@softspark/ai-toolkit 2.4.1 → 2.6.1

package/app/skills/seo-validate/reference/core-web-vitals.md

@@ -0,0 +1,445 @@
+# Core Web Vitals Reference
+
+Reference for `seo-validate` Category 5. Static-analysis patterns that cause LCP/INP/CLS regressions, plus resource hint guidance and above-the-fold heuristics.
+
+Source: web.dev Core Web Vitals, HTML Living Standard `<link>` rel types, W3C Resource Hints, per-framework image component documentation.
+
+## Thresholds
+
+| Metric | Good | Needs Improvement | Poor |
+|--------|------|-------------------|------|
+| LCP (Largest Contentful Paint) | <2.5s | 2.5–4.0s | >4.0s |
+| INP (Interaction to Next Paint) | <200ms | 200–500ms | >500ms |
+| CLS (Cumulative Layout Shift) | <0.1 | 0.1–0.25 | >0.25 |
+
+75th-percentile mobile values count for ranking. The skill can't measure these at runtime — it flags code patterns known to cause regressions.
+
+---
+
+## LCP (Largest Contentful Paint)
+
+The LCP element is typically the hero image or H1 text in the first viewport. Goal: render it as fast as possible.
+
+### Image dimensions
+
+Every `<img>` and `<video>` must declare `width` and `height` (or `aspect-ratio` CSS). Missing dimensions cause:
+- CLS as the image loads.
+- LCP delay (browser can't reserve space / estimate priority).
+
+```html
+<!-- Good -->
+<img src="hero.jpg" width="1200" height="630" alt="...">
+
+<!-- Bad (CLS + LCP) -->
+<img src="hero.jpg" alt="...">
+```
+
+### fetchpriority
+
+`fetchpriority="high"` on the LCP image tells the browser to fetch it before other resources. Introduced in Chrome 101 (May 2022).
+
+```html
+<img src="hero.jpg" width="1200" height="630" fetchpriority="high" alt="...">
+```
+
+`fetchpriority="low"` on below-the-fold images frees bandwidth for critical resources.
+
+### loading="lazy" on LCP = BAD
+
+`loading="lazy"` delays the image load until it's near the viewport. Applied to the LCP element, this is a HIGH-severity anti-pattern.
+
+Only apply `loading="lazy"` to images **below the fold**.
+
+```html
+<!-- Bad: LCP element marked lazy -->
+<img src="hero.jpg" loading="lazy" alt="Hero">
+
+<!-- Good: LCP element eager + high priority -->
+<img src="hero.jpg" fetchpriority="high" alt="Hero">
+
+<!-- Good: below-the-fold image lazy -->
+<img src="testimonial.jpg" loading="lazy" decoding="async" alt="Jane, happy customer">
+```
+
+### decoding="async"
+
+Signals that image decoding can happen off the main thread. Apply to non-critical images.
+
+```html
+<img src="sidebar.jpg" loading="lazy" decoding="async" alt="...">
+```
+
+Don't apply `decoding="async"` to the LCP image — it can delay paint. Use `decoding="sync"` or omit.
+
+### Preloading the LCP image
+
+When the LCP image URL is known at build time (hero background, article cover), preload it:
+
+```html
+<link rel="preload" as="image" href="/images/hero.jpg"
+      imagesrcset="/images/hero-640.jpg 640w, /images/hero-1280.jpg 1280w"
+      imagesizes="100vw">
+```
+
+### Font loading
+
+Fonts frequently gate LCP when text is the LCP element.
+
+```css
+/* Good: text renders in fallback, swaps when webfont loads */
+@font-face {
+  font-family: 'Inter';
+  src: url('/fonts/inter.woff2') format('woff2');
+  font-display: swap;  /* or 'optional' for aggressive LCP optimization */
+}
+
+/* Bad: FOIT (Flash of Invisible Text) blocks paint */
+@font-face {
+  font-family: 'Inter';
+  src: url('/fonts/inter.woff2') format('woff2');
+  /* font-display defaults to 'auto', which most browsers treat as 'block' = FOIT */
+}
+```
+
+Preload webfonts used above the fold:
+
+```html
+<link rel="preload" as="font" type="font/woff2"
+      href="/fonts/inter-var.woff2" crossorigin>
+```
+
+The `crossorigin` attribute is required even for same-origin fonts — fonts are always fetched with CORS.
