Nobody reads your blog. ChatGPT might.

Two years ago I started a blog and named it Nobody Read This, because I am a data scientist and I respect base rates. The name has held up. I published a couple of essays, told almost no one, and the analytics aged into a fair description of the project.

I still think the name was correct. What I got wrong was who "nobody" would turn out to be.

Here is the short version, since you probably asked a chatbot a question to get here. AI assistants recommend newsletters and blogs the same way a well-read friend does: they surface the source that answered a specific question cleanly, more than once, on a subject it clearly owns. To become that source you need four things. Write about one subject, answer real questions directly instead of circling them, keep your site technically readable by crawlers, and do it long enough that your name and your niche start appearing in the same sentence. There is no hack. There is mostly plumbing and consistency, which is good news, because plumbing and consistency are free.

The rest of this post is the honest version of each step, plus the parts I think are overhyped.

Why would ChatGPT recommend anyone's newsletter?

When someone asks an assistant "what should I read to understand board governance" or "who writes well about attention and technology", the assistant does not consult a mystical ranking of good writers. It searches, retrieves a handful of pages, and composes an answer from whatever it retrieved. If your essay is one of those pages, you get named and linked. If it isn't, you don't exist.

That retrieval step is the whole game. It favors pages that:

• answer a specific question, stated plainly, near the top
• contain passages that survive being quoted out of context
• come from a site that keeps answering questions in the same territory

Notice what's missing from that list: follower counts, posting streaks, whether you went viral in March. Retrieval does not know about your engagement rate. This is the most level playing field distribution has offered small writers in about fifteen years, and it quietly rewards the people who were writing substantive posts all along.

The industry named this GEO, generative engine optimization, because the industry names everything. Strip the acronym and it's this: write so a machine skimming ten pages can tell, in five seconds, that yours is the one that answers the question.

What actually makes a post quotable?

I've spent a decade doing data science at software companies, which mostly means explaining to people what the model actually rewards. Let me translate retrieval into writing advice.

One post, one question. Assistants retrieve chunks, not vibes. A post called "Some thoughts on growing an audience" is retrievable for nothing. A post that answers "how many subscribers do you need to make money from a newsletter" is retrievable for exactly that, forever. Before you write, say the question out loud the way a real person would ask it. That question is your assignment. Everything else is a different post.

Answer in the first hundred words. You can keep your story, your jokes, your dog (I certainly keep mine). But the direct answer has to appear early, phrased so it stands alone. If a stranger read only that paragraph, would they have the answer? That paragraph is what gets quoted.

Be the specific one. "Consistency matters" is filler. "I posted every other Tuesday for a year and 60 percent of my subscribers came from three posts" is a sentence a machine quotes and a human remembers. Numbers, prices, timelines, named tools. If a section of your draft contains no concrete fact, that section is decoration.

Headers are questions. Your H2s should read like the follow-up questions a curious reader would ask, because that is literally how retrieval queries arrive. Look at the headers in this post. I wrote them that way on purpose, and the format is working on you right now.

The plumbing: an afternoon of unglamorous work

None of this matters if crawlers can't read your site. The checklist is short and boring:

1. Don't block the bots you want. Check your robots.txt. OpenAI, Anthropic, Perplexity, and Google all publish their crawler names. Some platforms and CDNs block AI crawlers by default now, which is a reasonable stance for the New York Times and a self-inflicted wound for a newsletter with 300 subscribers trying to be discovered.
2. Have an RSS feed and a sitemap. Substack gives you both. If you're on your own site, generate them. They are how anything finds anything.
3. Structured data. Article or BlogPosting markup from schema.org tells crawlers your page's title, author, and date without making them guess. Most platforms emit it automatically; check yours does.
4. llms.txt if you like. It's a proposed convention: a plain-text index of your site written for language models. It takes ten minutes. I'd file it under "cheap, plausibly useful, unproven", which is also where I file most vitamins.
5. Own your domain if you can. Citations accrue to a domain. Ten years of essays on a domain you own is an asset; the same essays on a platform subdomain are a lease.

That's it. You do this once, then go back to writing.

Is GEO the new SEO snake oil?

Partially, yes. Wherever attention moves, a consulting industry materializes about four days later, and there are already agencies selling "AI visibility audits" for four figures. You do not need one. The measurement tools are immature, nobody outside the labs knows the retrieval weights, and anyone promising you a number one spot in ChatGPT is selling astrology with a dashboard.

Here's what I actually believe. The tactics above are worth doing because they are indistinguishable from good writing and basic web hygiene. Answering a clear question clearly was the right move in 2010 and it's the right move now. If GEO turns out to be half hype, you've lost nothing; you've just written a better blog. That asymmetry is the entire reason to care.

The part I'd bet on: the share of reading decisions that route through an assistant keeps going up. People already ask chatbots what to read, what to subscribe to, whose take to trust. Someone will be the answer to those questions in your niche. The only question is whether it's you or the person who kept publishing.

This is, full disclosure, the problem I'm building Pemberley around: it reads your newsletter or blog, learns your niche, and briefs you on the conversation your writing sits in, so being the citable one takes less than a decade. But the strategy above requires no tools at all. Just a question, an answer, and a schedule.

Questions people actually ask

Should I block AI crawlers to protect my work instead?

If you're a large publisher with licensing leverage, maybe. If you're an indie writer whose bottleneck is that nobody knows you exist, blocking the crawlers is optimizing for a problem you'd be lucky to have. You can always change your mind later; robots.txt is not a tattoo.

Does any of this help if I only have 12 subscribers?

It helps you most of all. Retrieval doesn't check your subscriber count before quoting you. A precise answer on a neglected question can get a nobody cited alongside names with a hundred times the audience. Small writers have never had a channel this indifferent to their smallness.

How long does it take to see anything?

Think months, and know that measurement is genuinely hard right now. Ask a few assistants your target questions every month or so and watch whether your name starts showing up. This compounds the way a savings account does. Checking it every hour will only make you miserable.

Can I just have AI write the posts?

You can have help with drafting, structure, and the plumbing. What you cannot outsource is having something specific to say, because specificity is the entire mechanism. A machine-generated post with nothing in it is retrievable for nothing, same as a human-written one.

The joke embedded in my blog's name turned out to have a second act. For fifteen years "nobody" meant the algorithm passed you over in favor of whoever was louder. Now some of the most influential readers of your work were never going to like, share, or subscribe, because they are not people. They just read carefully, remember who answered the question, and pass your name along.

Write for the reader who tells other people what to read. It was always good advice. It's just that now, some of those readers are machines.