create-content

name	create-content
description	TRIGGER when: user asks to create a new post, write an article, create a new page, or add new content blocks. Provides tone guidelines, SEO rules, content structure, MCP workflow, layout block conventions, and Lexical JSON patterns.

Content Creation

Tone & Content Guidelines

Auschwitz-Birkenau is a memorial and place of remembrance, not a tourist attraction. All content must reflect this.

• Somber, respectful tone — convey memory and witness, not leisure travel
• No marketing speak — no tourism framing, no promotional language
• Practical but dignified — help visitors prepare while honouring the site's significance

Wording Guidelines by Language

All languages: choose words conveying memory, witness, and learning — never leisure or entertainment.

Language	Prefer	Avoid
Polish	zwiedzanie, upamiętnienie, pamięć, poznanie historii	wycieczka, wyjazd, wypad, wizyta turystyczna
German	Gedenkstättenbesuch, Besuch der Gedenkstätte	Ausflug, Reise
French	visite commémorative, visite du Mémorial	excursion, sortie

"Book" = "reserve" (not "buy/purchase") — Auschwitz entry cards can be free. In all languages, translate "book" as "reserve" (zarezerwować in Polish, not kupić).

SEO Metadata Rules

• Title tag (pages): < 53 characters (frontend appends | {year}, ~60 visible in SERP)
• Title tag (posts): < 59 characters (no year appended — this IS the full SERP budget)
• CTR is the priority — use the full character budget. A short title wastes prime SERP real estate. Front-load primary keyword, include differentiators (transport modes, numbers, specifics), make every character earn its click.
• Meta description: 140–155 characters
• Clear, unique, relevant; primary keywords placed naturally — no keyword stuffing

Titles, Headings & Content Structure

These rules apply to SEO title tags, H1, and H2/H3 headings — title tags are the highest-CTR element in search results, so treat them as a priority.

• Clear, unique, relevant, engaging; primary keywords placed in front for better CTR
• Title tag: primary keyword front-loaded; include numbers where relevant — this is the first thing users see in SERPs (see char limits in SEO Metadata Rules above)
• H1: one per page, primary keyword front-loaded, < 60 characters
• H2/H3: use question format only when it genuinely matches search intent and adds value — e.g. "Can You Visit Auschwitz Without a Guide?" triggers featured snippets because people actually search that question. Do not force questions when a descriptive or number-based heading is stronger (e.g. "Driving to Auschwitz: Routes & Parking", "5 Booking Tips to Avoid Sold-Out Dates"). Never change the heading's topic to fit a question — the heading must match the content below it.
• Include numbers in titles/headings where appropriate — e.g. "5 Memorial Sites Near Auschwitz Worth Visiting" — improves CTR
• Current year — include only in the homepage H1 as a freshness signal (e.g. "How to Visit Auschwitz-Birkenau in 2026"). Do not add the year to H1s, H2/H3 headings, SEO title tags, or meta descriptions — the frontend already appends | {currentYear} to every rendered meta title tag automatically. For Posts: include the year in the H1/H2/H3 headings when relevant for SEO or AI search (e.g. policy changes, seasonal updates).
• English titles: Title Case — Polish titles: sentence case, only first word and proper nouns capitalised

AI Search & Rich Results

• Lead with a direct answer — start pages/posts with a concise 2–3 sentence answer before expanding; AI models and featured snippets pull this
• FAQ page — dedicated /faq/ page with H2 per question, direct 1–2 sentence answer first, then detail; add FAQPage schema only to this page — Google removed FAQ rich results for non-authority sites in 2023 but schema still helps AI parsers
• Structured content — prefer lists, tables, step-by-step formats; AI models and Google parse these better than wall-of-text paragraphs
• Cite authority — mention "licensed guide since 2006" or similar credentials; AI models favor authoritative, first-hand sources
• Long-tail keywords — cover specific visitor questions as H2s on the FAQ page — e.g. "Can I bring a backpack to Auschwitz?" — these dominate AI-assisted search

SEO Keyword Research

Before writing titles, H1/H2/H3 headings, or meta descriptions, read the keyword research memory at ~/.claude/projects/-Users-lucky-dev-visit-auschwitz-info-visitauschwitz-cms/memory/seo_keyword_research.md. It contains 100+ researched search phrases in 3 volume tiers (EN + PL) plus long-tail phrases ideal for H2 headings and FAQ questions. Pick phrases that match the content topic and weave them into headings and body text naturally.