+
+### Responsive images
+
+Images wider than ~600px should use `srcset` + `sizes` or `<picture>`:
+
+```html
+<img
+  src="hero-1280.jpg"
+  srcset="hero-640.jpg 640w, hero-1280.jpg 1280w, hero-2560.jpg 2560w"
+  sizes="(max-width: 768px) 100vw, 50vw"
+  width="1280" height="720" alt="...">
+```
+
+Over-fetching a 2560px image to a 375px mobile viewport wastes bytes and delays LCP.
+
+---
+
+## INP (Interaction to Next Paint)
+
+Replaces FID (First Input Delay) as the CWV interaction metric in March 2024. Measures longest interaction latency across the session.
+
+### Render-blocking scripts
+
+Scripts in `<head>` without `async` or `defer` block parsing:
+
+```html
+<!-- Bad: blocks parser + CSSOM -->
+<script src="/analytics.js"></script>
+
+<!-- Good: doesn't block parser, runs after parse but before DOMContentLoaded -->
+<script defer src="/app.js"></script>
+
+<!-- Good: downloads in parallel, runs ASAP (order not guaranteed) -->
+<script async src="/analytics.js"></script>
+```
+
+### Third-party scripts
+
+Analytics, chat widgets, ads, A/B test SDKs commonly cause INP regressions.
+
+Patterns to flag:
+- `<script src="https://www.googletagmanager.com/...">` without `async`
+- Intercom, Drift, HubSpot widgets loaded synchronously
+- Ads via direct `<script>` (most use `async` by default, verify)
+
+Next.js `<Script>` component with strategy:
+```jsx
+<Script src="https://analytics.example.com" strategy="lazyOnload" />
+```
+
+Strategies:
+- `beforeInteractive` — blocks hydration, rarely needed.
+- `afterInteractive` (default) — loads after hydration.
+- `lazyOnload` — loads during idle time. Best for analytics/chat.
+- `worker` — runs in Web Worker (Partytown). Experimental.
+
+### document.write
+
+Always HIGH. Modern browsers disable `document.write` for scripts loaded over 2G-like networks. Never use.
+
+### Hydration cost
+
+Heavy synchronous work during hydration blocks INP. Red flags:
+
+```jsx
+useEffect(() => {
+  // Large sync block — splits Long Tasks
+  heavyComputation();
+  anotherHeavyFunction();
+  yetAnotherOne();
+}, []);
+```
+
+Mitigation: `requestIdleCallback`, `scheduler.postTask`, or split across effects.
+
+### Bundle size
+
+Bundles >300KB gzipped gating interaction = INP risk. Check:
+- Imports of entire lodash / moment / date-fns (use tree-shaking or day.js)
+- Icon libraries imported wholesale (import only what's used)
+- PDF/charting libraries loaded eagerly (dynamic import)
+
+---
+
+## CLS (Cumulative Layout Shift)
+
+### Image/video/iframe dimensions
+
+Covered above — always set `width`/`height` or `aspect-ratio`.
+
+### Iframes
+
+YouTube embeds, maps, Twitter embeds cause CLS without reserved space:
+
+```html
+<!-- Bad -->
+<iframe src="https://www.youtube.com/embed/..."></iframe>
+
+<!-- Good: fixed container -->
+<div style="aspect-ratio: 16/9;">
+  <iframe src="..." width="100%" height="100%"></iframe>
+</div>
+```
+
+### Dynamic content
+
+Ads, embeds, and "related content" widgets injected after paint cause CLS. Reserve space with CSS:
+
+```css
+.ad-slot {
+  min-height: 250px;  /* Or aspect-ratio tied to ad size */
+}
+```
+
+### Font swap shifts
+
+`font-display: swap` causes a visible shift when the webfont loads (FOUT — Flash of Unstyled Text). Minimize by:
+- Using system-ui fonts as fallback with matching metrics (`size-adjust`, `ascent-override`).
+- Using `font-display: optional` (accepts missing-font in exchange for zero shift).
