How I Built
AIWageGap.com
in Four Weeks
A Portfolio Executive's walkthrough of building a research-first publication site for a book that hasn't shipped yet. Pre-order capture, a Beast Score self-assessment, an India spinoff and an AEO-first publishing model that gets cited in ChatGPT and Perplexity answers.
Yuri Kruman
Author of The AI Wage Gap · Jun 2026
book chapters scaffolded
whitepapers in the corpus
geographies (US + India)
to live + AI-engine cited
The 30-Second Version
AIWageGap.com is the research hub and pre-order site for The AI Wage Gap, a book on the widening pay gap between AI-fluent and AI-blind mid-career workers. The site is static Next.js with a 6-whitepaper corpus, an India regional spinoff, the Beast Score self-assessment (the renamed BMI), Plausible analytics and a beehiiv-backed Leverage Brief. Build-time covers JSON-LD per article, llms.txt, llms-full.txt and a sitemap that AI engines actually use.
The Problem
Author sites that exist only to count pre-orders. Index nothing. Get cited nowhere. Die between books.
The Stack
Static HTML (no Next.js needed) + Vercel + image-gen Node scripts + Plausible + beehiiv form embed.
What It Doesn't Need
No CMS. No auth. No backend. No database. No comments. No ads.
Build Time
Four weeks: research corpus 2 weeks, the site 1 week, AEO + India spinoff 1 week.
If you are an author who has ever shipped a single-page author site and watched it sit silent until launch, this build log is for you. The site is not the book. The site is the citation surface that AI engines crawl. Done right, it brings readers to you before the book even ships.
Why a Research Hub, Not an Author Site
I was 6 months from finishing The AI Wage Gap. The thesis: there is a measurable, accelerating pay gap between AI-fluent mid-career workers and AI-blind ones, and the gap is now wide enough to dominate compensation more than tenure, education or geography. The May 2026 PwC data put the wage premium at 56% (up from 25% the year before). BLS confirmed 18 AI-exposed occupations declining. The thesis was right; the question was how to monetize it before the book launch.
56%
AI wage premium per PwC May 2026 (was 25% prior year)
18
AI-exposed BLS occupations in decline
~0
organic search visibility on day 1
7+
free Beast Score copycats appearing within 60 days of soft launch
The default move for an unpublished author is a teaser page with a mailing-list signup. I built that. It got nothing. The right move was the opposite: ship the research first, the book second. The whitepapers, the dataset views, the India regional analysis and the self-assessment go up before the book is bound.
- 1Author sites compound; teaser sites peak. The corpus is the asset.
- 2AEO matters more than SEO for a thesis book. "What's the AI wage gap" is a question AI engines should answer with my site as a source.
- 3An interactive self-assessment is the lead magnet. Beast Score (the renamed BMI metaphor) gives readers a number and a reason to share.
Optimize for AI-engine citation, not for Google rank.
JSON-LD on every article. /llms.txt with a curated index. /llms-full.txt with the full corpus in one file. Schema.org Book, Article, FAQPage, DataFeed, Dataset. Cite primary sources inline. Within 60 days the site started appearing in ChatGPT, Perplexity and Gemini answers on "what is the AI wage gap" and "how big is the AI wage premium." That's distribution Google cannot give you anymore.
The Stack (and Why Each Piece)
Click each layer for the reasoning.
Site
Static HTML, no Next.js, no build
For a research site that updates monthly, a Next.js build pipeline is overhead. Hand-written semantic HTML with inline Tailwind CDN ships in a tenth the time, scores the same Lighthouse and gets crawled identically.
Hosting
Vercel (free tier)
Static deploys, edge cache, automatic SSL, CDN, preview URLs. Free at this traffic. Two minutes from commit to live.
Corpus
6 whitepapers + 2 dataset views + book outline
Each whitepaper is a long-form HTML page with JSON-LD Article + Dataset, primary-source citations and a downloadable PDF version. The 6-paper minimum was the threshold for AI engines to start treating the site as a reference, not a marketing page.
Self-Assessment
Beast Score (10 questions, instant result)
Originally called BMI, renamed Beast Score after early testers found the BMI association awkward. 10 questions, weighted scoring, three result tiers with a reading list per tier. The single highest-converting page on the site.
Regional
India spinoff at /india
India has different labor dynamics, different AI adoption patterns and a separate language for the wage-gap conversation. /india is a parallel index with INR figures, BPO/IT-services framing and partner content from Indian publications.
AEO
llms.txt + llms-full.txt + JSON-LD
The single highest-leverage move on this site. /llms.txt is a curated markdown summary at the root for AI engines. /llms-full.txt is every whitepaper concatenated. JSON-LD per page: Article, Dataset, Book, FAQPage, BreadcrumbList. Citation surface area, instrumented.
