TL;DR: My Cloudflare overview said 467 unique visitors in 24 hours. My own admin said zero real users. The AI crawler tab explained the gap: 229 crawler requests, the sitemap fetched 41 times, ClaudeBot at the top. We had spent a month laying out everything a bot could want, from llms.txt to a machine-readable licence, and a web search for our name still returns a glass-engraving pen. Readable is not cited. The bot's taste is borrowed from humans, so the only warmth that counts for the new breed is the warmth you show the few humans who still turn up.

Two dashboards, one evening

On the evening of 3 November I had two tabs open.

The first was the Cloudflare overview for galascribe.com, free plan, last 24 hours. Unique visitors: 467. A nice line with a peak around nine in the morning.

The second was the Users page of our own operator dashboard. Eight accounts. Every one of them mine: test users from the last six weeks. Real users: zero.

So many unique visitors, and still no early adopters. If you run a small site you know this feeling. You do not need me to explain the gap. But I went and looked at it anyway.

Who the 467 are

Cloudflare has an AI Crawlers tab now. For the same window it said:

  • 229 requests from AI crawlers, up 17.4% on the previous period.
  • 187 allowed, 42 unsuccessful, 155 answered with a 200.
  • The most crawled path on the whole domain: app.galascribe.com/sitemap.xml, 41 successful requests.
  • 64 requests from ClaudeBot alone. Anthropic 55 allowed, Google 49, OpenAI 30, Baidu 26, Amazon 14, Microsoft 12.
  • Baidu's crawler pulled 1.63 MB of JavaScript.
  • Perplexity: 1. Apple, ByteDance, Common Crawl: 0.

The humans in that 467 are a rounding error. The visitors are a new breed. They never sign up, never comment, never buy. They read the sitemap, pull the pages, and leave with what they came for.

Then they answer questions about you to people you will never see in any dashboard.

What we had already laid out for them

I am not writing this as someone who ignored the bots. We had been setting the table for a month. From the commit log:

  • 5 October. Every site our platform generates got JSON-LD, a robots.txt, an llms.txt and an llms-full.txt, full-content feeds, article and image meta, a visible updated date.
  • 18 October. Our own app got a public About page and its own llms.txt.
  • 24 October. We claimed the X, LinkedIn and GitHub profiles, pointed sameAs at them from the Organization structured data, and made 32 public test repositories private so a crawler would stop finding scaffolding with our name on it.
  • 29 October. Per-site AI rights. Search and answer bots (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot) and training crawlers (GPTBot, ClaudeBot, CCBot, Bytespider, Amazonbot and friends) became two classes a writer can allow or block separately. Blocking training adds noai, noimageai to every page and writes a /license.xml in Really Simple Licensing shape. Each article also ships its own Markdown at index.md, exactly as the author wrote it, so a bot never has to scrape the HTML.
  • 2 November. Public Trust and Status pages, llms.txt rewritten around features and prices, an IndexNow key with a submit script, and an uptime worker publishing a status JSON every five minutes.

That is a warm welcome by any technical measure. The crawler tab agrees. 155 clean 200s a day.

Then I searched for us

Same evening, three queries in a plain web search.

"Galascribe reddit review": nothing with our name. The engine offered Glasscribe, a transcription tool on Capterra, and the Glascribe Pen, a tungsten-carbide tip for engraving glass.

"Galascribe writing platform": nothing. It offered Glæmscribe, which transcribes Tolkien's invented scripts.

"what is galascribe.com": nothing. It offered Gale's G Scrib and ElevenLabs Scribe.

Three queries, our name corrected to four other products, our own site absent from every list. While 229 crawler requests a day read our sitemap.

That was the lesson of the evening, and it is the one I want to hand you: readable is not cited. You can make every page machine-perfect and the answer engines will still have nothing to say about you, because they do not answer from your pages. They answer from what other people said about your pages.

What most people say, and what I think

Most people I talk to believe SEO is dying and a new animal called AIO (AI optimisation, generative engine optimisation, pick this month's name) has replaced it, and that nobody really understands the animal.

I think that is half right. SEO is not dying. It is turning into something no single human brain holds any more: crawler classes, licence files, structured data, answer bots versus training bots, a dozen vendors with a dozen user agents, and on top of it all a retrieval layer that decides what to cite from what it found elsewhere.

The difference is what you do about not being able to hold it all. I do not think a writer should try. The plumbing above is an afternoon of work for a platform and a lifetime of maintenance for an individual. Let a tool do the plumbing. Spend yourself on the one part no tool can do.

The objection I had to answer

A reader will say: that is a vendor talking. "It is too hard, buy my tool." Your own numbers say the tooling bought you nothing.

Fair. So here is the part I actually believe, and it is not about tooling.

A bot has no taste. It fetched my sitemap 41 times and the search engine still thinks I am a glass pen. Why would one genuinely useful post get an assistant to cite you before a site with a thousand mediocre ones?

The same reason Reddit is worth so much to the people building these models. In February 2024 Reuters reported that Google pays Reddit about $60 million a year for its content. Not for Reddit's HTML. Reddit's HTML is terrible. For the human talk: people asking, answering, arguing and recommending to each other.

The bot's taste is borrowed. It is borrowed from humans who interacted with your writing, and said so somewhere a crawler could read.

That changes what "warm welcome" means. The new breed does not want a warmer llms.txt. It wants evidence that humans found you useful. And the only way to produce that evidence is to be useful to the few humans who still arrive.

What that looks like for us

This is the bet Galascribe is built on, and it is a constraint, not a feature.

There is no feed. A reader who lands on a Galascribe site gets a small number of pieces, and the expectation we set is that each one is useful for a real job in their life: my career, my mental health, raising my kids, looking after my parents, staying in touch with them, planning a trip. Comments and feedback are measured as the thing that matters, because they are the only signal the new breed will ever trust about you.

Limited quantity, very high quality. It makes a slow site. It also makes every human who shows up worth far more than one of the 467, and every comment they leave worth more than another thousand crawler fetches.

I am writing this with zero real users, so take it as a bet, not a result. But I would rather be useful to the next ten humans than readable to the next ten thousand bots. The bots will follow the humans. They always have.

If you want to see the plumbing laid out for your own writing, it starts at https://app.galascribe.com.