+
+### Hydration mismatches
+
+Server renders one thing, client renders another — content "flashes" or "jumps":
+
+```jsx
+// Bad — hydration mismatch potential
+function Component() {
+  if (typeof window !== 'undefined') {
+    return <ClientOnlyContent />;
+  }
+  return <ServerContent />;
+}
+
+// Good — use useEffect with loading state
+function Component() {
+  const [mounted, setMounted] = useState(false);
+  useEffect(() => setMounted(true), []);
+  return mounted ? <ClientOnlyContent /> : <Skeleton />;
+}
+```
+
+---
+
+## Resource Hints
+
+HTML Living Standard + W3C Resource Hints. Hints are suggestions — browsers can ignore them.
+
+### preload
+
+High-priority fetch for critical, known resources.
+
+```html
+<link rel="preload" as="image" href="/hero.jpg">
+<link rel="preload" as="font" type="font/woff2" href="/fonts/inter.woff2" crossorigin>
+<link rel="preload" as="style" href="/critical.css">
+<link rel="preload" as="script" href="/critical.js">
+```
+
+Rules:
+- Must include `as` attribute matching the resource type.
+- Must appear BEFORE the resource uses it in document order.
+- Don't over-use — >6 preloads starts to hurt.
+- Preloaded resource not used within a few seconds → browser console warning.
+
+### prefetch
+
+Low-priority fetch for probable-next-navigation resources.
+
+```html
+<link rel="prefetch" href="/next-page.js">
+```
+
+Framework-integrated prefetching:
+- Next.js `<Link>` auto-prefetches visible links in production.
+- Nuxt `<NuxtLink>` with `prefetch` prop.
+- SvelteKit: `data-sveltekit-preload-data="hover"` or `"tap"`.
+- Remix: `<Link prefetch="intent">`.
+
+### preconnect
+
+Establishes early connection (DNS + TCP + TLS) to a 3rd-party origin.
+
+```html
+<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+<link rel="preconnect" href="https://www.google-analytics.com">
+```
+
+Use for critical 3rd-party origins: font hosts, CDNs, analytics. Overuse contends with critical-path connections — limit to 3–4.
+
+### dns-prefetch
+
+Lightweight DNS-only resolution. Use for less-critical origins or as fallback for browsers without preconnect support.
+
+```html
+<link rel="dns-prefetch" href="https://some-analytics.example.com">
+```
+
+### modulepreload
+
+Preload ESM modules on the critical path (bypasses discovery cost for import graphs).
+
+```html
+<link rel="modulepreload" href="/build/chunk-main.js">
+```
+
+Build tools can generate these automatically (Vite, esbuild, Rollup plugins).
+
+### Ordering
+
+Preload/preconnect MUST appear before the resource that uses them — typically at the top of `<head>`, right after charset/viewport.
+
+```html
+<head>
+  <meta charset="utf-8">
+  <meta name="viewport" content="...">
+
+  <!-- Resource hints first -->
+  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+  <link rel="preload" as="image" href="/hero.jpg" fetchpriority="high">
+  <link rel="preload" as="font" type="font/woff2" href="/fonts/inter.woff2" crossorigin>
+
+  <!-- Then the resource (stylesheets) -->
+  <link rel="stylesheet" href="/styles.css">
+
+  <!-- Metadata -->
+  <title>...</title>
+  <meta name="description" content="...">
+</head>
+```
+
+### Anti-patterns
+
+- Preloading non-critical resources (every icon, every font weight).
+- Preload without `as` attribute — browser issues warning, may not prioritize.
+- Multiple `preload`s for images that aren't the LCP element.
+- Preload after stylesheets/scripts that already request the resource.
+
+---
+
+## Above-the-Fold Heuristics
+
+True "above-the-fold" detection requires rendering. In static analysis, treat these as ATF candidates:
+
+1. **First `<img>` / `<Image>` / `<NuxtImg>` / `<GatsbyImage>` / `<Image from 'astro:assets'>`** in a route/page component.
+2. **First child of `<main>`**, `<section>`, or `<header>`.
+3. **Components named** `Hero`, `HeroSection`, `Banner`, `Masthead`, `Jumbotron`, `CoverImage`, `SplashImage`, `HeaderImage`.
+4. **`<Image>` with `priority` prop** (Next.js) — developer has already declared ATF intent.
+
+For ATF elements, required:
+- `width` + `height` declared.
+- `fetchpriority="high"` (or framework `priority` equivalent).
+- NO `loading="lazy"`.
+- Matching `<link rel="preload">` if URL known at build.
+
+For below-the-fold images, recommended:
+- `loading="lazy"`.
+- `decoding="async"`.
