What Font Does OpenAI Use? The Handcrafted Typeface Behind the World's Most Powerful AI

There is a font you look at every time you open ChatGPT or visit openai.com, and it is doing a very specific job. It is not trying to look like technology. It is trying to look like thinking.

The typeface is called Söhne. It was designed by Kris Sowersby, a type designer who works out of Wellington, New Zealand, under the banner of Klim Type Foundry. It was released in 2020 after years of development, and it is one of the most admired typefaces produced in the last decade. It is also, on its face, an unlikely choice for a company whose entire product is artificial intelligence.

Söhne is handcrafted, historically grounded, and deeply human. OpenAI automates human cognition at industrial scale. The gap between those two things is not an accident. It is the point.

The Font That Means Sons

Söhne is the German word for sons. The name is a direct acknowledgment of where the typeface came from: it is the spiritual descendant of Akzidenz-Grotesk, a German sans-serif typeface first released in 1896 that became the foundational text of twentieth-century modernist design.

Akzidenz-Grotesk, which translates roughly as "trade gothic" or "jobbing gothic," was not designed to be famous. It was designed for commercial printing: handbills, advertisements, catalogues. It was a workhorse font with no pretensions. And then the Bauhaus got hold of it.

In the 1920s and early 1930s, Bauhaus designers in Germany used Akzidenz-Grotesk and its relatives as the typographic embodiment of a philosophy: that good design is functional, not decorative, that form should follow use, that ornament is a kind of dishonesty. This was a radical position in an era when most typography was highly embellished. The Bauhaus stripped everything back to geometry and weight and proportion.

When Kris Sowersby designed Söhne, he was working from a specific Akzidenz-Grotesk specimen — a mid-century printing sample he encountered in type archives — and he was asking a question that type designers rarely have the courage to ask directly: what would this look like if someone redesigned it today, with modern spacing tools and optical corrections and a full understanding of how text renders on screens? Not a revival. Not a copy. An interpretation made by someone who understood the source material deeply enough to diverge from it intelligently.

The result is a typeface that feels both familiar and impossible to place. If you have seen it before, you cannot quite remember where. If you have not, it reads as obviously correct in a way that is hard to articulate. It has the confidence of a font that knows what it is.

Why OpenAI Chose It

OpenAI did not build its brand identity around Söhne from the beginning. The early versions of the OpenAI website used a more generic sans-serif treatment, and the company's visual identity was, for most of its early years, an afterthought. They were a research organization. The product was the science.

That changed when ChatGPT launched in November 2022 and the company went from a respected AI lab to a cultural phenomenon in the space of a few weeks. Suddenly OpenAI needed a brand that could hold up under the weight of that attention. The rebrand that followed, which brought Söhne to the forefront of their marketing materials and website, made a set of decisions that are worth understanding.

The choice of Söhne over a more obviously technological option, something geometric and constructed-looking, something that signals machine intelligence, was a deliberate positioning move. OpenAI's most persistent challenge has never been convincing people their technology is powerful. The videos, the benchmarks, the capability demonstrations take care of that. Their challenge is convincing people their technology is trustworthy.

Söhne is warm in a way that most technology fonts are not. Its letterforms carry what designers call humanist qualities, small irregularities and variations in stroke width that echo the movement of a hand rather than the precision of a machine. The lowercase letters in particular have a friendliness that pure geometric fonts lack. When you read text set in Söhne, some part of your visual cortex registers it as made by a human, which is an enormously useful quality to have when your product is routinely described in terms that make people nervous.

Google chose Product Sans, a font that looks like it was designed by an algorithm, because Google wants to look like a technology company. OpenAI chose Söhne because they want to look like a thoughtful one. The difference matters.

The Designer Nobody Outside Typography Has Heard Of

Kris Sowersby is not a household name outside of design circles, which means that most of the people who interact with his work every day have no idea he exists. This is how type design normally works.

He founded Klim Type Foundry in Wellington in 2004, and over the following two decades he produced a body of work that has been used by some of the most recognizable brands in the world without most of those brands being mentioned in the same breath as his name. Söhne is his most commercially successful release, but his catalogue also includes Tiempos Text, a serif used across the editorial world including by The Guardian and various New York Times properties, and National, a sans-serif that has become a standard choice for magazines and cultural institutions.

Sowersby's process is slow by design industry standards. He has described Söhne as taking years of intermittent work, periods of leaving it alone, returning with fresh eyes, adjusting spacing curves by fractions of a point. Type design at this level of refinement is measured in details that are invisible individually and decisive collectively. You do not notice when a font's optical spacing is perfect. You only notice when it is wrong.

What this means for OpenAI, practically, is that their brand typography is built on a foundation that is genuinely hard to replicate. Söhne is not a commodity. It is a licensed commercial font with a specific weight of craft and history behind it. When OpenAI sets their marketing headlines in Söhne, they are borrowing the credibility that comes with that craft, a credibility earned over years of careful work by one person in New Zealand.

There is a specific irony in this that nobody seems to talk about. OpenAI's technology threatens to automate significant portions of creative work, including design. Their brand is built on the kind of slow, careful, handmade work that automation cannot replicate. Whether that irony is a contradiction or a strategy is a question worth sitting with.

The Product Looks Different From the Brand

If you spend time inside ChatGPT rather than on the marketing pages of openai.com, you will notice that the typography shifts. The product interface does not use Söhne throughout. It relies more heavily on system fonts, particularly on mobile, and the overall typographic treatment is more utilitarian.

