Programmatic SEO pre-launch checklist: 24 checks before you publish templated pages

What does Google actually say about scaled and AI-generated pages?

Google's spam policies (page last updated 2026-08-28) define scaled content abuse as producing many pages mainly to manipulate rankings rather than help users, usually with unoriginal content of little value, regardless of how it was produced. Its examples include using generative AI to make many low-value pages, scraping feeds or search results, stitching together content from several pages without adding value, and spreading the content over multiple sites to hide the scale. The same page defines doorway abuse to include near-identical pages and city or regional pages that funnel users to one destination.

Google's guidance on generative AI content (last updated 2026-10-01) says AI can help with research and structure, but producing many pages that add no value can breach the scaled content policy, all AI output should be fact-checked, and the same review applies to titles, meta descriptions, structured data and alt text. It also suggests explaining how content was created. Google's helpful-content self-assessment (last updated 2026-10-05) asks whether a page adds original value, shows who made it and how, and exists mainly to help people.

Google's May 2026 guide on generative AI features adds that creating a page for every possible query or fan-out variation to manipulate rankings or AI answers violates the scaled content abuse policy, and that more pages do not make a site higher quality. Templated pages are not banned: the test is value to the reader, not the method of production.

Gate A: does every page have unique value?

• 1. Every page has at least one element a reader cannot get by swapping the keyword or variable in the template (original data, analysis, a worked example or a downloadable asset). (Basis: Google spam policies: scaled content abuse)
• 2. A visitor arriving from search could reach their goal on this page without going back to the results. (Basis: Google helpful-content self-assessment)
• 3. No page exists only to rank for a query variation, a city or a region (no city pages). (Basis: Google spam policies: doorway abuse; AI optimization guide)
• 4. The page adds something beyond summarising or stitching other pages together. (Basis: Google spam policies: scraping and stitching)

Gate B: where does the data come from?

• 5. Every data field on the page has a named source you opened, with the date opened. (Basis: CiterLabs rule)
• 6. You are allowed to reuse the data; nothing is scraped from feeds, search results or other sites' content and then re-worded. (Basis: Google spam policies: scraping)
• 7. A last-verified date is shown on the page and a refresh owner is named. (Basis: CiterLabs rule; Google: time-sensitive information)
• 8. Rows with missing or weak data do not generate a page (no empty or near-empty pages). (Basis: Heuristic)

Gate C: who reviewed it and how?

• 9. A named human reviewer has read the page or, for a very large batch, a defined sample, and their name is on the page. (Basis: CiterLabs rule; Google: who/how/why)
• 10. Facts, titles, meta descriptions, alt text and structured data have been checked against sources (AI-drafted text included). (Basis: Google guidance on generative AI content)
• 11. The page says how it was made, including any use of AI or automation. (Basis: Google helpful-content self-assessment (How))
• 12. A 'What we do not know' section is present and specific to the page, not boilerplate. (Basis: CiterLabs rule)

Gate D: how different are sibling pages?

• 13. Pick 5 random sibling pages, mask the variable (name, number, place): the remaining text still differs in substance, not only in a few words. (Basis: Heuristic (ours))
• 14. A similarity measure across the batch was run and its failure threshold was written down before looking at results. (Basis: Heuristic (ours); Google publishes no threshold)
• 15. Title, H1 and meta description describe the specific content of each page and are unique within the batch. (Basis: Heuristic (ours))

Gate E: indexing and internal links

• 16. Pages sit in a clear, browsable hierarchy reached from a hub page, not as isolated near-identical URLs. (Basis: Google spam policies: doorway abuse)
• 17. Only pages that passed this gate are in the sitemap; failed pages are noindexed or not published. (Basis: Google spam policies: exclude from Search)
• 18. Status codes, canonical tags and robots rules are correct, and structured data matches visible content. (Basis: Google structured data guidelines)

Gate F: measurement and staged rollout

• 19. Success metrics and a baseline are recorded before wave 1 (indexed pages, impressions, clicks, engagement, assisted leads). (Basis: Heuristic (ours))
• 20. Wave 1 is small; the next wave is released only after the previous wave's indexing status has been checked. (Basis: CiterLabs rule)
• 21. Indexed Y/N is logged per URL with the check date (Page inventory sheet). (Basis: CiterLabs rule)

Gate G: kill criteria

• 22. Kill criteria with thresholds are written in the Kill criteria sheet before launch and signed off. (Basis: Heuristic (ours))
• 23. A named person can pause, noindex or remove a wave without further approval, and a manual-action notice triggers an immediate stop. (Basis: Heuristic (ours); Google: manual actions)
• 24. The bulk removal procedure (noindex or 410, sitemap update) has been rehearsed on a test URL. (Basis: Heuristic (ours))

Worked example: the rule set CiterLabs applies to its own resource pages

This is the rule set we apply to our own resource pages. It is stated as our policy, supplied by the CiterLabs team; we have not measured whether it improves indexing or citations.

How do I use the spreadsheet in 20 minutes?

1. Open Gate and mark each of the 24 checks pass, fail or na with evidence and your reviewer name.
2. Paste every planned URL into Page inventory with its template, unique element, asset and reviewer.
3. Write numeric limits in Kill criteria before publishing. The sheet has bracketed gaps on purpose.
4. Run the sibling-difference check (checks 13 to 15) on a random sample of five pages.
5. Publish wave 1 only if every Gate row is pass or na. Record Indexed Y/N after your chosen interval, then decide on wave 2.