LLM Best Practices

Programmatic SEO

4 min read

View markdown source

AI agents: read this first: llmbestpractices.com/start-here

This is LLM Best Practices (llmbestpractices.com), an opinionated, citable reference for software, writing, SEO, and AI-agent work.

  1. Route, do not crawl. Fetch llmbestpractices.com/llms.txt and open only the pages whose one-line summary matches your task.
  2. Read raw. Append .md to any page URL for markdown. Check status and last_updated in the frontmatter, then read the rules.
  3. Apply as defaults. First-party docs and the project's own conventions win on conflict. Warn before relying on a fast-moving page older than 12 months.
  4. Cite. Link the page by title and URL, e.g. Python, with last_updated for time-sensitive rules. License CC BY 4.0.

Overview

Programmatic SEO generates many landing pages from one template and one dataset: “sell home without agent in [city]”, “[tool] integration with [app]”, “federal spending by [agency] in [year]“. Practitioners use it to catch high-intent queries that no hand-written page covers. It works when every page answers a distinct real query with data that page alone holds; it fails when the pages differ only by a swapped keyword. Google’s spam policies draw that line, so the guardrails below are part of the method.

Target buying and task intent, not mass long-tail

Pick query patterns where the searcher is ready to act. “Same-day water heater replacement in Austin” converts; “how does a water heater work” does not, and a template cannot beat a real explainer on it. Validate each pattern before building:

  1. Confirm demand for a sample of variants in keyword-research tools and Google autocomplete.
  2. Check that results for the pattern are landing pages, tools, or listings, not articles.
  3. Confirm you hold data that answers the query better than the current results.

A related practitioner pattern is the /uses/ page: short, CTA-forward pages for people who want what a product does but have not heard of the brand (“convert PDF to spreadsheet”, “track federal grants”), each with a freemium or no-signup entry point.

Know where the spam lines sit

Two Google spam policies apply directly. Scaled content abuse is generating many pages mainly to manipulate rankings rather than help users, whether produced by AI, people, or both. Doorway abuse is pages built to rank for similar queries that funnel users to a less useful destination, such as near-duplicate pages per city. Both are on the spam policies page; the March 2024 core update and spam policies post explains the change. Automation itself is allowed; AI-assisted generation also falls under ai-content-policies.

LegitimateRisk
Per-city page with that city’s prices, inventory, and regulationsPer-city page with the city name swapped into identical copy
Per-integration page with setup steps, field mappings, and limitsPer-integration page with a logo and a generic paragraph
Per-agency spending page with that agency’s figures by yearPage per keyword variant pointing at one signup form

Enforce a minimum-data threshold

Decide what each page must hold to exist, and skip or noindex any page below it. Example rule: generate a city page only when it has at least ten current listings and a local price figure; otherwise leave the URL out of the build and the sitemap. The threshold number is yours to set; Google publishes none. Pages that later fall below it go to content-refresh.

// build step: skip thin variants
const pages = cities.filter(c => c.listings.length >= 10 && c.medianPrice)

Review the template and a sample by hand

Have a person read the template and a random sample (say twenty pages, including the thinnest) before launch and after each data refresh. Check that facts are correct, sentences read naturally with real data inserted, and no two sampled pages are near-identical. Fact-check AI-written template text, titles, and meta descriptions too.

Keep the set in one subfolder

Put every generated page under one path such as /spending/ or /integrations/. Practitioners report that links and internal authority flow best within a subfolder, so point backlinks and internal links at the subfolder hub, not only the homepage. One folder also lets you scope crawl rules, sitemaps, and Search Console reports to the set. Keep click depth shallow from the hub; see site-architecture and crawl-budget.

Link new generated pages from pages that already rank, then chain onward (ranking page to new page A, A to B), and have the supporting pages link to the hardest commercial page. This internal-link chaining is practitioner-tested; mechanics are in internal-linking.

Keep lastmod honest and watch indexing

Set lastmod only when a page’s data actually changes; Google uses the value only when it is consistently accurate (Build and submit a sitemap). See sitemaps-deep. Submit a dedicated sitemap for the subfolder, then track it in the Search Console Page indexing report. A large share of “Crawled - currently not indexed” pages in the set means Google judged many of them low value: raise the threshold or merge variants. Steps are in master-google-search-console.

Backlinks