start matching, burying the real hits in noise.

A workable habit is to start strict, look at the results, and loosen one step at a time until variants start appearing. Stop as soon as obviously unrelated matches show up — that is one step past the useful setting.

When it matters most

• Due diligence and background checks — where missing a mention is the entire failure mode.
• Disclosure and discovery review — documents written by many parties with no shared convention.
• Genealogy and archive research — historical spellings vary within a single family and a single document.
• Scanned and OCR’d collections — where the errors are the OCR engine’s rather than anybody’s.
• Invoice and reference lookups — where the number is right but the formatting is not.

For the review workflow around this, see searching a document production or disclosure set.

Frequently asked questions