Brand/Non-Brand Splitter

How to use it

1. Enter one brand term per line. Include the official name and genuinely distinct names people use; spacing and simple typos are handled automatically.
2. Choose a GSC or Ads CSV, or paste the table. The file must contain a recognized query or search-term column. To audit classifications, add a Reviewed class column with Brand or Non-brand.
3. Select Split rows. Check the row totals and the exact, contains, and fuzzy match counts.
4. Resolve any reviewed-label disagreements. Verify candidate aliases before adding them to the brand terms, then rerun and export.

Example output Example data

With brand terms Acme Widgets and Acme, these fictional rows demonstrate the actual matching order.

What you get

Brand rows and non-brand rows are mutually exclusive and keep the original header and columns. The method summary counts brand classifications by the first successful method: exact, then contains, then fuzzy. A review disagreement appears only when an imported human label contradicts the matcher. A candidate brand variant is repeated evidence to investigate, not a confirmed alias.

How it works

Brand terms and queries are lowercased and normalized for punctuation and spacing. For each term, the matcher also creates a compact version and adjacent-character-swap variants. It checks full-query equality first, substring containment second, and bounded edit distance against individual query tokens last. Rows are written back with CSV-safe escaping.

Features

• Reads common GSC, Ads, and keyword-tool query headers.
• Preserves every original column in two downloadable CSV files.
• Handles spacing variants, adjacent letter swaps, and limited misspellings.
• Compares optional independent reviewed labels with the automatic split.
• Suggests repeated terms from reviewed false negatives as aliases requiring approval.

Limitations

• Fuzzy matching can produce false positives for short or dictionary-word brands.
• Without an imported reviewed-class column, the tool cannot independently know that its own classification is wrong.
• Alias suggestions require at least two reviewed false negatives and still need human verification; frequent generic words can be poor aliases.
• The tool does not recognize brands semantically, translate names, or infer parent/sub-brand relationships.
• Aggregate “other search terms” rows are treated like ordinary input rows because this UI does not infer platform suppression semantics.
• Files larger than 10 MB are rejected; split them before analysis.

Frequently asked questions