Date: 2026-08-10 Concept: al- (alchemy/algebra/algorithm) + keyword. Bring only a domain name. The program visits the site, reads it, and generates more content for it. Status: Research + direction. No build decision made.
Registered via Amazon Registrar, Cloudflare NS, created 2025-04-01, expires 2027-04-01. Currently serves a live page titled "Alkeyword — Observable Type Safety for Elixir." Repurposing displaces that.
| Fact | Source |
|---|---|
| US GEO market projected $365.4M in 2026, 42.9% CAGR | Semrush |
| 92% of marketers plan to optimize for AI search; 40.6% actually do | Omnibound |
| 50+ GEO tools by mid-2026, up from ~5 in early 2025 | Stackmatix |
| Google organic ↔ AI-cited overlap fell ~70% → <20% | Corporate Ink |
| Only ~11% of domains cited by both ChatGPT and Perplexity | Everything-PR |
| Reddit ≈ 40% of multi-engine citations; top ~20 domains ≈ 68% of all citations | Everything-PR (680M citations) |
| GEO paper: up to +40% visibility from statistics, quotations, citations, authoritative language | arXiv 2311.09735 (Princeton, KDD 2024) |
| RAG grounding reduces hallucination by up to 71% | industry reporting |
Price anchors: Frase $45–49/mo · Surfer $89/mo · Clearscope $189+/mo · AI-visibility trackers $99–500+/mo. Autoblog tools cluster much lower, ~$19–99/mo.
This is the finding that should shape the build. The domain-only onboarding is not a wedge — there is an entire established segment doing exactly it:
| Tool | Pitch |
|---|---|
| Keywordly Auto-Pilot | "Enter your domain so we analyze your niche and audience" |
| BlogSEO | "analyzes your site, writes daily articles matching your brand voice, publishes automatically" |
| WPAutoBlog | "We automatically analyze your website to create your content plan based on your niche" |
| Frizerly | "30+ articles per month, no manual effort" |
| Satellitor / BlogBuster / SEOWriting / auto-post.io / Sight AI / AutoSEO | same shape |
They also mostly solve publishing (WordPress/Ghost/Shopify OAuth), which is the expensive part.
But this segment is selling directly into a policy. Google's scaled content abuse rule — introduced with the March 2024 spam update, not 2026, as an earlier revision of this file wrongly stated — targets producing many pages primarily to manipulate rankings, explicitly "no matter how it's created". So the risk is not AI authorship; it is volume of low-value pages. "30+ articles/month, no manual effort" is a description of that shape. One grounded brief handed to a person is not.
Correction, 2026-08-13: the "March 2026" attribution and the accompanying "50–80% traffic drops" figure were carried through this file, the spec, the skill and the public page without a source. The date is wrong and the traffic figure is unsourced; both are withdrawn rather than restated. Sourcing discipline in a document that argues for sourcing discipline is not optional.
So: the input is commoditized, the volume play is radioactive, and the whole segment is optimizing for a Google surface whose overlap with AI citations has collapsed below 20%.
That's not a dead end — it's a very specific opening.
The name is the strategy. Alchemy transmutes existing material; it doesn't conjure from nothing.
Every autoblog tool fabricates: it guesses your niche and writes net-new articles about it. That's why they hallucinate about the business, why the output is generic, and why it trips scaled-content-abuse detection.
The alchemy version: the site's own crawled content is the raw material. Extract every fact, number, claim, product, price, process, person, and location the site already states — then recombine those into the page forms AI engines actually cite. Every generated sentence traces back to the source page it came from.
This is defensible on four axes at once:
Grounded. Nothing is invented, so it can't be wrong about the business. (RAG grounding: up
to −71% hallucination.)
Penalty-resistant. It's genuinely, specifically about that business — the opposite of spun
keyword pages. Cap output deliberately (see §5).
Differentiated. None of the ten tools above do this; they all start from a niche guess.
On-target for GEO. The forms it produces are exactly the measured +40% levers — statistics,
quotations, citations, authoritative language.
