Skip to main content
FamilyPlans
Get the newsletter

Methodology

This page explains how FamilyPlans gathers, verifies, and presents data, and why some pages are not in our index.

Where our data comes from

Every published fact on FamilyPlans traces to a specific source: a council family-hub page, a leisure-centre operator's public listings, a Department for Education dataset, an Office for National Statistics file, or a charity research publication. The full source register is at /methodology/sources with each source's licence, refresh cadence, and current status.

How fresh the data is

Each page surfaces a last verified date that updates whenever the underlying source is re-checked; where no check has happened yet, the page says “not yet verified”. Our freshness policy targets flagging pages older than 60 days as stale and removing pages older than 180 days from our index until they are refreshed. This enforcement is not yet fully automated; checks are currently run as part of our scraping and review cycles. Public scraper status — including when each scraper last ran successfully — is at /methodology/scrapers.

How we decide what to publish

Every generated page starts noindex. To reach Google's index, a page must:

  1. Pass hard minimums for its page type (number of verified facts, presence of correction route, clear editorial responsibility, internal-link support).
  2. Score above the per-page-type threshold on a weighted quality score covering first-party data, usefulness, uniqueness, provenance, freshness, schema validity, internal links, Core Web Vitals, and editorial responsibility.
  3. Pass duplicate-similarity checks (no two pages may look like a place-name swap of each other).
  4. Pass schema-validation checks.

Full description of the gate, including thresholds and reason codes, is at /methodology/quality-gate.

Use of AI

AI assistance is used in two places. First, parts of our activity database were collected with LLM-assisted extraction of councils' and providers' public pages inside editorial sessions, with deterministic import and review before publication. Second, AI may assist drafting answer-first paragraphs, summarising data, and copy-editing — content is reviewed before publishing and a responsible editorial contact is accountable for every published page. AI is never used in production runtime: the site has no LLM API dependency. Our full AI policy is at /methodology/ai-use.

Corrections

Anyone can challenge any fact via /report-error. We aim to acknowledge within 2 business days. All factual corrections are logged at /research/corrections.

Datasets

Where we hold a useful dataset (Sure Start centre records, Family Hub rollout tracker, and others), we document it publicly; released downloads are licensed CC BY 4.0 with attribution. See /datasets.

What we do not do