Everjust Website SEO

EVERJUST Website SEO — Agent Skill

Make a live everjust.app tenant's public marketing site discoverable — by search engines AND by AI answer engines (ChatGPT Search, Perplexity, Google AI Overviews). You set per-page SEO metadata, control sitemap / crawl state, wire the site's social profiles (which feed the schema.org JSON-LD the platform injects on every page), edit robots.txt, verify Google Search Console, and enable IndexNow. Everything runs through the everjust_agent_mcp server as the connected Odoo user, so you need an admin / website-designer role (Odoo raises AccessError otherwise).

This is the discoverability layer that sits ON TOP of the page-editing skills. If the task is to change a page's copy or layout, that's [[everjust-website]] (Tailwind-QWeb ports) or [[everjust-website-snippets]] (the on-brand s_cd_* snippets). This skill is only about the metadata, indexing, structured-data, and crawler-facing surfaces. For opening a session against the right tenant DB and what each tool returns, see [[everjust-agent-mcp]]; for the platform invariants (COW, the /odoo debrand, everjust.public_website, re-sweep on -u, never self-escalate) read [[everjust-platform]] first.

When to use this skill

• Set / backfill a page's SEO metadata — <title>, meta description, keywords, the Open-Graph image, and the canonical URL slug (seo_name) — for one page or across the whole site (website.page).
• Control the sitemap and crawlability — website_indexed (is it in /sitemap.xml), website_published (is it live at all), and date_publish (schedule a go-live).
• Wire the site's social profiles — set social_facebook / social_linkedin / social_twitter / social_github / social_instagram / social_youtube / social_tiktok on the website so they become schema.org sameAs (a citation/authority signal AI reads), and set social_default_image as the fallback OG card.
• Edit robots.txt — the site-wide website.robots_txt (per-site override).
• Verify Google Search Console — the google_search_console token on website.
• Enable IndexNow — set the everjust_visibility.indexnow_key config param, serve the key file, and switch on the opt-in daily cron that pushes URLs to Bing/Yandex.
• Understand the JSON-LD feed — why the Organization + WebSite structured data on every page is auto-generated by everjust_visibility from the website record.

Do NOT use this skill for, and stop if the task is really:

• Page content / layout / snippets — copy, sections, CTAs, blocks. That's [[everjust-website]] (edit the Tailwind-QWeb arch) / [[everjust-website-snippets]] (the on-brand s_cd_* snippets). SEO metadata is separate from the arch even though it lands on the same view (see the model map).
• The Connect Domain PRODUCT (auto-DNS + SSL + reverse-proxy for a customer's own domain) — that's the control-plane under Customdomain/app/, unrelated to this Odoo site's SEO.
• Registrar DNS for connectdomain.app itself — GoDaddy / Route 53, see [[godaddy-api]]. (Search Console domain-property verification via a DNS TXT record is registrar work; the google_search_console field here is the URL-prefix HTML-tag / meta method.)
• The site's theme / palette / fonts — SCSS in everjust_theme / everjust_brand_website, shipped in code, not MCP-editable.
• Sending mail from a contact form — delivery is the everjust.mail.* stack ([[everjust-mail-ops]]).

The model map (real _names + live-verified fields)

One tenant DB = one website (on connectdomain: id 1, name "Connect Domain", domain https://connectdomain.app). SEO metadata is a mixin whose fields are delegated onto the page's VIEW — this is the single most important structural fact, and it's why COW matters for SEO too.

The delegation gotcha (READ THIS FIRST)

website.page.website_meta_title is related='view_id.website_meta_title', store=False, compute=True, writable — a live-writable proxy. The value lands on the page's ir.ui.view, the same view that holds the page's arch. Consequences:

• Setting page SEO via a generic update on website.page writes through to the view. Going through the website.page write is the correct, COW-safe path — prefer it.
• If you instead wrote website_meta_* directly on a module ir.ui.view (a view whose website_id is empty) without the site in context, you'd mutate the SHIPPED view for every site and it wouldn't COW — same trap as raw-editing arch ([[everjust-website]]). Set SEO on the website.page row, never on a raw ir.ui.view.
• is_seo_optimized is computed and read-only (true once title + description are both set). You never write it; you read it to confirm the page is done.

How SEO renders (what wins at page load)

website.layout renders <title> and the <meta>/OG tags from these fields, overriding any <title> you put in the arch. So set the page's title/description here, don't hand-roll <meta> in the QWeb ([[everjust-website]] Pitfall 4). Per-page website_meta_og_img wins for that page; if unset, the site falls back to website.social_default_image, then the company logo. seo_name is the clean canonical slug used in <link rel="canonical"> and the SEO URL.

