What Google — and now AI — actually reads on your website
Every day, machines scan your site and decide whether to show your business, quote it, or skip it. Most owners have never seen the layer they read. This note shows the layer — dated, sourced, and honest about what it cannot do.
The machine layer is real. Here it is in plain words.
Google does not read your website like a person. ChatGPT, Claude and Perplexity do not either. They read a machine layer: your title and description, small structured-data blocks (schema), your crawler policy file (robots.txt), and the plain text visible without scripts.
Most local business sites have large, fixable gaps in this layer. Closing them makes your business legible: machines can confirm who you are, what you sell, where you are, and whether they are allowed to read you at all.
- Title & description: the first sentence machines read.
- Schema blocks: machine-readable facts about your business.
- robots.txt policy: which crawlers you allow — and which you accidentally turned away.
- No-script text: what a machine sees when it never runs JavaScript.
- Q&A coverage: the buyer questions your pages answer in extractable form.
A machine first has to be able to read you. Only then can anyone choose you.
What is provably true this month — from our own policy radar
We watch 65 first-party sources (platform robot policies, IETF standards, search-engine documentation) hourly and timestamp every change with public SHA-256 hashes. Four items matter for owners right now:
robots.txt stopped telling the truth
Cloudflare retired its AI-bot blocks in favour of an “AI Preferences” model. Independent crawl measurement days later found ~80% of studied sites still block GPTBot and ClaudeBot at the edge — while their public robots.txt no longer says so. (Cloudflare platform change 15 Sep · dev.to measurement 23 Sep 2026)
AI content preferences are being standardized
The IETF is writing the vocabulary machines will respect (Content-Usage draft v05, 19 Aug; AI Preferences vocabulary v08, 13 Sep). Early sites get to speak it first. (IETF drafts, watched live by our radar)
Google shipped the September 2026 spam update
Its fourth confirmed spam action this year hunts exactly the thin, templated, machine-stuffed pages lazy “AI content” produces. Honest machine-layer work is the opposite of that. (Google Search Central · Search Engine Roundtable 25 Sep 2026)
86% of AI “best product” answers cite a commercial source
When people ask an AI for a good local provider, it mostly quotes businesses’ own web material — but only material it can read and verify. (GetCited analysis via PPC Land, 27 Sep 2026)
The full ledger is public: synapsein.com/policy-radar.php — every hash verifiable by anyone.
What this work will NOT do
It will not guarantee rankings
Nobody controls Google’s order. You can own whether your facts are declared for and readable by machines; the rest is probability. Anyone promising placement is guessing with your money.
It will not “put you in ChatGPT” overnight
AI systems update on their own schedule. Fixing the machine layer makes you eligible to be read and quoted; it does not schedule anyone’s model.
Check your own site — about 90 seconds
Where do I see proof this is real?
The proof page runs five checks on us in public, including our own monthly numbers with the bad months intact. Ninety seconds, no form.
I got a machine-audit email from you. What was it?
A one-time gift of notes, built from a live crawl of your site. No mailing list attached — reply “fix it” and we state plainly what we would do and what it costs, or hand the notes to any developer you trust.
Can SynapseIN guarantee rankings or AI citations?
No. The work improves clarity, usefulness, technical trust and citation readiness. We promise work, not outcomes.
Useful next steps
What is AEO? · What is GEO? · Policy radar ledger · Run the free audit · Contact