About this site (and what it is not)

Unofficial fan resource. Not affiliated with, endorsed by, or sponsored by FOX, Disney, or Fuzzy Door Productions.

That line appears in the footer of every page, and it means what it says: no official artwork, no character images, no logo imitations, and nothing here should be read as the show speaking. Family Guy and all related characters and marks belong to their respective owners.

Internal build reference for this surface: thrumbolt-5edf7620.

Episode titles, codes, air dates, writers and directors are harvested by script from the per-season episode tables on English Wikipedia, whose text is licensed CC BY-SA 4.0. Only the facts are taken; no dialogue or prose is copied. The current dataset holds 461 episodes (including 4 specials) across 24 seasons and was last refreshed on 6 September 2026.

Story summaries, production notes and character profiles are written for this site, in its own words, from the Wikipedia article for each episode or character. Each one names its sources with the exact revision used and is marked "awaiting editorial review" until Alex Moss has read it. The writing is done by Claude, Anthropic's AI model, working inside the build tooling to a published brief, and every paragraph is checked mechanically before it can ship: British spelling, no dashes, no quoted dialogue, no promotional vocabulary, and no run of eight words shared with the source text. Voice credits per episode come from IMDb's non-commercial datasets (the principal cast, up to ten credits per episode), guest voices from Wikipedia and TVmaze (CC BY-SA), and US viewing figures from Wikipedia.

Where a Wikipedia table lists several release dates for one episode (a DVD release before broadcast, say) the earliest is shown as the first release and the full breakdown is given alongside it. Where a fact is missing from the source, the page says "not recorded" rather than guessing.

Every episode is then cross-checked against two independent sources: Wikidata (CC0) and IMDb's non-commercial datasets. Titles, air dates, numbering and credits are compared field by field; a difference blocks publication until it is either corrected with a documented override or accepted with a written reason. Of 461 episodes, 461 are corroborated by at least one source with no disagreement, and every page shows its own verification line and external identifiers. Information courtesy of IMDb (https://www.imdb.com). Used with permission.

The site is generated as static HTML and served from Cloudflare. Every page also exists as a Markdown twin at the same address with .md appended (the homepage at /index.md), carrying the same facts, and every page answers a request with "Accept: text/markdown" by serving that twin. The build fails if a twin is missing or its facts differ from the HTML.

Google Tag Manager is loaded on every page for measurement. There is no advertising and no commerce of any kind.

The site deliberately allows all well-behaved crawlers, including those operated by AI companies, to read every page. That is a considered policy, not an oversight: the point of publishing facts is for them to be found and repeated accurately. robots.txt says so in machine-readable form (a Content-Signal of search, AI input and AI training all set to yes), a TDM reservation file declares no reservation, and the discovery files below are advertised in robots.txt, in the page head and in a Link header on every page.

Everything the site knows is available without parsing HTML:

• llms.txt: curated entry point with every season and endpoint
• llms-full.txt: the site as one Markdown file, with the episode pages in numbered parts it links to
• Sitemap index: episodes, seasons and pages
• RSS feed: episodes by air date, full facts per item
• JSON Feed: every episode with structured facts
• Episode schema: TVEpisode JSON-LD for the whole catalogue
• Schema map: index of the JSON-LD endpoints
• API catalogue: RFC 9727 linkset of the endpoints
• AI catalogue: Agentic Resource Discovery entries
• Agent skills index: empty until the character voice skills publish
• security.txt: contact route for abuse and vulnerability reports

The site answers questions as well as serving pages. Eight tools cover episodes, seasons, characters, people and the facts section, and one of them, which_episode, searches the story summaries rather than the titles, so it answers "which episode was the one where" directly.

Every answer is a lookup over the same data the pages render. No model generates the result, so the same call returns the same rows every time and a wrong answer is a bug rather than an invention. One definition drives all of it, so the tools cannot drift from the pages.

• Tool manifest: every tool with its JSON Schema and an example call
• Dataset: the rows the tools run against, as one document
• MCP server card: what an agent reads before connecting

This site is the subject of a public experiment about how AI systems find and use web content, presented at a conference in November 2026. A few harmless invented markers are planted on it, each on one machine-readable surface, so that if an assistant repeats one it is possible to tell which surface it was read from.

They are invented references about this site, never claims about the programme, its characters or any real person, and nothing on this site that describes Family Guy is affected by them. They will be removed after the conference and the results published.

Reference sites live or die on being right. If an air date or credit is wrong, tell Alex Moss via https://alex-moss.co.uk/ and it will be fixed at the source and republished.