The SEO and Content Infrastructure Behind 400+ Blog Articles and 19,000 Programmatic Pages

Tenders-SA serves over 19,000 unique URLs: tender detail pages, province-category combinations, organisation profiles, blog articles, tool pages, intelligence articles, and more. Getting these pages to rank well in search engines — and to display useful rich results when they do — requires an SEO infrastructure that goes far beyond meta tags.

This article covers the key architectural decisions behind the content system: the JSON-based blog engine, the programmatic SEO page generator, the structured data system, and the AI-powered SEO metadata pipeline.

The stakes here are not abstract for the businesses that rely on this content. A construction contractor in Polokwane searching for guidance on CIDB grading, or a cleaning company in Nelspruit trying to understand B-BBEE set-aside rules, is far more likely to find that guidance through a search engine than by browsing the platform directly. If the underlying pages are thin, poorly structured, or invisible to crawlers, that supplier simply never discovers the tools that could help them win their next contract. SEO infrastructure, in that sense, is not a marketing afterthought bolted onto the platform — it is the on-ramp that connects South African SMMEs to procurement guidance and tooling they didn't know existed.

The Blog Content System

The blog uses a JSON-based content system instead of MDX. Each article is a JSON file in /content/blog-json/ containing metadata and structured content blocks:

Copy
1// Each blog post is a strongly-typed JSON structure
2export type ContentBlock =
3    | HeadingBlock    // { type: 'heading', level: 1-6, text, id? }
4    | ParagraphBlock  // { type: 'paragraph', content, className? }
5    | CodeBlock       // { type: 'code', language, code, title?, showLineNumbers? }
6    | ListBlock       // { type: 'list', ordered, items: ListItem[] }
7    | ImageBlock      // { type: 'image', src, alt, caption?, width?, height? }
8    | VideoBlock      // { type: 'video', platform, videoId }
9    | BlockquoteBlock // { type: 'blockquote', content, citation? }
10    | TableBlock      // { type: 'table', headers, rows, caption? }
11    | DividerBlock;   // { type: 'divider' }
12
13export interface BlogPostJSON {
14    slug: string;
15    metadata: BlogPostMetadata;  // title, excerpt, author, category, tags, SEO, howTo, FAQ
16// ... (truncated)
Show More (5 more lines)
TYPESCRIPT

This approach was chosen over MDX for several reasons: type safety (every content block is validated against TypeScript interfaces), portability (JSON is language-agnostic — we could switch rendering frameworks without rewriting content), AI-generability (the structured format makes it straightforward for AI to generate valid articles), and performance (JSON parsing is faster than MDX compilation, and we can cache the parsed result aggressively).

A build script (scripts/generate-blog-manifest.js) reads all JSON files and generates /content/blog-manifest.json — a master index that drives blog listing pages, RSS feeds, and sitemap generation.

Programmatic SEO Pages

Beyond the 400+ blog articles, we generate programmatic SEO pages for every province-category combination: /sa-tenders/gauteng/construction, /sa-tenders/western-cape/healthcare, etc. These pages are database-backed and render at runtime using Incremental Static Regeneration with a 2-hour refresh window.

Each page includes: a listing of active tenders for that province and category, procurement statistics (number of active tenders, awarded contracts, award rate), an AI-generated description specific to the province-category combination, cross-links to related categories and provinces, and structured data (CollectionPage + ItemList + BreadcrumbList) for rich search results.

We also maintain dedicated landing pages for high-value SEO terms: /government-tenders-south-africa, /cleaning-tenders-south-africa, /construction-tenders-south-africa, /security-tenders-south-africa, /tender-winners-south-africa, /ai-tender-matching-south-africa, and more.

The Structured Data System

Every page on the platform emits structured data appropriate to its content type. The system is built around reusable generators in src/lib/seo/:

• Tender pages: BreadcrumbList, WebPage with paywall schema (isAccessibleForFree: false with CSS selectors for gated content), and GovernmentService or BusinessAudience
• Blog pages: Article with author, published date, and featured image; FAQPage for articles with FAQ sections
• Listing pages: CollectionPage + ItemList with per-item metadata
• Organisation pages: Organization with contact info, identifiers, and award statistics
• Pricing page: Service types with ZAR pricing
• Tool pages: SoftwareApplication with feature descriptions and pricing
• Site identity: Organization + WebSite with SearchAction on every page

The Tender Radar pages are particularly structured-data rich: they emit Organization, WebSite + SearchAction, BreadcrumbList, WebPage with speakable selectors, a Service schema describing the company archetype, and an ItemList of live matching tenders.

