Scores and updates. The source is Google Fonts’ public tags/all/families.csv. Data-driven pages request the current file when opened. If the request fails or exceeds five seconds, the site uses a local copy saved October 20, 2026 (CSV). The status below identifies which version is in use. Counts, scores, and data-driven examples update with the loaded file.
Connecting to source…
Fixed research dates. Letterform measurements were primarily taken on October 14, 2026 (terminal measurements on November 1); the WHO TAGS? history runs through October 2026; and the six pairs in 02.3 were selected using the October 20, 2026 copy. These analyses do not automatically change when the live tags change.
What a score means. The CSV records font families, assigned tags, and scores from 0 to 100, sometimes at particular positions on a variable font’s weight or width axis. Google Fonts describes its tags as human-assessed. Its tagging guide explains that font onboarders use GF Tagger to compare fonts carrying the same tag and assign relative percentage values. The public commit history records revisions. The guide does not define what “Business” should mean. A score represents a classification judgment, not the percentage of people who agree, a quality rating, or a measurement of suitability.
Why this library matters. The HTTP Archive’s 2025 Web Almanac found Google Fonts on approximately 54% of desktop sites and 47% of mobile sites in its crawl.
A variable font can contain continuous variation along axes such as weight or width. In the loaded tagging file, 392 of the 1,025 Business-scored families are variable according to metadata in the google/fonts repository (checked October 30, 2026). Families newly added to the tag file also count as variable when scored point by point.
Google assigns some scores to a whole family and others to selected axis positions, for example weights 100, 400, and 900. Different feelings need not be scored at identical positions. A family scored at multiple axis positions is variable; optical size, slant, and other axes do not receive separate scores in this file and are read at their defaults.
01 / Collection. ANY STYLE uses the highest recorded score and EVERY STYLE the lowest. The specimen is shown at the deciding position when possible. “Every” refers to recorded scores, not every point in a variable font’s continuous range.
02 / Perception. Letterforms are measured from Regular, with variable fonts at weight 400 and default width and optical size. Weights per family counts the standard weights a family offers: a variable weight axis extending from 100 to 900 counts as nine, equivalent to nine separate standard-weight builds.
03.1 / Feeling associations. A design uses its highest available score for each feeling, even when those maxima come from different axis positions.
03.2 / Briefs. Business and the two brief-tag scores are evaluated at one scored position wherever possible. A family-wide score applies to every position. Among jointly scored positions, the site selects the one closest to meeting both thresholds (Business 60+, brief fit 50+); ties favor higher brief fit, then higher Business score, then the position nearest Regular. If one brief tag is unscored at a position and the other is 0, the missing score cannot change the resulting minimum of 0.
For the few families in each brief without a jointly scored position, each tag is read at its nearest scored position, and the dot identifies that exception. Specimens, candidates in their settings, and the font carried into 04 follow the selected position, with its style identified when it is not Regular.
Which families count? The Collection includes every family with a /Expressive/Business score, counted once. For families with multiple recorded scores, ANY STYLE takes the highest and EVERY STYLE the lowest. Ties go to the position nearest Regular (weight 400, normal width). These filtering rules and the cutoff are choices made for this project; Google publishes no minimum Business score.
Among families scored at multiple positions, Regular holds the highest score for most. Some were scored only at the extremes and have no recorded Regular score.
How the scale works. Each bar counts families at a given Business score, and selecting a bar or dragging the handle sets the minimum. A score between whole numbers (for example, Martel Sans at 90.6) belongs to the lower bar: it qualifies at 90, but not at 91. The scale is linear.
Families versus designs. Different families sometimes share the same Latin letterforms. To create the DESIGNS view, the letters in “I mean business” were shaped with HarfBuzz and rendered from Fontsource Latin-subset builds at the same size, using the weight nearest 400. Families were grouped when all letters overlapped by at least 99%. The period was excluded because script-specific editions may draw it differently. Close but nonmatching pairs remain separate; Zen Kaku Gothic New and Zen Maru Gothic, for example, overlap by approximately 98%. Groups were computed October 14, 2026; families added since are treated as distinct.
How the grid is arranged. Every family meeting the cutoff appears. As fewer families qualify, their specimens grow and spread out. Ordering remains stable but deliberately arbitrary; neither size nor position indicates rank. Fonts load as their specimens approach the visible area.