+- `fetchpriority="low"` (optional).
+
+---
+
+## Framework-Specific Image Components
+
+### Next.js — `next/image`
+
+```jsx
+import Image from 'next/image';
+
+// LCP image
+<Image src="/hero.jpg" alt="..." width={1200} height={630} priority />
+
+// Below fold
+<Image src="/thumb.jpg" alt="..." width={400} height={300} loading="lazy" />
+```
+
+Benefits: automatic WebP/AVIF, responsive srcset, lazy by default, supports `fill` + `sizes`.
+
+Flags:
+- `<img>` used instead of `<Image>` in route components → WARN (unless explicit raw HTML is needed).
+- `<Image>` without `priority` on LCP element → HIGH.
+- `<Image>` without `sizes` when using `fill` → WARN.
+
+### Nuxt — `<NuxtImg>` / `<NuxtPicture>`
+
+```vue
+<NuxtImg src="/hero.jpg" width="1200" height="630" preload />
+<NuxtImg src="/thumb.jpg" width="400" height="300" loading="lazy" />
+```
+
+Requires `@nuxt/image` module.
+
+### Astro — `<Image>` from `astro:assets`
+
+```astro
+---
+import { Image } from 'astro:assets';
+import hero from '../assets/hero.jpg';
+---
+<Image src={hero} alt="..." width={1200} height={630} loading="eager" />
+```
+
+### Gatsby — `GatsbyImage`
+
+```jsx
+import { GatsbyImage, getImage } from 'gatsby-plugin-image';
+
+const image = getImage(data.file);
+<GatsbyImage image={image} alt="..." loading="eager" />
+```
+
+### SvelteKit — `@sveltejs/enhanced-img`
+
+```svelte
+<script>
+  import heroImg from '$lib/images/hero.jpg?enhanced';
+</script>
+<enhanced:img src={heroImg} alt="..." />
+```
+
+---
+
+## References
+
+- web.dev LCP: https://web.dev/lcp/
+- web.dev INP: https://web.dev/inp/
+- web.dev CLS: https://web.dev/cls/
+- Resource Hints: https://www.w3.org/TR/resource-hints/
+- HTML Living Standard `<link>` rel: https://html.spec.whatwg.org/multipage/links.html#linkTypes
+- fetchpriority: https://web.dev/fetch-priority/
+- font-display: https://developer.mozilla.org/en-US/docs/Web/CSS/@font-face/font-display
+- Next.js Image: https://nextjs.org/docs/app/api-reference/components/image
+- Nuxt Image: https://image.nuxt.com/
+- Astro Image: https://docs.astro.build/en/guides/images/
+- Gatsby Image: https://www.gatsbyjs.com/docs/reference/built-in-components/gatsby-plugin-image/

package/app/skills/seo-validate/reference/geo-aeo-patterns.md

@@ -0,0 +1,259 @@
+# AEO — Answer Engine Optimization
+
+Reference for `seo-validate` Category 6 (GEO/AEO). Companion to [geo-guidelines.md](geo-guidelines.md). Focuses on optimizing content to appear in AI-generated answers — not just ranked pages.
+
+All AEO findings are severity `INFO` — emerging practice, no known penalties.
+
+---
+
+## 1. What is AEO
+
+Traditional SEO ranks pages. AEO targets AI-synthesized answers that cite or paraphrase your content without requiring a click.
+
+**Key answer engines:**
+
+| Engine | Mechanism |
+|--------|-----------|
+| Google AI Overviews | Retrieval-augmented summarization over organic index |
+| Bing Copilot | GPT-4 grounded on Bing index; inline citations |
+| Perplexity | Real-time retrieval + LLM synthesis; explicit source cards |
+| ChatGPT Search | OpenAI web retrieval; used in ChatGPT Plus |
+
+**Difference from classic SEO:** AI engines do not rank pages against each other — they extract the most useful fragment. A page ranked #8 with a direct-answer paragraph can beat #1 in AI responses.
+
+---
+
+## 2. Content Structure for AI Consumption
+
+### Direct-answer paragraphs (first-sentence-answer pattern)
+
+AI engines extract the first 1–3 sentences after a heading as the answer to the implied question. Put the fact first.