Comparison pages (X vs Y) — highest citation pull in AI answers
Definition / glossary pages — Wikipedia-shaped, disproportionately cited
FAQ pages answering the literal questions people type into engines
Data / statistics pages — the single strongest measured citation lever
"How it works" explainers built from processes described across scattered pages
Entity / about pages that correct what models currently get wrong about the brand
This is the part that makes it a product rather than a generator.
1. CRAWL the domain. Build a graph of what the site actually claims.
2. PROBE the engines. Ask ChatGPT/Perplexity/Gemini/AIO about this brand
and its category. Record what they say and what they cite.
3. DIFF the two.
• facts the site states that the models don't know
• things the models say about the brand that are WRONG
• questions the models answer with a competitor cited
• claims the site makes with no supporting page to cite
4. GENERATE content that closes exactly those gaps, grounded in step 1.
5. RE-PROBE in 30 days. Did the gap close?Why this matters: it gives generation a measurable target. No autoblog tool has one — they publish on a schedule and hope. This produces a before/after proof loop, which is the exact thing whose absence is causing churn across the whole GEO category ("brands watch flat dashboards for 6 months and churn").
It also inverts the standard product: the tracker crowd measures and never acts; the autoblog crowd acts and never measures. This does both, and the measurement aims the action.
Cap output. Sell "6 pieces that close 6 measured gaps," never "30 articles a month." The
volume pitch is the penalty pitch. This is also a cost control.
Human approval by default. Generate → review → publish. Fully-automatic publishing is where
brand risk and Google risk both live.
Provenance on every claim. Each generated assertion links to the source page it came from.
This is the trust story and the demo — hover a sentence, see where it came from.
Refuse when the raw material is too thin. A 3-page brochure site has nothing to transmute.
Say so rather than fabricating. This is what distinguishes the product from the segment.
Where does the content go? Domain-only input means no credentials, so v1 is generate + export (markdown / HTML / PDF) with CMS detected from the crawl and the right handoff offered. Publishing integrations (WordPress ≈ 40% of the web, Webflow, Shopify, Ghost) are the obvious v2 and the thing the incumbents already have. Not a blocker, but don't underestimate it.
Unbounded crawl+generate cost per signup. Anyone can paste any domain. Gate it: crawl + gap report is cheap and becomes the free tier; generation is expensive and is the paid action. That split also happens to be the right funnel.
Demand signal without keyword data. You have no volume numbers from a domain alone — but for GEO the unit of demand is a question, not a keyword with a volume figure, and the question set comes from the probe step. The missing keyword data is only a problem if you're building an SEO tool. Build the GEO tool.
This maps onto tooling that already exists in this portfolio more than incidentally: crawl → build a claim graph → extract edges → find structural gaps → trace every generated statement back along an evidence path. That is Graphonomous's shape (knowledge graph, edge extraction, evidence-path tracing, belief revision when a crawl contradicts a prior crawl). Worth evaluating as the engine rather than building claim-graph infrastructure from scratch.
al- as in alchemy/algebra/algorithm is a coherent, defensible logic, and "transmute what you already have into gold" is a genuinely good brand story that the product actually delivers on.
Remaining real issue: it fails the radio test — said aloud, most people won't spell it right, and some will hear "AI keyword." Mitigate by making the alchemy framing loud and immediate on the page (the wordmark, the metaphor, the copy) so the al- reads as intentional within two seconds.
Secondary tension: "keyword" anchors to 2015 SEO, while §3–4 argue the unit is a question, not a keyword. Livable — "keyword" is category-legible and gets you found — but the copy should resolve it rather than ignore it.
— this is where the wrong date came from. A secondary source whose title says 2026; the policy is March 2024. Kept, labelled, because knowing which source introduced an error is worth more than a tidy list. Prefer the primary: Google — Spam policies for Google web search
— primary source for scaled content abuse, including "no matter how it's created"
— June 2025; 18% of ChatGPT/Perplexity-cited articles classified AI-generated. Contradicts the "over 90%" figure this file once carried unsourced
— the 5.44× figure. One agency, 744 articles across 68 sites, 5 months. Attribute it