So far every page was handmade — one brief, one draft, one audit. But some needs repeat with structured variation: every city wants its events, every public company deserves a profile, every tool comparison answers the same questions with different facts. Programmatic pages meet those needs with data plus a template. Done honestly, they are a public utility. Done lazily, they are a page factory that search engines rightly ignore.

Data plus template, for genuinely distinct needs

Programmatic content means pages produced from structured data through a shared template, where each output serves a distinct user need. The template holds the structure; the data supplies the difference.

Useful families you already recognize:

• Searchable tool directory: each tool gets features, pricing shape, evidence links, and who it fits — because "which tool handles filings?" differs per tool.
• City event pages: each city gets current, verified listings with dates, venues, and sources — because Friday in Austin is not Friday in Berlin.
• Data-backed company profile: each company gets filing-sourced facts with links to cited paragraphs — because Apple and a small-cap biotech raise different questions.
• Product comparison: each pairing gets actual tested differences — because "Sonariq vs. generic notes app" and "Sonariq vs. terminal data" are different decisions.

Contrast the factory version: a thousand city pages with swapped place names and identical paragraphs, company profiles with stale prices and no sources, comparisons where only the logos change. The template shipped; the value did not. Readers bounce, search engines notice the pattern, and the whole family sinks together.

The six-part quality gate

Before generating page one, every programmatic family must pass this gate. Fail any part and do not publish:

1. Unique user purpose: each page answers a question someone actually asks differently per page. Test: could you write a distinct title and bookmark reason per page without forcing it? 2. Reliable data: facts come from named sources with refresh dates — filings, verified calendars, tested products. "Our model guessed" is never a source. 3. Useful structure: the template organizes a decision — verdict, evidence, differences, limits, next step — not just slots for keywords. 4. Meaningful updates: stale pages get refreshed or retired. An events page with last year's dates is worse than no page. 5. No empty near-duplicates: if two outputs differ only by a swapped noun, merge them or add the missing differentiating data. Hundreds of twins help nobody. 6. A path to deeper action: every page offers a sensible next move — the full brief template, the booking link, the detailed review, the signup with a sample.

Run the gate as a table during planning:

| Gate | Pass looks like | Fail looks like |

|---|---|---|

| Purpose | "Friday AI meetups in Austin" searched monthly | Swapped city names, same text |

| Data | Venue + date verified this week, source linked | Scraped list, no dates |

| Structure | Filter, map, details, add-event action | Paragraph + ads |

| Updates | Nightly refresh + "verified" stamp | 2024 dates in 2026 |

| Uniqueness | Each company profile cites different filings | Same generic description |

| Action | Save seat / download brief / compare | Dead end |

AI speed is not value — and the duty stays yours

AI makes factories tempting: ten thousand pages before lunch. Google's guidance on AI-generated content is explicit that the question is not who — or what — wrote the words. It is whether the page gives people something helpful and original. Mass-producing pages to chase queries, with or without AI, violates the people-first standard from Lesson 76.2; a well-built data system with verified sources and real differentiation does not.

So assign AI the jobs it does well and keep humans on the ones that carry responsibility:

• AI may: normalize data, draft descriptions from verified fields, flag stale entries, group duplicates, suggest template improvements, check coverage against queries.
• Humans must: choose and verify sources, set the "do not publish" rules, review samples before a full generation run, own corrections, and retire families that no longer earn attention.

Connect this to the Lesson 70.2 pattern you already know: preserved originals, AI drafts, human review, published output with links back. A company profile generated from filings should link to the cited paragraphs so a reader — and a reviewer — can check the machine's work. Speed without that trail is not leverage. It is risk at scale.

Exercise: specify one page family

Propose exactly one programmatic family and write PROGRAMMATIC-PAGE-SPEC.md:

# PROGRAMMATIC-PAGE-SPEC.md — [Family name, e.g. City AI-event pages]

1. User question (per page): ___
2. Page count + why each deserves to exist: ___ (e.g. 30 cities with ≥5 verified monthly events)
3. Data fields: ___ (venue, date, source URL, verification date, …)
4. Template sections (in order): ___
5. Source + update process (who verifies, how often, stale rule): ___
6. Differentiation test (what varies meaningfully per page): ___
7. Deeper action (per page): ___
8. Do-not-publish-if: ___ (e.g. fewer than 3 verified events, no update in 30 days, duplicate of existing page)
9. Sample review (3 sample URLs reviewed by ___ on ___): ___

Worked sketch — Neighborhood Events city pages: question "what AI events happen in [city] this month?"; fields include event name, date, venue, organizer, source link, verification date; template runs filter → map → curated list → submit-event action; updates are weekly with a visible "verified" stamp; do-not-publish if fewer than three verified upcoming events or nothing verified in 30 days. Three sample cities get full human review before the remaining twenty-seven generate.

Finish line: a PROGRAMMATIC-PAGE-SPEC.md with fields, question, sections, source-and-update process, differentiation test, and an explicit do-not-publish rule.

Verify quickly: generate three samples and cover the city or company name. If a reader cannot tell which page they are on, the family fails uniqueness — add differentiating data or shrink the family.

Common failure mode: approving the template and skipping the data contract — "we will fill sources later." Later never comes, and a thousand sourced-looking pages ship with invented details. No verified source pipeline, no generation run.

Check your understanding

1. What separates a useful page system from a page factory? 2. Recite the six-part quality gate and name which part most directories fail. 3. Under Google's AI-content guidance, what actually determines whether AI-assisted pages are acceptable?

Next

You can build one good page and plan a family of them. Lesson 76.5 closes the class with the loop that keeps search honest over time: measure, refresh, and retire.