Back to Blog

Robot Journalism at

Reading Time 4 mins | December 12, 2019 | Written by: Anne Geyer

Reading time: 2 minutes

Today I ask Dominik Liebsch from about his experiences with text automation. 

Hello Dominik, you are responsible for text automation at What are your tasks?

Hello Saim, on the one hand, I take care of the entire project management of the implementation of text automation projects into our existing business processes. That alone is quite complex: I keep in touch with you and your team, interact with our data providers, and keep an eye on quality assurance and the competition. 

Over the past two years, however, I've also been "digging into" your NLG cloud and learning ATML3, data analysis, and the programming itself. To be honest, the learning process at this point is ongoing on for me, and I am trying to develop further projects besides the constant optimization.

Which departments use the automated text and contributions?

Currently, we have created our own department for TV previews of the daily program, which is completely filled only with automated text that comes from the NLG cloud. In 2017 we started with football: We text the games from the 1st Bundesliga [a professional soccer league] up to the 3rd league and are planning an additional extension with the

text of the games of the lower leagues.

Did the content exist before automation?

With our match reports in football, we have in any case opened up a completely new area for ourselves, which was not previously available in our editorial department -- at least, there was no editor who would have discussed football matches or the like.

The situation is different with the content of the TV program: this content was indeed already there before, written by hand by the editors. Now this text is written by the software--of course, in greater volume, currently about 70 articles per day, with four different versions per article.

Corina Lingscheidt said at the Medientage in Munich that is also profiting from this new area. What do the figures say? Do readers read this content?

In May 2019 we were able to announce the break-even point in my area. The situation is as follows with the football text: we have now increased the traffic from the start days in summer 2017 about tenfold, and the time spent on the content is only slightly longer than with the editorial, non-automated text.

However, the biggest - and most constant - traffic generator is now the automated TV previews, because here we have new content every minute, if you like, which we can produce in contrast to football matches.

A few weeks ago we were able to reach a milestone that was very important for me personally: On a Sunday, traditionally the busiest day of the week for us, the most read article of the day came for the first time from the pen of the Textrobot and was not written by a human being.

You have been using AX Semantics for quite some time. What has improved for you in terms of use?

I've become more and more secure in using the Cloud, and I'm looking forward to your not-so-new editor version, "NEXT", which you've been offering for a year now. Up to now we could not migrate our projects from "Cockpit" to "NEXT" because we needed a certain function. This function is now also available in NEXT, and so we can move soon.

I'm very excited about the new interface, all in all. I know so far, you have had a lot of things to say about Data visualization or process visualization being optimized. I think this is especially important and helpful!

You have a new colleague as of a few days ago. What will his job be, and what will come next if you want to reveal that?

My new colleague fills the role of Content Developer. In the future we want to create still further jobs for this role, which will take care of text automation projects without having to worry about all the stuff around the project work (that will definitely remain my part). At the moment I give my new colleague all the time in the world to familiarize himself with your software and above all to try it out a lot. This is often forgotten in the whole discussion about "robot journalism": It is actually a lot of fun to work in an NLG cloud, to work on logical problems in order to be able to achieve the desired results at some point. If my colleague is ready, we will definitely implement further projects that are already in the pipeline. What these are, I have already hinted at elsewhere in the interview give away ?. Beyond that we also want to concern ourselves still more in detail with questions of the internal data preparation tailored to our interests. Because at NLG, without good data, there are no good automatically generated texts!

Dominik, thank you for the interview.

By the way, the Stuttgarter Zeitung is also successful when it comes to text automation.

And the APA is also very successful with election reports in Austria.

Anne Geyer