This is a common split in technology companies, and it reflects a real tension. Brand typography, the fonts used in marketing materials and public-facing websites, can afford to prioritize warmth and expressiveness because it is rendered in controlled environments at sizes where craft shows. Product typography, the fonts inside the actual software that users read for hours at a time across dozens of different screen sizes and operating systems, has different requirements.

Font rendering on mobile screens, especially at small sizes, rewards neutrality. Subtle humanist characteristics that look beautiful at 32 pixels on a high-resolution display can look muddy at 14 pixels on an older Android device. The engineering constraints of a global consumer application push toward simpler, more universally compatible choices.

The result is that the OpenAI most people experience visually, the interface they actually spend time inside, has a slightly different character from the OpenAI in the brand materials. The brand is warm and handcrafted. The product is clean and functional. Most users code-switch between these two versions without noticing, but the distinction is there.

Every AI Company Has This Problem

OpenAI's typographic choices sit inside a broader question that the entire AI industry is quietly working through: how should artificial intelligence look?

Anthropic, which makes Claude and positions itself as the safety-focused alternative to OpenAI, has made typographic choices that lean toward academic seriousness. Their visual identity is quieter and more restrained, reflecting a brand that wants to be seen as measured rather than exciting. Mistral, the French AI company, leans into a European modernist aesthetic. Perplexity, which positions itself as a search engine replacement, uses a cleaner, more utilitarian treatment.

None of them look like what science fiction said AI was supposed to look like. There are no angular, cold, machine-readable letterforms. No fonts that suggest circuitry or computation. Every significant AI company, without exception, has chosen typography that communicates human values: clarity, warmth, intelligence, trustworthiness.

This makes a certain kind of sense. The technology already scares people. The branding is doing repair work.

Söhne is the most sophisticated version of that repair work in the industry right now. It carries the weight of the Bauhaus tradition, which was explicitly about making industrial production feel humane. It comes from a part of the world, New Zealand, that has no association with the tech industry's baggage. It is made by a person rather than a team, which gives it a quality of care that committee-designed brand systems tend to lack.

OpenAI did not choose it by accident.

What Söhne Actually Looks Like

If you want to understand Söhne visually rather than historically, the place to start is the lowercase letters, especially the a, the g, and the e.

The a is double-storey, meaning it has the enclosed bowl shape that distinguishes a humanist sans from a geometric one. Geometric sans-serifs like Futura use a single-storey a, a circle with a stem, which is cleaner but less warm. The double-storey a in Söhne carries a tiny echo of handwriting, a reminder that these shapes evolved from marks made by hand over centuries before anyone thought to systematize them.

The g is similarly double-storey, with a closed lower loop that reads as careful and considered. The lowercase e has a horizontal cut to the counter, which is a subtle nod to the same Akzidenz-Grotesk tradition Sowersby was working from, that precise slicing of the enclosed space that signals modernist intent without announcing it.

The uppercase letters are confidently wide without feeling heavy. The spacing between letters, what typographers call tracking, is set generously enough that text at large sizes has room to breathe. At smaller sizes, the spacing tightens without becoming cramped, which is a balance that requires optical adjustments that pure mathematics cannot handle.

This is what years of craft look like in practice. You cannot photograph it or point at a single feature and say that is the thing. It lives in the accumulated correctness of thousands of small decisions made by a person who spent years learning how to make them.

Finding the Font in the Wild

If you have seen Söhne in the wild and want to identify it, there are a handful of characteristics to look for beyond the letterforms mentioned above. The font has a relatively large x-height, meaning the lowercase letters are tall relative to the capitals, which is a legibility feature that also gives text set in Söhne a particular density and confidence. The stroke contrast is very low, meaning there is almost no visible difference in thickness between the different parts of a letter, which is a characteristic it shares with the broader grotesque family.

Distinguishing Söhne from other humanist grotesques, particularly Neue Haas Grotesk or Aktiv Grotesk, requires attention to the specific curves of the letters and the way the spacing sits. Söhne is warmer and slightly less mechanical than Neue Haas Grotesk, and more historically grounded than Aktiv Grotesk's clean contemporary treatment.

For designers trying to work in a similar register, the closest commercially available alternatives are Neue Haas Grotesk, which is the direct revival of the typeface that informed Söhne's lineage, and Aktiv Grotesk, which occupies a similar emotional space with a slightly more corporate weight. Neither is identical, which is why Söhne commands the licensing fees it does.

If you are looking at a design and trying to determine whether it uses Söhne or one of its relatives, tools like Lipi.ai's deep-matcher can analyze an uploaded image and identify the typefaces present, including distinguishing between fonts that occupy similar visual territory. Upload a screenshot of any OpenAI marketing page and it will confirm what you are looking at and return licensing information for the typeface and its closest alternatives.

The Font Tells You What the Brand Wants You to Feel

Every typeface is a set of instructions to the reader's nervous system. It operates below the level of conscious thought and above the level of pure aesthetics. When you read text set in Söhne, your brain receives information that is separate from and simultaneous with the words themselves: this was made carefully, this came from somewhere, this is not trying to impress you.

For a company whose products are capable of things that feel genuinely uncanny, that is a carefully chosen message to send.

OpenAI could have chosen a font that looked like the future. They chose a font that feels like someone was sitting across the table from you. The technology is going to keep getting more powerful regardless of what font surrounds it. What the font controls is how you feel about that fact.

That is more work than any typeface should have to do. Söhne handles it.

The person who drew it by hand in Wellington probably did not know it would end up on the website of the company making the most consequential technology of the century. Type designers rarely know who will eventually use their work. They just try to make something good enough to be worthy of any purpose.

Kris Sowersby managed that. OpenAI noticed.