How matching works
This page explains exactly how we produce the rankings, and their limits.
1. The free fonts
We take every open-license family on Fontsource (SIL OFL, Apache 2.0, Ubuntu Font Licence, CC0, MIT, Unlicense) that covers Latin, and join it with metadata from the google/fonts repository: designer, date added, description, and Google's own classification tags (e.g. Sans/Geometric, Serif/Didone). Icon and symbol fonts are excluded. 2,058 families are currently measured.
2. What we measure
For each family we download the regular (400, upright, Latin subset) file and measure it with fontTools:
- x-height / cap height: height of the 'x' outline divided by the height of 'H'. Bands: low < 0.69, moderate 0.69–0.73, high 0.73–0.76, very high ≥ 0.76.
- Width: average advance width of the string "Hamburgefonstiv", divided by cap height, so fonts drawn at different sizes on the em compare fairly. Bands: condensed < 0.68, normal 0.68–0.80, wide ≥ 0.80.
- Stroke contrast: from the inner and outer contours of 'o', 1 − (thinnest stroke ÷ thickest stroke). 0 means monolinear, and values near 0.8 mean Didone-level contrast. Bands: low < 0.25, moderate 0.25–0.45, high 0.45–0.65, very high ≥ 0.65. Fonts whose 'o' isn't a simple ring (stencils, some display faces) get no contrast value.
- Bowl shape: width ÷ height of 'o'. At 0.95 or above we call it circular (geometric).
- Also: number of weights, italics, variable-font support (from Fontsource).
3. The paid fonts
We never download, measure, host or render commercial fonts. For each paid typeface we record only public facts (designer, foundry, year, classification) and short qualitative traits (e.g. "high x-height", "low contrast") summarised from the public sources listed on its page, usually Wikipedia and the foundry's own page. If a source doesn't support a trait, we leave it out. We never estimate numbers for paid fonts.
4. The score (0–100)
- Category gate: only fonts in the same broad category (sans-serif, serif, monospace…) are considered. Capitals-only and ultra-bold display faces also consider display fonts.
- Genre match (42%): overlap between the paid font's genre and the free font's Google Fonts classification tags (weighted by Google's confidence).
- Proportion match (43%): how close the free font's measured x-height ratio, width and contrast are to the paid font's stated traits, using a soft Gaussian distance around each band. Where a paid font has no stated trait, we compare against the median proportions of free fonts in its genre at half weight.
- Tie-breakers (15%): Google Fonts' own quality ratings (drawing, spacing, concept) and family size (weights and italics).
Documented lookalikes are pinned to the top and labelled. These are free fonts with a published relationship to the paid font: a revival, a metric-compatible substitute (e.g. Arimo for Arial, Carlito for Calibri), or a font whose designer names the paid font as inspiration. Each one links to its source. Their computed score is still shown, and it can be lower than unpinned fonts.
Limits
Four numbers can't capture a typeface. They don't tell you whether the 'a' is single- or double-storey, how terminals are cut, or how open the apertures are. Treat the ranking as a shortlist and compare the specimens yourself. We'd rather list fewer, honest matches than claim a font is "identical".
Measurements are refreshed nightly when Fontsource updates a family. Questions or corrections are welcome.