+
+```html
+<!-- Good -->
+<h2>What is a canonical URL?</h2>
+<p>A canonical URL is the preferred page version search engines should index when near-duplicates exist. Specify it with <code>&lt;link rel="canonical" href="..."&gt;</code> in <code>&lt;head&gt;</code>.</p>
+
+<!-- Bad: answer buried after filler -->
+<h2>What is a canonical URL?</h2>
+<p>When building websites, you often encounter situations where multiple URLs serve similar content...</p>
+```
+
+### FAQ schema with concise Q&A pairs
+
+`FAQPage` is the highest-precision signal. Each `acceptedAnswer` is extracted as a direct response to its `name` question. Pair with matching visible HTML.
+
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "FAQPage",
+  "mainEntity": [{
+    "@type": "Question",
+    "name": "How do I add a canonical tag in Next.js?",
+    "acceptedAnswer": {
+      "@type": "Answer",
+      "text": "In App Router: export const metadata = { alternates: { canonical: 'https://example.com/page' } }. In Pages Router: use <Head><link rel=\"canonical\" href=\"...\" /></Head>."
+    }
+  }]
+}
+```
+
+Rules: question text = natural-language query; answer = self-contained under 300 chars for voice engines.
+
+### HowTo schema with numbered steps
+
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "HowTo",
+  "name": "How to configure robots.txt for a Next.js site",
+  "step": [
+    { "@type": "HowToStep", "position": 1, "name": "Create the file", "text": "Add robots.txt to public/." },
+    { "@type": "HowToStep", "position": 2, "name": "Add directives", "text": "Set User-agent, Disallow, Allow, and Sitemap URL." }
+  ]
+}
+```
+
+Match with a visible `<ol>` — never schema without HTML steps.
+
+### QAPage schema for single-question articles
+
+Use `QAPage` for knowledge-base and support docs structured around one primary question. Include `author` + `answerCount` fields.
+
+### Entity-first writing
+
+Define the subject before explaining it. AI engines build knowledge-graph connections from entity definitions.
+
+```html
+<!-- Good: entity defined first -->
+<h2>Core Web Vitals</h2>
+<p>Core Web Vitals are three Google-defined UX metrics: LCP, INP, and CLS. They are used as ranking signals in Google Search since 2021.</p>
+```
+
+---
+
+## 3. Technical Signals
+
+### llms.txt / llms-full.txt
+
+The `llms.txt` convention (analogous to `robots.txt` for AI agents) lets agents discover curated content. Serve at domain root.
+
+```
+# /llms.txt
+> ExampleCo builds developer tooling for SEO automation.
+
+## Docs
+- [API Reference](https://example.com/docs/api/llms.txt): Full API docs
+- [Getting Started](https://example.com/docs/start/llms.txt): Setup guide
+```
+
+`llms-full.txt` = concatenated full-text for one-shot context loading.
+
+### robots.txt AI bot directives
+
+Known AI crawler `User-agent` strings:
+
+| Bot | Owner | Purpose |
+|-----|-------|---------|
+| `GPTBot` | OpenAI | ChatGPT training + search |
+| `OAI-SearchBot` | OpenAI | ChatGPT Search real-time retrieval |
+| `ClaudeBot` | Anthropic | Claude training |
+| `anthropic-ai` | Anthropic | Claude real-time retrieval |
+| `PerplexityBot` | Perplexity | Real-time answer grounding |
+| `Google-Extended` | Google | Gemini training (NOT AI Overviews) |
+| `Googlebot` | Google | Organic + AI Overviews |
+
+Blocking `Googlebot` blocks AI Overviews. Block `Google-Extended` to opt out of Gemini training only.
+
+```
+User-agent: GPTBot
+Allow: /docs/
+Disallow: /
+
+User-agent: Google-Extended
+Disallow: /
+
+Sitemap: https://example.com/sitemap.xml
+```
+
+### Sitemap with lastmod
+
+Include `<lastmod>` reflecting actual content change dates — not today's build date. AI engines prefer recently-updated sources.
+
+### Clean HTML semantics
+
+AI parsers rely on heading hierarchy. Rules:
+- One `<h1>` per page — becomes the answer title in AI responses.
+- `<h2>` maps to extractable sub-questions.
+- Never skip heading levels.
+- Use `<article>`, `<section>`, `<main>` over `<div>` wrappers.