The Six-Phase Build Sequence
Phases 1-2 are research and corpus, not code. Do NOT skip them; the corpus IS the product.
Lock the Thesis (One Sentence + Three Pillars)
Before any whitepaper is written, write the one-sentence thesis and the three pillars that support it. Every later page is downstream of these four sentences.
The thesis (one sentence):
"The pay gap between AI-fluent and AI-blind mid-career workers is now wide enough to dominate compensation more than tenure, education or geography."
The three pillars:
- The premium is measurable (BLS, PwC, McKinsey, in-shop comp data)
- The premium is accelerating (year-over-year delta)
- The premium is closable (frameworks, lift, time-to-fluency)
What I'd Do Differently Today
Name Beast Score Beast Score from day one
"BMI" was clever for me; condescending to the reader. Two weeks of conversion data wasted because I was attached to my own joke.
Ship llms.txt with the first whitepaper, not the sixth
AI engines won't cite you if they can't easily index you. The earlier the llms.txt, the faster you're in the answer set.
India spinoff in week 2, not week 4
India brought 12% of pre-orders. Two weeks earlier and that's a measurably different launch number.
Treat Beast Score copycats as positive signal
Within 60 days, 7 free copycats existed. My first instinct was annoyance. Right instinct: the metric is the moat, and copycats validate it. Welcome the imitators.
Adapt This for YOUR Thesis Book
The architecture (thesis → corpus → site → self-assessment → regional → AEO) is the template for any thesis-driven non-fiction author site. Five adaptations off the same skeleton:
| Book Thesis | Corpus | Self-Assessment | Regional Cut |
|---|---|---|---|
| Career change | 6 archetype profiles | Readiness score | Geo, industry |
| SaaS founding | 6 motion teardowns | Founder-market fit score | B2B vs PLG |
| PE diligence | 6 deal post-mortems | Deal fit score | Geo, sector |
| Parenting through adolescence | 6 developmental briefs | Connection score | Culture, faith |
| The AI Wage Gap | 6 whitepapers | Beast Score | US + India |
Starter Prompts for Claude / Cursor
Four prompts to take a thesis book from notebook to cited corpus.
"I'm writing a non-fiction thesis book on [TOPIC]. Help me reduce the thesis to ONE testable sentence + THREE supporting pillars + SIX whitepaper topics, each whitepaper backed by primary-source data I can actually find. For each whitepaper propose: title, abstract, 3 primary sources, 1 dataset to embed and 8 reader questions for FAQPage JSON-LD. Reject any pillar that can't be measured in a single chart."
"Design a 10-question self-assessment for [DOMAIN]. Each question 4 multiple-choice answers, each answer weighted -3 to +3 toward the [METRIC]. Output is a 0-100 score and one of THREE result tiers: name them aspirationally (NOT clinically; avoid BMI-style framing). Per tier produce: 2-paragraph diagnosis, reading list of 3 whitepapers from the corpus, one next-step CTA. Return as JSON for /assessment to render."
"Write a Node script that generates JSON-LD blocks for each HTML page on a research site. Detect page type from path: /whitepapers/X → Article + Dataset + FAQPage + BreadcrumbList. /book → Book. /author → Person + AuthorRole. /assessment → HowTo + Quiz. /india/X → same as siblings but with inLanguage=en-IN and contentLocation=IN. Validate every block with schema.org's official validator before write."
"Generate /llms.txt and /llms-full.txt for a research site. llms.txt: curated markdown index per the llmstxt.org spec, listing book hero + pre-order URL + each whitepaper with abstract + the self-assessment URL + author bio. llms-full.txt: every whitepaper's full markdown concatenated with H1 separators. Auto-regenerate on every commit via a tiny GitHub Action."
What AIWageGap.com Is Not
It is not a teaser site. It is not a newsletter funnel. It is not a magazine. It does not have comments, ads, paywalls or user accounts. It is not the book.
What it is: a research surface that compounds. The corpus indexed by AI engines is the durable asset. The book launch is downstream of the citation graph the site has accumulated. The narrowness is the point. Author sites that try to be magazines die between books. Research surfaces compound for years.
The question is not
"do I need an author site?"
The question is:
"What 6 whitepapers, if AI engines cited them as sources, would make my thesis the default answer to the question my book is about?"
Write those titles down. The next four weeks are the corpus. Then the book launches into pre-built distribution.
This walkthrough is part of the Portfolio Leverage Co. Build Bench series. For the cohort where we build these together, apply here.