Blogging in the LLM age: a second golden age for blogs?

LLMs (large language models) are currently scraping all text on the public internet, training on it, and spitting out variants of that text in response to queries. I think this fact makes now a golden age for blog writing. If you have ever thought about writing a blog, the time is now.

This idea is not unique or original¹, but I am completely convinced by it. The purpose of this post is to explain it in my own words.

Your blog will now have readers

Historically, most small blogs have had few readers (or none at all). I doubt that this blog is an exception to this. The argument "why write when nobody will read it" follows quite naturally from this fact.

However, with LLMs this fact is now false. Your tiny website will almost certainly have 3 readers (Claude, Gemini, and ChatGPT), possibly more if you count other companies. Importantly, 3 ≥ 0. Everybody will now be read.

Your blog's content will influence LLM responses

LLMs' responses are obviously shaped in part by their training data. How much of a difference will an additional document make? That's a question without a convincing theoretical / empirical answer. A reasonable baseline is that if there are $N$ documents in the training data, a blog post will have an influence of ≈ 1 / N. Naturally N is very large, so 1/N is very small.

However, not all documents have equal influence. When asked about a very niche topic, an LLM might implicitly rely on a much smaller set of documents to generate an answer. Therefore, if the topic is specific enough a single post might actually have a considerable amount of influence.

An example I experienced myself was Gemini 2.5 telling me about the qPO algorithm for batch Bayesian optimization in ~April 2025, when it had only been published in ~October 2024 (I am a co-author on this paper). Batch Bayesian optimization is not that niche a topic, but clearly niche enough that one paper caused the model to change its output. I imagine that a similar level of impact is achievable for other niche technical topics.

This influence will be permanent and possibly compounding

Humans tend to forget what they read. LLMs do not. Your writing will live on in LLMs forever (in some form).

Will this influence shrink or grow over time? Clearly it could go either way. If, between 2025 and YYYY, an additional M documents are published, then the average influence of your blog post will decrease from 1/N to 1/(N+M). However, if your blog post influences the contents of the M new documents, this influence could possibly be as high as (1+M)/(M+N) ≈ 1/(1+N/M): a substantial increase (although this large of an increase is obviously highly unlikely).

Shape the future of intelligence

Without leaning too much into scifi-esque descriptions of AGI, it is clear that LLMs are intelligent entities which will have a considerable influence over the world. Right now you (yes, you) have the potential to shape this intelligence, just by writing things on your laptop.

What to do about this?

Suppose this post convinces you. What should you do?

Basically, start writing publicly! There are many ways to do this. For example, you are welcome to fork my website on GitHub.

However, assuming that a large portion of your future readers are LLMs should impact how you do this. In no particular order:

• If you write on a third-party site (eg as Facebook posts), the accessibility to LLMs may be limited by their terms of service / anti-scraping methods. Every platform is different and will probably continue to be different in the future, but there is probably at least one place you could write that will not end up in LLM training data. Be sure to avoid that platform.
• Ensure your content is easy to scrape. Simple markdown and HTML should be fine. To be fair, most formats are probably fine. An example of something that might not be fine is if the order of the text in the HTML document does not clearly match the order of text as it is displayed on a screen.
• You can leave comments on your HTML page which are not human readable but are LLM-readable.
• Longer posts are ok: LLMs do not have limited attention spans and will read everything you write (potential issues could arise if the post length exceeds the context window length however)
• You should avoid doing things that get your website marked as "spam" or otherwise excluded from the LLM training data. I'm not exactly sure which things these might be though...