Ticket Policy (March 2026)

• Entry cards available only online at visit.auschwitz.org since March 1, 2026
• No on-site ticket sales at the Museum entrance
• Reservations up to 3 months in advance; free entry cards bookable 7 days ahead
• Guided tour tickets available until sold out; real-time availability for last-minute visits
• visit.auschwitz.org is the only official reservation system — Museum does not cooperate with external booking entities
• Source: https://www.auschwitz.org/en/museum/news/visit-auschwitz-org-entry-cards-to-the-memorial-available-only-online-from-1-march,1819.html

MCP Workflow

MCP Tool Priority

• Always check for visitauschwitz-cms-local first (dev server). Only fall back to visitauschwitz-cms-prod if local is unavailable.
• If MCP tools are not available and the workflow requires them, ask the user to reconnect (/mcp) before proceeding — do not attempt workarounds.

MCP Partial Updates (Critical)

Block IDs: When the user provides a block ID (e.g. 69b82f6c82e7f12ebd4a1938), it refers to the blockName field in the CMS, not the MongoDB _id. Use blockName to locate blocks, then use the actual id field for partial updates.

WARNING — Layout arrays are REPLACED, not merged. MCP updatePages/updatePosts passes layout to MongoDB via $set, which replaces the entire array. Sending a partial layout (only some blocks) deletes all blocks not included. The DeepPartial TypeScript type only makes fields optional at compile time — it does NOT merge arrays by id at runtime.

Safe approaches for layout field updates:

1. Full array: Fetch the complete layout, modify in-place, send back the full array. Use for MCP-based updates.
2. MongoDB positional filters: Use scripts/update-locale.ts with $[identifier] array filters to target specific fields within specific blocks without touching the array. Use for bulk field updates (e.g. translating headings across many blocks).

For non-array fields (title, slug, meta, hero), MCP partial updates work fine — send only the fields you want to change.

• Always use select param to limit response size (e.g., select: { "layout": true }).

MCP Layout — Patched

The MCP plugin's Zod schema conversion (json-schema-to-zod@2.6.1) fails on block discriminated unions (anyOf/oneOf with blockType). This affects both create and update for any collection with layout blocks.

Fix applied: pnpm patch @payloadcms/plugin-mcp@3.77.0 — patches sanitizeJsonSchema.js to replace block union items schemas with {} (accept any), skipping Zod validation for blocks while keeping all other validation intact. Patch file: patches/@[e-mail].

With the patch, MCP handles layout fields normally — use MCP for all create/update operations (it handles versioning, hooks, localization automatically).

MongoDB fallback (scripts/update-locale.ts) is still useful for targeted positional array updates on existing docs (e.g., updating a single heading across many blocks). Run with npx tsx scripts/update-locale.ts. Includes Lexical JSON helper functions (richText, paragraph, heading, textNode, linkNode, linebreak).

Pages — Content Block Layout

• 1/6 column = decorative spacer (hidden on mobile, margin on desktop). Never put content in it.
• Column sizes: full, half, twoThirds, oneThird, oneSixth
• richText (above image) = regular inline links (amber underline). Supports ordered/unordered lists.
• richTextEnd (below image) = styled pill/button links via CSS. Best for CTA lists. Supports ordered/unordered lists.
• Always add an empty paragraph after headings (frontend uses padding-top, not margin-bottom)
• Lead answer block — addMarginTop: false, addMarginBottom: false — the block sits directly below the hero without extra spacing

