Methodology
We eat our own dogfood: how Hidden Layer is built to be cited
A research blog that tells you how to be discovered by AI should be discoverable by AI. Here is exactly how we apply our own findings to this site — and how to verify every claim yourself.
On this page
We publish a lot of advice about how to be discovered, read, and cited by AI systems. The obvious test of that advice is whether we follow it ourselves. We do. This post is the audit of our own site against the same standards we hold everyone else to — and, because we are a fact-checked publication, every claim below is one you can verify in your own terminal in under a minute.
We render on the server, so agents see the whole page
The single most common reason a page is invisible to AI is the render gap: the content exists only after JavaScript runs, and most AI crawlers do not run JavaScript. 3 Hidden Layer is a static export. Every article ships as complete HTML on the first request — the same bytes a curl command receives are the bytes a model receives. There is no client-side hydration standing between our research and an agent reading it.
You can confirm there is no render gap directly: fetch a post with no browser and check that the prose is already present in the raw HTML.
curl -s https://hidden-layer-blogs.pages.dev/post/what-actually-moves-ai-citations \
| grep -o "What actually moves AI citations"
# the headline and body are in the raw HTML — no JS execution requiredWe allow the crawlers we tell you to allow
Our robots.txt explicitly permits the AI crawlers that drive citations — GPTBot, ClaudeBot and Claude-SearchBot, PerplexityBot, Google-Extended, OAI-SearchBot, and Applebot-Extended — and blocks the ones that take content without giving discovery back, including CCBot, Bytespider, and AhrefsBot. We also declare a Content-Signals line (allow-training, allow-search) so our intent is machine-readable rather than implied.
This is the posture we recommend for any publisher that wants to be in AI answers: be selectively open, not closed. Check it yourself.
curl -s https://hidden-layer-blogs.pages.dev/robots.txtWe ship the AI-readable surfaces in full
Beyond robots.txt, we expose the discovery surfaces an AI-era site should have: an llms.txt summary, a complete llms-full.txt containing the plain-markdown text of every article we have published, an XML sitemap with last-modified dates, an RSS feed with full content, and the agent-facing endpoints — /.well-known/mcp.json, /.well-known/agent-card.json, and an OpenAPI spec for our audit API.
The llms-full.txt is the one most sites skip. It is a single file a model can ingest to get our entire corpus without crawling page by page. We generate it at build from the same source as the site, so it never drifts.
- https://hidden-layer-blogs.pages.dev/llms.txt — the summary and index
- https://hidden-layer-blogs.pages.dev/llms-full.txt — every article, in full, as markdown
- https://hidden-layer-blogs.pages.dev/sitemap.xml — all URLs with lastmod
- https://hidden-layer-blogs.pages.dev/feed.xml — full-content RSS
We use schema as eligibility plumbing — honestly
We have published, with receipts, that Schema.org markup produces roughly zero direct lift in AI citations: large-language models tokenize JSON-LD as raw text rather than parsing it as a semantic layer. 2 We emit it anyway — every post carries a schema.org Article block with author, publisher, dates, and word count — because schema remains a hard prerequisite for Shopping-feed eligibility, Knowledge-Graph indexing, and Perplexity. Calling schema "infrastructure, not a citation lever" and then shipping it is not a contradiction. It is the exact distinction we ask you to make.
We go one step further than most: our Article JSON-LD includes a machine-readable citation array, so the sources behind a post are resolvable as structured data, not just visible as footnotes. Our receipts are part of the markup.
A site that explains how to be cited by AI should be a site AI can cite. We do not publish a standard we have not met ourselves.
We write the way the evidence says to write
The strongest causal levers on AI citation are about content substance: explicitly citing sources, adding statistics, using direct quotations, and writing in fluent, well-formed prose. 1 Every flagship on this site follows those rules by construction. Statistics carry inline citations. Claims trace to named sources. We do not pad with rhetorical questions or hype, because over-optimized and keyword-stuffed text measurably performs worse.
And the differentiator no competitor has: every statistic we publish is decomposed into atomic facts and run through an adversarial truth engine before it ships, with a per-claim verdict, confidence, and corroborating sources attached. The "receipts" you see on our living-knowledge posts are not decoration — they are the audit trail.
Hold us to our own standard. If you find a gap between what we recommend and what we ship, that is a bug, and we want to fix it.
Footnotes3