Figment has been turning Figma files into live websites since 2021. Over that time 35,077 people signed up, 31,402 websites got created, and 20,048 of them got published to the internet.

I've always been curious what's actually inside those files. Not what designers say they do, but what the node trees say they did. So I pulled a random sample of 3,000 website records out of our production database and counted things.

Everything below comes from that sample. I've also published the raw aggregates as JSON so anyone can check the arithmetic. There's no customer data in any of this: no names, no emails, no domains, no page text. Counts only.

The sample

3,000 website documents drawn at random from the 31,407 sitting in the collection on September 13, 2026.

Of those 3,000:

That 62.8% published rate lines up with the 63.8% we see across all 31,402 websites, which is a decent sign the sample isn't skewed.

The quarter of records that are empty is worth sitting with. A lot of people create the project, then never come back with a design.

Almost nobody builds a multi-page site

This was the biggest surprise in the whole pull, and it's a little painful given how much work went into multi-page support.

93.9% of designs with any content are a single page. Among designs that actually got published, 99.1% are a single page. The median is 1 and so is the 95th percentile. The largest design in the sample had 46 pages.

Whatever people think a website is, in practice what they build in Figma is one long scrolling page.

The fonts

I walked every text node in the sample and read the font family off it. 1,984 designs contained at least one text node with a font, and across all of them I found 880 distinct font families.

Share of designs using each family (a design counts once for a family no matter how many times it appears):

Honest caveat on the top slot: Inter is Figma's default typeface, so some of that 37.7% is text somebody typed and never restyled. The four names at the top are all free Google Fonts, and counted by individual text element rather than by design, they cover 34,080 of the 69,158 text elements that carry a font. Four typefaces, 49.3% of all the text.

The tail is long and mostly dead. Of the 880 families in the sample, 558 appeared in exactly one design, and 684 appeared in two designs or fewer. So 63% of the font families in use show up in a single design and nowhere else.

Per design, the median number of font families is 1. 998 designs used exactly one family, 571 used two, and 219 designs had no text in them at all. The maximum was 13 families in a single design.

Grouping designs by the year they were last edited, Inter's lead widens and Roboto's fades:

Type sizes and colors

I counted 69,845 font-size declarations across the sample. The median is 18px, the 90th percentile is 40px.

Six sizes cover 57% of all text on these sites. Everyone is building on the same handful of steps.

Text color is even more concentrated. Across 68,168 color declarations I found 3,463 distinct colors, but 27.8% of all text is pure white and 22.3% is pure black. Exactly half the text in the sample is #000000 or #FFFFFF. Widen that to any greyscale value, where red, green and blue are identical, and it's 66.6%.

Mobile barely exists

Figment stores a desktop frame and a mobile frame separately. Of the 2,203 designs with content, 62 had a mobile frame. That's 2.8%. Among published designs it's 15 out of 1,882, or 0.8%.

That doesn't mean the rest are desktop sites, though. Look at the width of the primary frame:

13.9% of designs are drawn at phone width and have nothing else. Those people designed a mobile site and called it done.

The single most common canvas width is 1440px at 34.5% of designs, then 1920px at 14.7%, then 1512px (the MacBook Pro 14 logical width) at 3.9%. So the practical picture is that about 14% design only for phones, about 3% draw both, and everyone else draws one desktop frame and lets it scale.

How big is a design, really

A typical website design is 39 elements arranged four levels deep, with 13 pieces of text totalling about 300 characters of copy. That's a hero and a couple of sections under it. The biggest single design in the sample had 3,425 elements, and the biggest single page had 1,575.

Across all 204,801 elements I counted:

Component instances are 5.6% of all elements, and the whole sample contained only 104 component sets and 663 components. For all the effort the design tooling world puts into design systems, the people shipping real websites are mostly drawing rectangles and typing into them.

On the content mix, 30.6% of designs are image-heavy (more than half the leaf content is an image), 13.4% are text-heavy (under 20%), and the rest sit in between. 129 designs had no images at all, 213 had no text at all.

Two thirds of these designs have nowhere to click

67.6% of designs with content contain zero links of any kind, counting both Figma prototype links and text hyperlinks. Among published sites it's 66.2%. The median number of links in a design is 0. The most in any single design was 56.

That number stayed stubborn no matter how I sliced it:

Big designs, recent designs and published designs all land in the same band. A design can look completely finished and still have nothing clickable in it.

What people skip

Out of all 3,000 sampled sites, 6 had set a custom site title, 3 had written a meta description, and 1 had uploaded a custom favicon. Figment fills in defaults for all three, so what this measures is how many people bothered to override a default rather than how many sites have titles. Still, 6 out of 3,000.

Of the 1,885 published sites in the sample, 24 (1.3%) are on a subdomain they chose. The other 98.7% kept the auto-generated one.

Things I checked that turned out boring

In the interest of not cherry-picking:

• Custom domains. The field exists in the schema and was empty on all 3,000 sampled sites, so custom domain adoption is not measurable from this collection. I've left it out rather than report a zero I don't trust.
• Page-level SEO settings. Empty on every single page in the sample. Dead field.
• Ownership and collaboration. The owner list was empty on all 3,000 documents, so the field isn't populated and collaboration can't be measured from it. Same story for the stored Figma file link, present on 29 of 3,000.
• Video and inline SVG. Zero of the 204,801 elements carried stored video or SVG data, so there is nothing to count at the design layer.

Methodology

Source. The websites collection in figment.so's production database, read-only, on September 13, 2026. The collection held 31,407 website documents at that moment.

Sample. 3,000 documents, selected by running a seeded Fisher-Yates shuffle (mulberry32, seed 42) over the full list of document IDs and taking the first 3,000. Every document was fetched and parsed in full. At this sample size, a percentage measured across all 3,000 has a margin of error of about 1.7 percentage points at 95% confidence. The 31,402 figure quoted at the top is our count of websites created; the collection itself enumerated 31,407 documents on the day of the pull.

What counted as what. An "element" is one node in the stored Figma node tree, including frames, groups, shapes, text and images. A "design with content" is a sampled website whose stored tree has at least one node (2,203 of 3,000). "Published" means the record has at least one entry in its published URLs list (1,885 of 3,000). Font shares are computed over the 1,984 designs that contain at least one text node carrying a font, and a design counts once per family regardless of how many times that family appears.

Exclusions. 895 font-size declarations stored as Figma's mixed-value symbol were dropped from the type size numbers. Page-level settings objects were empty everywhere and were dropped. Custom domain data was empty everywhere and was dropped.

Privacy. No personally identifying information was read into the aggregates or retained: no names, emails, domains, site URLs, document IDs or page copy. Every figure here is a count or a percentile over the sample.

Raw data. The full aggregate output is published as JSON at figment-design-research.json in our repo. If you want to cite any of this, cite the numbers here and link back. If you think I've counted something wrong, tell me and I'll rerun it.

If you want to build one of these yourself, Figment publishes your Figma file as a real website. See how it works.