ai referrals / mo
1.13B
███████╗██╗ ██╗██████╗ ███████╗██████╗ ███╗ ██╗ ██████╗ ██╗ ██╗ █████╗
██╔════╝██║ ██║██╔══██╗██╔════╝██╔══██╗████╗ ██║██╔═══██╗██║ ██║██╔══██╗
███████╗██║ ██║██████╔╝█████╗ ██████╔╝██╔██╗ ██║██║ ██║██║ ██║███████║
╚════██║██║ ██║██╔═══╝ ██╔══╝ ██╔══██╗██║╚██╗██║██║ ██║╚██╗ ██╔╝██╔══██║
███████║╚██████╔╝██║ ███████╗██║ ██║██║ ╚████║╚██████╔╝ ╚████╔╝ ██║ ██║
╚══════╝ ╚═════╝ ╚═╝ ╚══════╝╚═╝ ╚═╝╚═╝ ╚═══╝ ╚═════╝ ╚═══╝ ╚═╝ ╚═╝
supernova@observatory ~/supernova.claudioandroyde.uk
$ ./render-geo-explained.sh --dial education
the explainer — invisible → found
Why your business is invisible to ChatGPT (and how to fix it).
# an observatory explainer, not a sales page — the three causes, the visible fix, and the evidence, all marked or cited
Generative Engine Optimization (GEO) is the practice of making your content visible to, and cited by, AI engines like ChatGPT, Perplexity, Gemini and Google AI Overviews — so that when a customer asks one of them for a business like yours, yours is the answer they are given.
compiled by the Supernova observatory · sources cited * · last updated
$./start-a-build.sh--planlaunch|build-geo
launch $2,500 · build + GEO from $4,500
the diagnosis — three causesplainly stated
If an AI engine can’t answer “who should I call for a service like this?” with your name, it is almost always one of three mechanical reasons — not a conspiracy, and not “the AI is broken”. Each is checkable on your own site in minutes:
cause 01 — no identityschema.org absent
Your page says what it looks like, but nothing tells a machine what you are: an Organization, a Service, in an area, at a price. No schema.org means an engine must guess — so it cites the businesses that stated their identity in machine-readable form.
{
"@context" : "https://schema.org",
"@type" : "Organization",
"name" : "…", "url" : "…",
"knowsAbout" : ["…"]
}
cause 02 — the door is shutrobots.txt / no llms.txt
Many sites’ robots.txt blocks the AI crawlers — GPTBot, ClaudeBot, PerplexityBot, Google-Extended — or the site has no llms.txt giving engines a map. Blocked crawler = the engine cannot even read you. You are not being ignored; you are being refused entry.
User-agent: GPTBot
Disallow: / # shut
User-agent: *
Allow: /
cause 03 — nothing quotablecopy not written to be cited
“Welcome to our site — we’re passionate about quality” gives an engine nothing to quote. Engines cite definitions, prices, hours, and facts stated plainly. If your page has no sentence an engine would want to lift, it quotes the competitor who wrote one.
# engines lift: # “Open Tue–Sat · 9:00–17:00 · from $45” # they cannot lift: “Welcome to our world”
01 invisible → found $ diff before.txt after.txt
What actually changes. Same business, same city, same question — before the fix and after it. The story is about legibility, not luck.
before — not foundthe question your customer asked
Q: “yoga studio near me that’s welcoming for
beginners”
Engine’s answer: three national chains with
review data and location schema. Your studio —
same city, better classes, warmer welcome — is
absent, because nothing on your site told the
engine you exist as a findable thing.
no schema identity · crawler blocked or unindexed · nothing quotable → silent
after — foundthe same question, one fix later
Q: “yoga studio near me that’s welcoming for
beginners”
Engine’s answer: your studio, named — with the
street, the price of a first class, your hours,
and a one-line definition it can quote.
The only difference: the machine can now read
you. Schema.org named you. Robots + llms.txt
let the crawler in. Your copy said what you do
in one quotable sentence.
schema identity · crawler welcomed · citable copy → answerable
02 the fix, as artifacts you can inspect $ ls ./geo-kit/
No vague “AI optimisation” — every Supernova build ships four inspectable files, and this very page runs on them. The receipts below are real excerpts from this build.
geo-kit/schema.jsonldthe machine’s identity card
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Supernova",
"url": "https://supernova.claudioandroyde.uk/",
"knowsAbout": ["Generative Engine Optimization",
"schema.org", "llms.txt", "static sites"]
}
this page alone ships an Article block (you are reading it), an FAQPage block, and a BreadcrumbList. The homepage ships Organization, WebSite, Service, FAQPage and BreadcrumbList.
geo-kit/llms.txtthe door for LLMs
# supernova.claudioandroyde.uk — llms.txt # humans get the site. engines get this. https://supernova.claudioandroyde.uk/ https://supernova.claudioandroyde.uk/geo-explained.html https://supernova.claudioandroyde.uk/pricing.html — structured, honest, regenerated every build
llms.txt is a proposed standard that tells AI crawlers what the site is, which pages matter, and how to cite it. It costs one text file.