Why Business? Using each family’s highest scored position, Business has the most families at every cutoff from 46 to 90 except 51 and 52, where Calm edges ahead. Above 90, Artistic leads. Below 46, Rugged and Vintage lead: they are given more often, mostly with low scores. The comparison includes twenty expressive tags. This statement and the WHY BUSINESS? chart (families scoring 80+) on the Collection page use the currently loaded data.
What is compared? Form sets the range covers 1,663 distinct designs across Google Fonts, including those never tagged Business. Families with matching letterforms count once; the displayed design takes the highest Business score of its grouped families. Each a, n, or o is set from a Latin Regular build at the same nominal em size. For variable fonts, Regular means weight 400 at default width.
Letterform measures. X-height is the height of lowercase x relative to the em. Width is the width of “business” at 1 em. Weight is the stem thickness of n divided by its x-height, measured in Regular. Contrast is the thickest divided by the thinnest stroke of o. Slant is the angle of l from vertical, estimated by rotating its main strokes upright; under 2° is counted as upright, and letters without a clear axis are excluded. Spacing measures side space beside n and o relative to x-height.
Category measures. Classification is Google’s first listed catalog category when a family has several: serif, sans serif, display, handwriting, or monospace. Terminals describes the top hook of a two-storey a as flat, angled, round, tapered, or ball/serif; one-storey a counts as “no hook.” Weights per family comes from the family’s metadata in google/fonts (checked October 30, 2026); a variable weight axis counts the standard weights from 100 to 900 that it supports.
Typical. This is our measure of how close a design lies to the middle of the library on six scalable letterform properties: x-height, width, weight, contrast, slant, and spacing. For each property, percentile distance is 0% at the median and up to 50% at either extreme; Typical averages the available distances. Each design has at least four of the six measurements. Missing measurements are omitted from its average. The result describes statistical centrality within this dataset, not how familiar or suitable a typeface appears. Classification and Terminals are categories, while Weights per family describes range, so none enter the Typical calculation.
How outlines were measured. The six scalable letter measurements were taken October 14, 2026 from Fontsource builds of Google Fonts releases. Terminals were measured November 1, 2026 from Regular files in google/fonts; six subsequently withdrawn families retain Fontsource measurements. The automated terminal analysis distinguishes one-storey from two-storey a, then reads the hook’s outline: a cut has a straight edge between corners; a serif or beak creates an inward corner; a ball widens beyond the supporting stroke. Samples of each type were checked visually, and some borderline endings can fit more than one category. Editions that incorporate instructional lines, knitting grids, arrows, or dots use their corresponding plain editions for this measure.
The typographic measurements were developed for this project, independently of Google’s Feeling scores. Google supplies its font files and catalog classifications, but not these measurements. The three notes above the chart in 02.1 give additional display and measurement details. The little drawings at either end of an axis exaggerate what the measure records; they are diagrams, not specimens from the library.
02.2 / The top of the scale. This view shows the families at the first whole-number Business cutoff admitting ten or fewer (currently —). Their browser-rendered outlines share a baseline and nominal type size.
02.3 / Measured alike, scored apart. Six pairs were selected from Business-scored families using the October 20, 2026 copy. Within each pair, the Regular styles closely match on x-height, word width, stem thickness relative to x-height, and stroke contrast, yet their Business scores differed by at least 40 points. Each pair emphasizes a different visible distinction. Current scores and which is higher come from the loaded file, so subsequent changes can alter the result; the reveal indicates when a gap has narrowed. The interpretive descriptions are our observations, not explanations of Google’s reasons for its scores.
Which designs count? Sections 03.1 and 03.2 use designs in the tag file, with matching-letter families grouped once and absent tags counted as 0. That gives 1,680 designs, compared with 1,663 measured in 02.1. 17 have no Latin a, n and o that could be measured (such as emoji, fonts for Khmer, Lao or Tamil, a barcode and a dot-matrix face). These still count in the tag analyses but cannot appear as measurable Latin specimens.
03.1 / The company Business keeps. Only designs carrying a Business score are shown, grouped into ranges 1–39, 40–79, and 80–100. For each other Feeling tag, the square’s distance from Business shows the percentage of designs in the range that carry it; the word size shows their average score for that feeling. Color reflects a change in the combined association (share × average score) from the lowest to the highest Business range: a gain of at least 10 points moves in (neon), a loss of at least 10 drifts away (white), and smaller changes are outlined. Tags run clockwise from largest gain to largest loss.