Posts — Layout Blocks

• Text block (blockType: "Text", style: text | quote | emphasis, rich text field: content)
  • emphasis = bordered box, semibold, larger text. Always use as first block (intro/summary) + mid-article callouts.
  • quote = card with large serif quotation mark. Italic quote text (format: 2). Use for credibility if applicable — spoken quotes from survivors or authorities, but also cited written texts (catalogue introductions, official letters, written statements). Attribution format: — Name, in bold (format: 1), then Role, Organisation, Description and source link in regular (format: 0). Important: search for relevant quotes on the official websites: https://www.auschwitz.org/ or similar and add at least 2 per post if possible, always use exact quotes from the sources, do not paraphrase.
  • text = regular prose paragraphs.
• Length of paragraphs: devide long texts into paragraphs of around 100 words / 600 characters; important: add short descriptive H3 heading for each such paragraph.
• Lists in Text blocks: Use unordered lists for non-sequential items (rules, tips, what to bring) and ordered lists for sequential steps (directions, booking process, itinerary). Lists improve readability and are preferred by AI search and featured snippets. Available in all Text block styles.
• Linking — important: Every Text block must have a short, descriptive blockName (e.g. getting-there, what-to-bring, booking-tips). This serves two purposes: CMS admin readability (blocks are labelled in the sidebar) and anchor targets for internal linking within posts. Links use the simple #blockName structure.
• Table of contents — important: Immediately after the lead emphasis intro block, add a text-style Text block (blockName: table-of-contents) containing an unordered list of anchor links to each main H2 section in the post. Each link points to #blockName of the corresponding section's Text block. Example: [Getting There](#getting-there). Keep the ToC concise — only H2-level sections, no H3 sub-sections.
• Image block (blockType: "Image", media: media ID, caption: richText — not a plain string)
• Block source mapping: Banner (import) → slug "Text", Code (import) → slug "Image" — config files: src/blocks/Banner/config.ts, src/blocks/Code/config.ts
• H2 spacing: add empty paragraph after every H2
• H3 spacing: add empty paragraph before and after every H3, exception: no spacer paragraph before a H3 heading that starts a block
• No spacer between paragraphs — empty paragraphs are only for heading spacing
• Typical pattern: emphasis intro → table of contents → text + H2 → image → more text/quote/image → emphasis callout

Image Placement

• Every post must include 1 hero image (meta.image) + 3–5 Image blocks spread through the layout
• Before using the placeholder, check the media collection (findMedia) for a relevant existing image
• If no relevant image exists, use placeholder 67be70ae35ec329c954f5410
• Placeholder in Image block: set the caption rich text to describe what the image should show (its future contents) — the user will replace the placeholder and use the caption as alt text for the real photo
• Placeholder as hero image (meta.image): append ALT: <desired image alt text> at the end of meta.description after the real SEO description — the user will swap the hero photo and use this alt text

Lexical JSON Conventions

• Empty paragraph after headings: { type: "paragraph", children: [], ... }
• Heading tag: tag: "h2" / "h3"
• Bold: format: 1, Italic: format: 2, Underline: format: 8
• Links: type: "link" wrapping text, fields: { url, newTab, linkType: "custom" }
• Unordered list: type: "list", tag: "ul", listType: "bullet", start: 0
• Ordered list: type: "list", tag: "ol", listType: "number", start: 1
• List items: type: "listitem", value: 1 (increments per item), checked: null; children can contain text nodes, formatted text, and link nodes

Content Accuracy

• Only use verified information. When basing content on an existing page, extract actual links, numbers, and details from that page. Never invent URLs, prices, or details. If unsure, ask the user in a [QUESTION] comment.
• Accordion content is rich. Always extract and use content from accordion blocks — they contain detailed, verified information (timetables, bus lines, prices, step-by-step directions).
• Try token-saving methods first. Use small targeted commands (regex, jq, grep) before spawning agents or reading entire files. Always prefer the simplest approach.

Global Content Rules

• Always present full post content for user review before injecting into CMS. Draft all texts, images, links, SEO metadata, and block structure in the plan/conversation first. Only call createPosts after the user approves. Use [QUESTION] inline comments for anything uncertain.
• Placeholder image: media ID 67be70ae35ec329c954f5410
• Posts always created as draft
• Author: always set authors to Łukasz (675f51ab4d074485ad8b59af) when creating posts
• SEO: title < 53 chars for pages, < 59 for posts (see CTR rules in SEO Metadata Rules), description 140–152 chars