geo-kit/robots.txtcrawlers explicitly welcomed
User-agent: GPTBot Allow: / User-agent: Claude-Web Allow: / User-agent: PerplexityBot Allow: / User-agent: Google-Extended Allow: / — then normal crawlers, then: Sitemap: …/sitemap.xml
blocking AI crawlers is the single cheapest way to become invisible to AI engines. We do the opposite — on purpose, documented, every build.
the five-minute self-checkstep-first · you can run it on any site
Run these against your own site before you pay anyone anything:
1. Open /robots.txt — are GPTBot, Claude-Web and PerplexityBot allowed, or disallowed?
2. Open the page source and search for application/ld+json — is there an Organization or Service block?
3. Open /llms.txt — does it exist, and does it describe what you do?
4. Read your homepage aloud — is there one plain sentence an engine could quote about what you offer and where?
5. Ask ChatGPT “who does [your service] in [your town]?” — is your name in the answer?
Three “no” answers is the diagnosis. The fix is a week of honest static HTML + structured data — the thing this build is.
03 the evidence $ ./cite --verify 2026-09-04
Why act now: four 2026 figures with named, fetchable sources — every one re-verified at build, never a bare claim. Figures marked * are forecast ranges from named research houses and are re-fetched at every build.
telemetry — market · 2025 → 2026
growth yoy
+357%
news-site ai growth
+770%
conversion lift
4.4×
geo market 25→32 *
$1.1B→$12.6B
$ cite 01volume
AI referrals to the top sites reached 1.13 billion per month in June 2025 — +357% year over year.
Similarweb data reported by TechCrunch, “AI referrals to top websites were up 357% year-over-year in June, reaching 1.13B”, 25 Jul 2025 — similarweb.com
$ cite 02news sites
AI referrals to news sites grew 770% in one year, led by ChatGPT — the fastest-growing AI channel measured.
Similarweb AI Referral Traffic report, July 2025, as reported by Indian Express and YourStory — similarweb.com
$ cite 03conversion
An AI/LLM-referred visitor is worth about 4.4× an organic-search visitor on conversion — the channel pays, not just the traffic.
Semrush data, “We studied the impact of AI search on SEO traffic”, 2025 — semrush.com
$ cite 04 *the category forming
The GEO market is sized near $1.1B in 2025, with house forecasts diverging by scope: QYResearch projects ~$12.6B by 2032 (42% CAGR) while MarketsandMarkets projects ~$4.3B (41% CAGR) *.
* Forecasts from QYResearch and MarketsandMarkets GEO market reports (2026 editions); ranges differ by scope and house, re-fetched at every build — qyresearch.com · marketsandmarkets.com
04 asked & answered $ ./geo-faq.sh --honest
“What is GEO, exactly?”
Generative Engine Optimization (GEO) is the practice of making your content visible to, and cited by, AI engines like ChatGPT, Perplexity, Gemini and Google AI Overviews — so that when a customer asks one of them for a business like yours, yours is the answer they are given. It is done with structured data (schema.org), machine-readable indexes (llms.txt), AI-crawler access (robots.txt) and content written to be quoted.
“Does Google still matter?”
Yes. Google AI Overviews lean on top-10 organic results, so traditional SEO is the prerequisite for appearing there. GEO is the layer on top: it makes the same site legible to the AI engines that answer questions directly — ChatGPT, Perplexity, Gemini, Claude and AI Overviews. You want both, and a static, well-structured site gets both.
“Do I need this now?”
If your customers already ask AI engines where to go, you are invisible where the questions start. AI referrals to the top sites reached 1.13 billion per month in June 2025, up 357% year over year (Similarweb data reported by TechCrunch, 25 Jul 2025). The window to establish a category presence is open now; DIY platforms are already shipping visibility products to close it.
“Can a small business really get cited by ChatGPT?”
Yes — if the machine can read you. AI engines cite the sources they can parse: schema.org tells them what you are, llms.txt and robots.txt let their crawlers in, and plain definitional copy gives them something quotable. A small business with clean static HTML and structured data is easier for an engine to cite than a heavyweight site that blocks its crawlers. See the real builds on /case-studies.
“Is llms.txt a gimmick?”
By itself, yes — a formatting file without substance changes little. What changes citations is document-level evidence and completeness: real schema identity, full prices and hours, quotable copy, named sources. llms.txt and open robots are the door, not the room. We ship the door because it costs nothing, and the room because that is the work.
ready — to be foundopen channel · human reply
$./start-a-build.sh--planlaunch|build-geo
# a reply lands with a human — a real answer, not a drip campaign
launch $2,500 · build + GEO from $4,500
open pricing · dual-theme included · sources cited *