A dashed tag represents fewer than five designs; faint squares show where the tag falls in the other ranges. Overlapping labels are shifted outward, with a fine line back to their plotted square. Selecting a feeling shows the highest-scoring applicable typefaces with measurable Latin letters. Values update with the loaded data.
03.2 / Where the label runs out. The six example businesses and their briefs are invented for this project. Each brief is translated from two everyday descriptions into two Google Feeling tags. Its brief-fit score is the lower of those two scores, ensuring that one high score cannot compensate for a low score on the other quality. A design meets both displayed thresholds when Business is at least 60 and brief fit at least 50. User-created briefs use the same calculation.
For variable fonts, the three tags are normally evaluated at the same scored position; see VARIABLE FONTS above for exceptions. The plotted point is therefore a comparison under our scoring rules, not a prediction of real-world suitability. The selected pair of tags, the 60/50 thresholds, and the translation of the brief into Google's vocabulary are our decisions.
How the candidates are chosen. Candidates use designs with measurable Latin letters. When fewer than 30 designs meet both thresholds, the options are: the highest Business score among designs with brief fit below 20 (FAVOR BUSINESS); the best brief fit among designs with Business below 20 and a brief fit above 0 (KEEP THE BRIEF); and a compromise maximizing the lower of Business score and brief fit, shown only when that value reaches 20 (SPLIT THE DIFFERENCE). If some design meets both thresholds and none of these does, the best brief fit among those meeting both is added. When 30 or more meet both, the scores cannot choose, so three of them are offered instead: the best brief fits, with at most one per Google structure tag (such as old-style serif or humanist sans) so that they differ in kind. Ties go to the higher Business score, then the name. Any dot selected on the chart can be added as a further candidate. These rules illustrate different compromises; they are not typeface recommendations.
The settings. Each business is tried in one setting written for this project: a patient message, a booking card, an account statement, an app screen, a tasting menu, a menu board, and a shop sign for your own brief. Within a business every candidate gets the same words, sizes, layout and colors, so only the typeface changes. In each business’s DESIGNS tab the candidates share one frame, and buttons above it switch between them in place. All text is set in the candidate’s scored style, with no bold or italic added, and the type size is held constant, so faces with smaller x-heights look smaller, as they would in use. Names and scores are hidden until a choice is made. The settings show how a typeface looks in use; they do not measure how readers respond to it.
What the data can’t decide. The eight questions, their alternative answers, and the two sample positions are authored for this project. They are prompts for examining priorities, not survey results or an assessment of design ability. Each question presents two defensible approaches; uncertainty and intermediate positions are allowed.
Each wheel segment begins in the middle (undecided). Positions move inward toward answer A or outward toward answer B; one step indicates leaning and two indicate a firm stance. The meanings of A and B differ across questions. Responses are not added into a total score and never select a font on your behalf. Comparing with a sample position overlays its stances without changing yours.
The decision note. The note in 04.2 restates the brief and the typeface chosen in 03.2 (the most recent choice, or another you select), lists up to three of your stances, firm ones before leaning ones, in wheel order, and gives what you accept: the score trade-off of your choice and what each listed stance gives up. The sentence that ends it is yours. Nothing in the note is scored or ranked.
The site does not collect or transmit your brief, guesses, choices, wheel positions, or decision note. Sliders and guesses reset when you leave their pages. Your custom brief, choices in 03.2, stances in 04, and the sentence in your decision note are stored in this browser tab using session storage so they carry between chapters. Closing the tab normally ends that session; the clear control in 03.2 and START OVER in 04 also reset the relevant saved choices. Copying, downloading and printing the note happen in your browser.
The site still makes external requests for fonts from Google Fonts and for tagging data from GitHub. Those providers receive ordinary request information, which can include your IP address. Your choices are not included in those requests.
Interface labels and figures use IBM Plex Mono, which the loaded Google tagging file scores 30 for Business. Headings and reading text use IBM Plex Sans, scored 78 to 90 depending on the style. This pairing is itself an interpretive choice.
Each Collection specimen attempts to show “I mean business.” in its own typeface. Fonts without the required Latin letters, or fonts that fail to load, appear in the browser’s fallback face. The Collection dims these specimens and labels them STAND-IN FONT.