Inkitt, a self-publishing platform using AI to develop bestsellers, books $37M led by Khosla | TechCrunch

Everyone has a story in them, as someone famous once said. A startup called Inkitt believes that it can use AI to turn the strongest of these into blockbusters and to build a new “Disney” for the 21st century around that content. Now, it’s raised $37 million in aid of that ambition.

The startup’s eponymous app lets people self-publish stories, and then, using AI and data science, it selects what it believes are the most compelling of these to tweak and subsequently distribute and sell on a second app, Galatea.

Its business has already attracted 33 million users and dozens of bestsellers, the company said. The new funding that it’s raised, a Series C, will be used to expand the kind of content it produces: AI to write stories based on your original ideas, and to produce versions of its fiction personalized for specific readers; a move into games and audiobooks; and more video content adapted from fiction published on its platform — video that is produced with humans today but will, eventually, also be generated using AI.

The longer-term vision, says CEO and founder Ali Albazaz, is to expand its content library and to build multimedia empire around it. Less a competitor to the likes of Wattpad and more “the Disney of the 21st century,” as he calls it.

The $37 million, a Series C, is being led by Vinod Khosla of Khosla Ventures. Previous backers NEA and Kleiner Perkins, along with other undisclosed investors, also participated.

The investment brings the total raised by Inkitt to date to $117 million (other rounds included a $3.9 million seed; a $16 million Series A and a $59 million Series B). It’s already been approached by at least one publisher to be acquired, and at a time when there are precious few examples of consumer-focused startups raising funding, we understand investors are already approaching it for another round, if Inkitt chooses to raise it. This current Series C values Inkitt at around $400 million post-money.

Inkitt’s ambitions are, in a way, a notable zig to the publishing industry zag.

With the rise of mobile phones and apps that are built to occupy minutes (or sometimes way more) of consumers’ leisure time, reading has been in decline for the last two decades. In 2022, the most recent year tracked by the Bureau of Labor Statistics, the average American adult read only around 15 minutes per day, down from more than 20 minutes several years prior. The average number of books read (and completed) has also dropped to around 5.

Inkitt is betting against that trend by, ironically, leaning into it. It believes that it can pull in more significant amounts of reading time and engagement from its users, by focusing on more innovative ways of delivering books, building out chapters that are shorter and easier to read on mobile devices and incorporating a number of effects (such as sounds) throughout the text to make the reading experience more dynamic.

And also by making the books more fitting to readers’ tastes. It runs A/B tests around every aspect of the work from titles and story arc, to first lines and cliffhangers. That gives Inkitt a trove of data not just for specific books, but around how fiction in general is performing, and what performs best — learnings that it applies in turn more immediately to subsequent books getting published.

All of that appears to be paying off. Some services that flourished during the height of the Covid-19 pandemic — “instant” deliveries, online shopping, virtual events, and streams of everything — have deflated, or at least returned to more tame growth patterns, in the years since. Not so for Inkitt, Galatea and the startup’s newest addition, Galatea TV, which have all see engagement time go up.

“People have so much going on now,” Albazaz said of the post-pandemic years. “They are looking for an escape, and we think that’s why we have been thriving.”

He said that Inkitt, in aggregate, ranks as the number 11 best-selling publisher globally, above household names like Penguin Random House, claiming that its algorithms give it a “20x” higher success rate than traditional publishers for publishing a best-selling book.

Revenues have doubled in the last year (although it doesn’t disclose an actual number). And while newer projects like Galatea TV are more an offshoot of the reading enterprise,  they are yielding big returns in their own right. A TV series based on the Galatea book Beautiful Mistake alone has generated revenues of $500,000, he said. The company is, he added, “almost profitable.”

Inkitt, founded in Berlin and now based in San Francisco, is focussed on fiction — but it wouldn’t be totally accurate to say it’s aiming for literary fiction. In fact, after a famous author dabbling on the platform several years ago walked away in horror and disgust after Inkitt suggested a number of edits to him for an upcoming novel, Albazaz said that Inkitt prefers to save itself the headaches of working with big names and big egos to focus on the long tail of undiscovered talent.

It will be interesting to see how amenable that long tail is to succumbing to the will of the algorithm. Albazaz said that AI-generated stories, based on catchy treatments crafted by talented humans, is very much part of what it hopes to do more of in the future, along with personalizing stories using AI algorithms.

Personalizing is still very much a work in progress, he added, with Inkitt trying out different degrees of changes that it might implement on an original work, or even potentially giving over the controls to readers themselves.

To see out this vision, the company has been experimenting with a number of LLMs, including APIs from OpenAI, Anthropic and Mistral AI (the startup’s current favorite for shorter passages, Albazaz said) for narrative construction, building its own customizations around them, he said.

Every book also gets automatically published in 10 languages: DeepL is the primary AI that’s used for this purpose. It is using ElevenLabs text-to-speech voice generator for its audiobooks, and currently Leonardo for cover art (a recent switch from Midjourney, he said). The main idea seems to be to stick with a mix to triangulate for the best result.

“These LLMs write very bad to average content,” he said. “The point is that LLMs alone are not able to create best-selling content. That’s where we come in with the data that we have gathered over the last couple of years.”

Interestingly, there is an IP reason behind this, too: Inkitt doesn’t want to lean on any single LLM for a full novel because, even though platforms insist they have no intention of using proprietary data to train their models, Inkitt prefers not to take a chance, he said. (To that end, it also sometimes asks very intentionally flat prompts, to avoid teaching the LLMs too much about Inkitt’s best work.)

Even while navigating those tricky waters to keep from drowning, the startup has definitely found opportunities to float. Its focus and interest in building an AI-powered content empire helped the startup secure its lead investor. Vinod Khosla at the end of last year wrote on how he believed the future of entertainment would be hyper-personalized, making Inkitt’s mission (and success so far) a close fit.

Vinod himself declined to be interviewed for this story, but provided a statement: “With the advent of AI, entertainment could be plentiful and personalized,” he said. “Inkitt is doing just that with stories, creating content that is hyper-personalized and meaningful to every person.”