Meta Description Generation

With 19,000+ pages, manually writing meta descriptions is impossible. We built an AI-powered fallback generator that builds meaningful descriptions from page content, with word-boundary-aware truncation at 160 characters. Category pages mention specific opportunity types; province pages mention relevant sectors; organisation pages lead with the active tender count; forensic pages use procurement risk and compliance keywords.

The fallback description length was raised to 155 characters (up from 100) in July 2026, ensuring every page has a fuller search snippet. The terminal fallback was also fixed — previously an empty string, it now builds a meaningful description from the tender reference, organisation, and province.

Sitemap Architecture

We maintain 7+ sitemap types served as dynamic routes: /sitemap.xml (index), /sitemap-tenders.xml, /sitemap-blogs.xml, /sitemap-categories.xml, /sitemap-provinces.xml, /sitemap-province-categories.xml, /sitemap-static.xml (static pages, tools, Tender Radar), /sitemap-docs.xml, /sitemap-awarded-tenders, /sitemap-closed-tenders, and /sitemap-company-intelligence.

For timely notification, we use IndexNow — pushing URL changes to search engines as they happen. When a new tender is ingested or an article is published, the IndexNow API is called immediately. This ensures new content appears in search results within hours rather than days.

Structured data also earns its keep well beyond the search results page. A well-formed Article schema on a blog post, or an ItemList on a province-category listing, is what lets that page appear as a rich result rather than a plain blue link — with an author name, a publish date, or a count of listed items visible directly in the search results. For a blog covering technical procurement topics like CIDB grading or B-BBEE consolidation, that extra context in the search snippet often makes the difference between a click and a scroll past, because it signals to the searcher that the page is a substantive, attributed piece of content rather than a thin aggregator page.

AI-Ready Content Delivery

AI platforms like ChatGPT, Gemini, Claude, and Perplexity need clean, well-structured content to cite accurately. We serve markdown versions of 70+ public routes at /_ai/* mirror paths, with JSON feeds, JSON sitemaps, and llms.txt files for AI platform discovery. Well-behaved AI bots get 60 requests per minute; abusive scrapers are limited to 10. A daily bandwidth cap monitors AI traffic and alerts if thresholds are exceeded.

Content Generation and Quality

Blog content is generated through a structured pipeline. Topics are organized in clusters covering: basics (starter guides, compliance fundamentals), proposals (winning strategies, document preparation), sector-specific (construction, IT, healthcare, cleaning, security), and provincial (each province's procurement landscape). Each article is authored by one of 19 subject-matter experts (from the AUTHORS.md registry), reviewed, and published through an admin workflow.

Quality controls include: a novelty gate that blocks near-duplicate articles from being published, tag capping at 3 tags to prevent spamming, manual SEO metadata review for feature articles, and automated audit scripts (npm run blog:audit) that check for common issues.

Why Programmatic Pages Need Real Data Behind Them

Programmatic SEO has a well-earned bad reputation in the wider web when it is done by templating a page shell and stuffing keywords into it with no genuine content behind them. Search engines have become increasingly good at identifying and demoting exactly this pattern, and rightly so — a page generated purely to rank, with no real information for the visitor who lands on it, is a poor experience for everyone except the site operator. The reason the province-category pages on Tenders-SA can exist at the scale they do without falling into that trap is that each one is backed by genuinely live, queryable data: real active tenders in that specific province and category, real procurement statistics computed from the underlying award database, and cross-links that route a visitor to genuinely related content rather than an arbitrary internal-linking scheme. The AI-generated description on each page is there to add context, not to manufacture the appearance of substance where none exists.

This is also why the platform actively excludes thin pages from search indexing rather than publishing everything indiscriminately. A province-category combination with only one or two listings genuinely does not have enough content to be a useful landing page, and indexing it anyway would dilute the overall quality signal search engines associate with the rest of the site. Better to keep a handful of low-value combinations out of the sitemap than to have that low value drag down how search engines evaluate every other page on the domain — a lesson that applies just as much to individual blog posts as it does to programmatic listing pages.