+
+---
+
+## 4. Content Quality Signals
+
+### E-E-A-T
+
+| Signal | Implementation |
+|--------|---------------|
+| Experience | First-person case studies, original screenshots, tested procedures |
+| Expertise | Author byline + `Person` schema with `sameAs` links (LinkedIn, GitHub) |
+| Authoritativeness | `Organization` schema with `sameAs` to Wikidata/Crunchbase |
+| Trustworthiness | HTTPS, `dateModified` on articles, correction notices |
+
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "Article",
+  "headline": "How to Optimize robots.txt for AI Crawlers",
+  "author": {
+    "@type": "Person",
+    "name": "Jane Doe",
+    "sameAs": ["https://github.com/janedoe", "https://linkedin.com/in/janedoe"]
+  },
+  "datePublished": "2026-03-01",
+  "dateModified": "2026-04-10"
+}
+```
+
+### Citation-worthy content
+
+- Statistics with sources: "76% of developers use React (State of JS 2024)" — not "most developers."
+- Dated claims: include year on every statistic.
+- Expert quotes in `<blockquote cite="...">` with named speaker.
+- Outbound links to primary sources (w3.org, ietf.org, schema.org, developers.google.com, developer.mozilla.org, arxiv.org) — AI engines treat authoritative outbound links as a credibility signal.
+
+---
+
+## 5. Grep Patterns for Automated Detection
+
+All patterns below yield `INFO` + `heuristic` findings.
+
+### Missing FAQ schema on FAQ-like content
+
+```
+Detect: (?i)(frequently asked questions|faq|common questions)  in *.html,*.tsx,*.jsx,*.vue,*.svelte,*.astro
+Then check: "FAQPage"  in same file or its layout
+Flag if: FAQ heading found, no FAQPage schema present
+```
+
+### Missing llms.txt
+
+```
+Glob: llms.txt  in public/, static/, site root
+Flag if: not found
+```
+
+### robots.txt missing AI bot directives
+
+```
+File: robots.txt
+Pattern: GPTBot|ClaudeBot|PerplexityBot|Google-Extended|OAI-SearchBot|anthropic-ai
+Flag if: robots.txt exists but no AI-specific User-agent rules found
+```
+
+### How-to headings without HowTo schema
+
+```
+Detect: (?i)<h[2-3][^>]*>how (to|do|can|should)\s  in *.html,*.tsx,*.jsx,*.vue,*.svelte,*.astro
+Then check: "HowTo"  in same file or layout
+Flag if: how-to heading found, no HowTo schema present
+```
+
+### Missing author/date metadata on articles
+
+```
+Detect: (?i)<article|"@type"\s*:\s*"Article"  in *.html,*.tsx,*.jsx,*.vue,*.svelte,*.astro
+Then check: datePublished|rel="author"|<time datetime
+Flag if: article detected, neither datePublished in JSON-LD nor <time datetime> in HTML
+```
+
+---
+
+## Checklist (Category 6 AEO additions)
+
+- [ ] `llms.txt` present at domain root.
+- [ ] `robots.txt` has explicit AI bot directives.
+- [ ] Sitemap `<lastmod>` reflects actual content change dates.
+- [ ] FAQ content uses `FAQPage` schema with self-contained answers.
+- [ ] How-to content uses `HowTo` schema with `step` array + visible `<ol>`.
+- [ ] Single-question pages use `QAPage` schema.
+- [ ] Articles have `datePublished` + `dateModified` in JSON-LD.
+- [ ] Author bylines use `Person` schema with `sameAs` links.
+- [ ] Statistics include inline source citations with year.
+- [ ] `<h2>` headings phrased as natural-language questions where applicable.
+- [ ] Entity terms defined on first use.
+
+---
+
+## References
+
+- llmstxt.org spec: https://llmstxt.org
+- OpenAI GPTBot docs: https://platform.openai.com/docs/gptbot
+- Google-Extended opt-out: https://developers.google.com/search/docs/crawling-indexing/overview-google-crawlers
+- Google E-E-A-T guidance: https://developers.google.com/search/docs/fundamentals/creating-helpful-content
+- Schema.org FAQPage: https://schema.org/FAQPage
+- Schema.org HowTo: https://schema.org/HowTo
+- Schema.org QAPage: https://schema.org/QAPage
+- Schema.org Article: https://schema.org/Article
+- Princeton/Georgia Tech GEO paper (Aggarwal et al., 2023): https://arxiv.org/abs/2311.09735