The JSON-LD feed — auto-generated from the website record

The everjust_visibility addon (auto_install=True on any tenant that has a website) inherits website.layout and injects, into every public page's <head> (via t-out="website._everjust_jsonld()" xpathed //head), two <script type="application/ld+json"> blocks built by website._everjust_jsonld():

• an Organization block: name (company name, else website name), url (base), logo ({base}/web/image/res.company/{id}/logo), sameAs (the site's set social URLs — see below), and email (only if company.email is set).
• a WebSite block with a SearchAction targeting {base}/website/search?search={term}.

You never author this markup. You influence it by setting the website record's social fields and the company name/email/logo. The generator is wrapped in try/except and returns empty Markup("") on ANY error (or when base/name is missing) — so it can never break the layout, and a bad value fails silently (verify by fetching a page and grepping for application/ld+json).

Exactly the seven fields in the addon's _SOCIAL_FIELDS reach sameAs: social_facebook, social_linkedin, social_twitter, social_github, social_instagram, social_youtube, and social_tiktok. social_discord is a field on the website record but is NOT in that list — it never reaches sameAs (see Pitfall 7). A value must startswith("http") to be emitted.

The extra discoverability routes (everjust_visibility)

New public routes the addon adds (all sitemap=False, and they DON'T override Odoo's own controllers): /llms.txt (an llmstxt.org-format index of published pages), /.well-known/security.txt (RFC 9116, contact = company.email), and /everjust-indexnow-key.txt (404 until a key is set, then serves the key). robots.txt and /sitemap.xml remain Odoo's own — this addon deliberately does NOT override robots (the AI-crawler Content-Signals override is a documented, un-shipped follow-up).

The tool surface (dedicated vs. generic)

There is no dedicated SEO tool — SEO/indexing/social are all generic ORM writes through everjust_agent_mcp. The dedicated website_* tools cover the adjacent content ops. Route each recipe through the MCP against the tenant (see [[everjust-agent-mcp]]).

Confirm gates that apply here: delete always needs confirm:true; writing ir.config_parameter and ir.cron (structural models) needs confirm:true; a non-read call (e.g. the IndexNow submit) needs confirm:true. website.page / website updates are NOT confirm-structural. Granting admin and writing secret config params are hard-blocked regardless — everjust_visibility.indexnow_key is not classified secret, so it's writable with confirm.

Recipes

Route each through the everjust_agent_mcp MCP against the tenant ([[everjust-agent-mcp]]).

1. Audit the site's SEO / index state before touching anything

// tool: website_pages   (no args → the tenant's one website)
{}
// → { website:{id,name,domain},
//     pages:[{id,url,name,view_id,published,indexed,in_menu,website_specific}], _count }

Then pull the SEO columns the page tool doesn't return, in one read:

// tool: get  (generic, on website.page)
{ "model": "website.page", "ids": [/* all page ids */],
  "fields": ["url","name","website_published","website_indexed","date_publish",
             "is_seo_optimized","website_meta_title","website_meta_description",
             "website_meta_keywords","website_meta_og_img","seo_name","visibility"] }

Read is_seo_optimized per page — false means the page is missing a title or description (the two the compute checks). That's your backfill worklist. Two live signals worth spotting on connectdomain right now (13 pages, 5 published, 11 website_indexed): pages that are website_indexed:true but website_published:false (drafts still marked crawlable — harmless while unpublished, but they'll appear in /sitemap.xml the moment they go live, so decide each deliberately), and two rows with the same url (/ often exists as both the module view and a COW fork — edit/set SEO on the site-specific one).

2. Set / backfill a page's SEO metadata

Write the delegated fields on the page (they proxy to its view — the correct path):

// tool: update  (on website.page)
{ "model": "website.page", "ids": [<page_id>],
  "values": {
    "website_meta_title": "Pricing — Connect Domain",                 // <title> + og:title
    "website_meta_description": "Bring-your-own-domain onboarding from $0. Automatic DNS, SSL and edge.",
    "website_meta_keywords": "custom domain, DNS, SSL, Entri alternative",
    "seo_name": "pricing",                                            // clean canonical slug
    "website_meta_og_img": "/web/image/website/1/social_default_image" // or a page-specific image URL
  } }
// is_seo_optimized flips true once title+description are both set (computed; read to confirm)

Keep titles ~50-60 chars and descriptions ~150-160; the rendered <title>/metas come from here and override any <title> in the arch — never hand-roll <meta> in the QWeb. website_meta_og_img expects an image URL/path (a /web/image/... route or a static asset), NOT a binary. Leave it unset to fall back to the site's social_default_image, then the company logo.

