People often ask me about my experiences with blogging: why I do it, how much time it takes, benefits, etc. I thought I would write a “meta” blog responding to such questions.

Why I blog?

I started this blog in late 2016, largely because I was worried by much of what I saw in academia and NLG research. At the time, NLG (like the rest of NLP) suffered from “LSTM mania”, with lots of scientifically dubious papers at top venues (early blog) and considerable hostility to papers which didnt use LSTMs. Despite the fact that LSTMs were really bad at the kind of NLG I am interested in! Evaluation was also poor (early blog), dominated by BLEU and similar metrics. And it was difficult to publish unusual papers, even though these often have the most long-term impact (early blog).

So I decided to create a blog “soapbox” where I could express my concerns, try to make constructive suggestions, share excitement about positive development, and also try to educate people. Ive been running it for almost 9 years now, and the underlying objectives are still there

• Technology: Neural NLG has massively improved since LSTMs, but it still has limitations, which people need to appreciate and understand
• Evaluation: Techniques are much better (eg, LLM-as-Judge is much better than BLEU), but poor research culture still encourages weak evaluations.
• Academic life: NLG and NLP communities have exploded in size, which raises serious concerns about things like reviewing quality. And it is still difficult to publish unusual papers. I also now write occasional blogs about what my students and colleagues are doing (example), because it gives them more visibility (and in some cases has led to useful contacts or even seminar invitations for my students). I also write personal blogs (example) once in a while, usually about holidays, this is mostly for my own benefit.

I am considering “deprecating” out-of-date blogs. For example, my most read blog is How do I Build an NLG System: Tools?. I wrote this in 2017 and it is completely out of date in 2025.

How much time do I spend on blogging

Obviously the time I spend on blogs varies a lot week by week, but in general I try to put out a blog once every two weeks, and writing the blog takes 2-4 hours. So 1-2 hours per week. Which sounds like a lot, but its often something I do to relax when I need a break from marking, reviewing, admin nonsense, etc.

Its also worth keeping in mind that writing blogs is much faster than writing formally-published position papers. In 2019 I took a blog (NLG and Explainable AI) and turned it into a position paper for a workshop (Natural Language Generation Challenges for Explainable AI). The blog took a few hours to write and has been viewed 1078 times. The workshop paper took maybe 2 weeks to write (including revision, camera-ready, and preparing the presentation) and got 53 citations. After this experience, I decided it was a mistake (not time-effective) to turn blog posts into formal papers. I did however incorporate a lot of blog material into my recent book.

Benefits of blogging

One thing that has surprised me is that when I meet a researcher I dont know, they often mention my blog. Ten years ago when I met a researcher who was familiar with my work, they might say that they liked some of my research papers (or open-source software). But in the past five years, they are more likely to say that they like my blog. I’ve even sometimes seen my blog posts cited in other people’s formal research papers. So it has been good for my visibility in the community.

Of course, the bigger question is whether my blog has helped addressed the above concerns, eg encouraged people to do better evaluations. I like to think that the blog has helped in a small way. Many people have told me that they really appreciated the fact that I was airing such concerns, and in some cases seeing that I shared these concerns gave such people more confidence.

Almost all (97%) of my non-personal blogs get viewed at least 100 times (I don’t really expect other people to want to read about my holidays!), and 20% of them have been viewed over 1000 times. Hopefully this means that people find them to be useful.

I guess blogging about the above concerns also gives me the feeling that I am trying to do something constructive about them, instead of just “complaining in the pub”.

Should you blog?

Sometimes people ask me if they should start a blog. Of course blogs are very different! I have settled on producing biweekly blogs which are around 1000 words and which mostly address scientific issues but are intended to be easy to read. Other people write blogs which are longer or shorter, more/less frequent, more technical or more business oriented, etc. So the first question for potential bloggers is what kind of blog they want to write. Of course this can evolve over time! Sometimes people start with sporadic and quite diverse posts, and then over time settle on a particular pattern.

Of course bloggers want their blogs to be read. I write my blogs with WordPress (which I might drop if I was starting again) and publish them on ehudreiter.com; I “advertise” them on Twitter, Bluesky, and LinkedIn. I suspect I would have gotten more readership, especially initially, if I used a blogging/publishing platform such as Medium or Substack. I did not do this because I wanted full ownership and control (I have had some bad experiences in past where I used a web service which initially was great but then shut down or degraded), but Medium or Substack could make sense for many people.

Another possibility is just to publish blogs as LinkedIn posts. I actually do this when I write something shorter (100-200 words), but I like my longer 1000-word blogs to be more “formally” published and indexed.

Final Thoughts

I have written 245 blogs over 9 years. If it takes 3 hours to write a blog, this is 735 hrs, ie about 20 weeks of effort over these 9 years. Which is a lot! But overall its been an enjoyable experience which hopefully has helped some people, so I am glad I went down this path.