Programmatic SEO Without AI-Generated Content: Real Numbers from a SaaS Test

Programmatic SEO is having a moment. Build a database, generate hundreds of pages with AI, hit publish, watch traffic compound. The pitch is seductive: scale that doesn't require writing. The dirty secret: Google has gotten very good at detecting AI-generated thin content. The 2024 helpful-content updates demoted entire sites built on this pattern. By 2025 the demotion rate accelerated. By 2026 the model is broken for most niches.

So what's replacing it? Programmatic SEO is still the right strategy — long-tail keywords still convert higher than head terms. What changed is the quality bar. The new approach: programmatic structure, hand-written content per page. This piece walks through our test on i3dify — 30+ landing pages, none AI-generated, real numbers on what ranked, what didn't, and the framework that emerged.

The test setup

i3dify is a tool that turns PDFs and images into interactive 3D viewers. SaaS, $0-$6/month tiers, heavily reliant on organic. We shipped 30+ programmatic landing pages over 6 weeks across five formats: free-tool landings (6 pages, ~1500 words each), competitor comparison pages (3 pages, ~1800 words each), industry landings (5 pages, ~1200 words each), platform integration guides (6 pages, ~1400 words each), niche template landings (12 pages, ~1000 words each).

All content hand-written. No AI batch generation. We used a config-driven page template (TypeScript config per page), but the content in each entry was written manually. Total time: ~50 hours over 6 weeks.

The numbers (first 90 days)

• ●Pages indexed by Google: 100% within 14 days (helped by IndexNow plus sitemap)
• ●Pages ranking in top 10 for target query: 11 of 30 (~37%)
• ●Pages ranking in top 3: 4 of 30 (~13%)
• ●Organic traffic generated: ~12,000 visits in month 3
• ●Signups attributed: ~320 in month 3
• ●Conversion rate: 2.7% visit → signup (higher than 1.4% homepage)
• ●Cost per signup: $0 (no paid acquisition)

What ranked best

Specific intent plus specific answer

The /free/3d-restaurant-menu page ranks #2 for free QR menu maker because it answers a specific question with a specific answer. Generic AI-generated content can't do this — the specificity requires actually understanding the niche. A generic best menu makers in 2026 listicle we tried with AI as control: ranked nowhere, demoted within 30 days.

Honest comparison with real numbers

The /vs/issuu page ranks page 1 for Issuu alternative because it includes real pricing for both products, a TL;DR telling the reader when each tool is the right choice, a Pick Issuu if section that doesn't dunk on Issuu, a FAQ addressing migration concerns. Pages that dunk on competitors read as biased and get demoted. Pages that fairly explain when the competitor is the better choice get rewarded.

Tutorial pages with concrete steps

The /integrations/notion page ranks page 1 for embed flipbook in notion because it walks through the exact steps with code snippets and platform-specific gotchas. The more concrete and platform-aware the content, the better it ranks.

What didn't rank

About 60% of pages haven't broken into top 10 yet. Pattern 1: too-narrow long-tail with no search volume. Some templates targeted niches with very low search volume (e.g. school yearbook template, ~50 monthly searches). Even ranking #1 doesn't generate meaningful traffic. We ship these for internal linking support but they're not direct drivers.

Pattern 2: too-broad without differentiation. Initial /free/portfolio-builder ranks nowhere — free portfolio builder is dominated by big-domain sites (Wix, Squarespace). New domain has no chance. Rewrote as /free/3d-portfolio-builder. Ranks page 2 for 3d portfolio builder within 30 days. The differentiation in the query made the niche winnable.

Pattern 3: pages lacking internal links. First 10 pages weren't cross-linked. Six weeks in we added a Related pages component linking 4-6 thematically-related landings. Within 3 weeks, average ranking of the cluster moved up. PageRank flows through internal links.

The framework

• ●Pick winnable queries — specific intent, 500-5000 monthly searches, not dominated by mega-domains
• ●Write one substantive page per query — 800-1500 words, structured, specific, honest
• ●Build a structured page template — TypeScript config plus generic page component
• ●Add structured data — WebPage, HowTo, FAQPage, BreadcrumbList JSON-LD on every page
• ●Cross-link aggressively — Related Pages section with 4-6 hand-curated links per page
• ●Push IndexNow on every publish — cuts indexing time from 1-3 weeks to 1-3 days
• ●Verify each page with Lighthouse — score >95 on SEO, mobile-first, loading <2s

The anti-pattern: AI batch content

We ran a small control: 10 blog posts generated with Claude Sonnet, edited lightly, published. Results after 90 days: 9 of 10 indexed, 1 of 10 ranking top 10, ~80 visits/month total across all 10, 1 signup attributed. The hand-written work outperformed AI batch by ~15x on traffic and ~20x on signups. Cost differential (AI gen ~$5 in tokens vs hand-written 50 hours over 6 weeks) makes AI look cheap on the surface, but actual return per dollar is far worse.

Why? Google demotes AI content (2024 helpful-content updates specifically target thin AI-generated copy). And AI content lacks specificity — it says generic things in fluent language. Specific things — the cover slide of your deck should match your one-liner — require actual understanding. Use AI for first-draft outlining (then heavily rewrite), translation, SEO meta description, completeness review. Don't use it for generating page body, comparison content, expert claims.