Backfill across website.page (whole site, one pass):

1. Read all pages (recipe 1) with is_seo_optimized, website_published, name, url.
2. Filter to is_seo_optimized == false AND website_published == true (fix the live/indexable pages first; drafts later).
3. For each, derive a title from page.name (mirror the tenant's existing pattern — several pages already carry "{Name} | Connect Domain") and a one-line description from the page's purpose; set seo_name where the URL isn't already a clean slug.
4. update in a loop, then re-run recipe 1 and confirm every published page now reads is_seo_optimized:true.

3. Control sitemap membership + publish + schedule

// publish + include in /sitemap.xml            // tool: website_publish
{ "url": "/pricing", "published": true, "indexed": true }
// keep live but REMOVE from the sitemap (e.g. a thin/duplicate page)
{ "url": "/welcome", "published": true, "indexed": false }

website_published = live vs draft (computed on the page); website_indexed = whether it appears in /sitemap.xml / is offered to crawlers (stored on the page). They're independent — a page can be live but out of the sitemap, or a draft flagged indexed. Because website_indexed is a stored field, a raw update on website.page also works:

{ "model": "website.page", "ids": [<page_id>], "values": { "website_indexed": false } }

To schedule a future go-live, set date_publish and leave it published-forward:

// tool: update  (on website.page)
{ "model": "website.page", "ids": [<page_id>],
  "values": { "date_publish": "2026-08-01 13:00:00", "website_published": true } }

4. Wire the site's social profiles (feeds JSON-LD sameAs) + default OG image

Set the URL fields on the website record — the seven in _SOCIAL_FIELDS become schema.org sameAs on every page's Organization block (an AI citation/authority signal):

// tool: update  (on website)
{ "model": "website", "ids": [1],
  "values": {
    "social_linkedin": "https://www.linkedin.com/company/connectdomain",
    "social_twitter":  "https://x.com/connectdomain",
    "social_github":   "https://github.com/ever-just",
    "social_youtube":  "https://www.youtube.com/@connectdomain",
    "social_tiktok":   "https://www.tiktok.com/@connectdomain"
  } }

facebook / linkedin / twitter / github / instagram / youtube / tiktok all reach sameAs; social_discord does not (Pitfall 7). Each value must start with http or the JSON-LD generator skips it. Set the default social share card (used when a page has no website_meta_og_img): social_default_image is a binary — write a base64 image string to it via update on website (or set it in backend Website settings); has_social_default_image then reads true. Verify JSON-LD after: fetch a public page and confirm two application/ld+json blocks with your sameAs (the generator swallows errors silently, so "no error thrown" is not proof).

5. Edit robots.txt and verify Google Search Console

robots_txt is an html field on website (a per-site override of the default):

// tool: update  (on website)
{ "model": "website", "ids": [1],
  "values": {
    "robots_txt": "User-agent: *\nAllow: /\nSitemap: https://connectdomain.app/sitemap.xml" }
}

Keep the Sitemap: line pointing at the site's real /sitemap.xml. Don't Disallow: / a live public site. Verify Google Search Console (URL-prefix HTML-tag method) by setting its token — Odoo renders it as the verification <meta> in the head:

{ "model": "website", "ids": [1],
  "values": { "google_search_console": "<the token from Search Console>" } }

(Domain-property verification instead uses a registrar DNS TXT record — that's [[godaddy-api]] / Route 53, not this field.)

6. Enable IndexNow (Bing/Yandex → also feeds ChatGPT Search) — CONFIRM-GATED

IndexNow is inert until you (a) set the key config param and (b) switch on the cron. The key file at /everjust-indexnow-key.txt returns 404 until the key exists, then serves it.

