I sell a bundle of 30 business checklists (~450 items). People assume a team wrote them. No — a Python script did, in an afternoon. Here's the exact generator architecture, what I learned about programmatic content, and where the approach breaks.

The data-driven approach

The whole product lives in one Python data structure. Each checklist is a dict; each item is one actionable sentence:

CHECKLISTS = [ { "id": "cash-flow-management", "title": "Cash Flow Management", "audience": "founders, freelancers", "intro": "Thirteen weeks of runway visibility...", "sections": [ { "name": "Setup & Visibility", "items": [ "Open a dedicated business bank account...", "List every recurring subscription...", ... ] }, ... ] }, ... ] 

~15 items per section, 3 sections per checklist, 30 checklists. Everything downstream (PDF, SEO pages, previews) renders from this single source of truth.

Generating the PDFs with reportlab

from reportlab.lib.pagesizes import A4 from reportlab.lib.units import cm from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer from reportlab.lib.styles import getSampleStyleSheet def render_checklist(cl, out_path): doc = SimpleDocTemplate(out_path, pagesize=A4, title=f"{cl['title']} — Checklist") styles = getSampleStyleSheet() story = [Paragraph(cl["title"], styles["Title"]), Spacer(1, 0.4*cm), Paragraph(cl["intro"], styles["BodyText"])] for section in cl["sections"]: story.append(Paragraph(section["name"], styles["Heading2"])) for i, item in enumerate(section["items"], 1): story.append(Paragraph(f"☐ {i}. {item}", styles["BodyText"])) doc.build(story) 

Printable A4, checkbox glyphs, consistent styling. 30 PDFs in under 10 seconds.

The SEO layer renders from the same data

Every checklist gets a preview page: the first 3 items are free (real value, Google-indexable), the rest are locked. One template, 30 pages, cross-linked:

def render_preview(cl, template): free_items = cl["sections"][0]["items"][:3] locked_count = total_items(cl) - len(free_items) return template.format( title=cl["title"], free_items=free_items, locked_count=locked_count, related=[c["title"] for c in CHECKLISTS if c["id"] != cl["id"]][:5], ) 

This is programmatic SEO done honestly: every page has genuinely useful free content, not doorway spam. Whether Google agrees is a live experiment — 32 URLs submitted via IndexNow, results pending.

What I learned about programmatic content quality

  1. Items must be verbs. "Set up weekly cash-flow review" beats "cash flow awareness". Every item is a checkbox action, not a topic.
  2. Specificity > coverage. 15 concrete items beat 40 vague ones. The temptation is to pad — don't.
  3. The data structure is the product. Fixing one typo in the source fixes PDF + SEO page + preview simultaneously. Single source of truth isn't a nice-to-have; it's the whole maintenance story.
  4. LLM-assisted drafting, human-format curation. Drafting 450 items is where an LLM shines; curating which survive (cutting duplicates, vague phrasing) is where quality actually comes from.

Where it breaks

  • Long-tail only. Programmatic pages win niche queries ("cash flow checklist for freelancers"), never head terms. Expect weeks, not days, for traction.
  • Same-ish structure across pages. Google can pattern-detect templated pages. Mitigation: real unique intro text per page, different item counts, genuine cross-links.
  • Zero editorial serendipity. A human writer would have added the weird, specific advice that makes content memorable. The fix: mine real user questions and add curated answers per checklist over time.

The stack summary

Python 3 + reportlab (PDFs) + string templates (SEO pages) + static hosting. Total cost: $0. Total build: one afternoon. The product hasn't needed a content change since launch — because the source of truth is stable data, not prose.

If you're building a content product, start with the data structure, not the design. Everything else renders from it. Free sample of the result is on my profile; happy to answer reportlab/layout questions in comments.