// (a) set the key — any string; it's the shared secret AND the key-file body.
// tool: create on ir.config_parameter (or update if it already exists) — STRUCTURAL → confirm:true
{ "model": "ir.config_parameter",
  "values": { "key": "everjust_visibility.indexnow_key", "value": "<32-64 char key>" },
  "confirm": true }

// (b) enable the daily submission cron (ships active=False) — STRUCTURAL → confirm:true
// xmlid everjust_visibility.ir_cron_everjust_indexnow (id 35 here); code = model._everjust_indexnow_cron()
// tool: update on ir.cron
{ "model": "ir.cron", "ids": [<cron_id>], "values": { "active": true }, "confirm": true }

Once both are set, /everjust-indexnow-key.txt returns the key (confirm by fetching it), and the cron calls _everjust_indexnow_cron() daily → for each website _everjust_indexnow_submit() POSTs _everjust_published_urls() to https://api.indexnow.org/indexnow. To submit once immediately without waiting for the cron, call the submit method:

// tool: call  (non-read → confirm:true)
{ "model": "website", "ids": [1], "method": "_everjust_indexnow_submit", "confirm": true }
// → the HTTP status int (200/202) on success; False if no key is set or the POST failed

IndexNow reaches Bing + Yandex, NOT Google — Google ignores it; Google discovery is via the sitemap + Search Console. Don't promise "instant Google indexing."

Pitfalls

1. SEO metadata stores on the page's VIEW, not the page table. website_meta_* / seo_name on website.page are related='view_id.…', store=False (writable). Set them via update on the website.page row (correct, COW-safe path). Never write website_meta_* onto a raw module ir.ui.view (empty website_id) — that mutates the shipped view for every site and doesn't COW, the same trap as editing arch ([[everjust-website]] / [[everjust-platform]]).

2. Set the title/description here, not in the QWeb <head>. website.layout renders <title> and metas from website_meta_*, overriding any <title> in the arch. Don't hand-roll <meta> tags in the page body — they'll be shadowed. (Content edits are [[everjust-website]] / [[everjust-website-snippets]]; metadata is this skill.)

3. website_published and website_indexed are independent — and shaped differently: website_published is computed (publish state); website_indexed is stored on the page (sitemap membership). Publishing a page does not add it to the sitemap-intent you want, and marking website_indexed doesn't publish it. On this tenant 11 pages are indexed but only 5 published — decide both per page (recipe 3), and beware drafts marked crawlable that will surface in /sitemap.xml the instant they publish.

4. is_seo_optimized is computed and read-only. It flips true only when BOTH title and description are set. Don't try to write it — write the two fields and re-read it to confirm. Use it as your backfill checklist (recipe 2).

5. og_img is a URL/path; social_default_image IS a binary. Per-page website_meta_og_img takes an image URL (/web/image/... or static). The site-wide fallback website.social_default_image is a binary (base64). Different field types, different write shapes. The fallback chain is page og_img → site social_default_image → company logo.

6. You don't author the JSON-LD — you feed it. The Organization/WebSite structured data on every page is generated by website._everjust_jsonld() from the website record (company name/email/logo + social sameAs). To change it, change those fields — do NOT inject your own <script type="application/ld+json"> into a page arch (you'd get duplicate/competing blocks). The generator swallows every error and emits empty markup, so a bad social value fails silently — always verify by fetching a page and grepping application/ld+json.

7. Seven social fields reach sameAs; social_discord does NOT. The addon's _SOCIAL_FIELDS is facebook, linkedin, twitter, github, instagram, youtube, tiktok. social_discord exists as a field on website but is NOT in that list, so it never appears in the JSON-LD sameAs — set it if you want, but it won't be a citation signal. Values must startswith("http"). (If a task genuinely needs Discord in sameAs, that's an addon-code change to _SOCIAL_FIELDS in everjust_visibility, not an MCP write — flag it; see [[everjust-website-snippets]] for the "author QWeb, flag py/JS" boundary.)

8. IndexNow is Bing/Yandex-only, confirm-gated, and OFF by default. The key config param, the cron toggle, AND the one-shot _everjust_indexnow_submit call each need confirm:true (structural models / non-read call), and the cron ships active=False, so nothing submits until you explicitly enable BOTH the key and the cron. It never reaches Google. Setting the key also starts serving /everjust-indexnow-key.txt (was 404) — that's expected. Don't enable it on a site whose /sitemap.xml is wrong or whose pages aren't ready to be crawled.

9. everjust_visibility does NOT override robots.txt — Odoo's website.robots_txt does. The addon deliberately ships no robots override (the AI-crawler Content-Signals override is an un-shipped follow-up); its extra routes (/llms.txt, /.well-known/security.txt, /everjust-indexnow-key.txt) are all NEW and sitemap=False. To change robots, edit website.robots_txt (recipe 5); don't look for an addon route. Keep the Sitemap: line and don't Disallow: / a live public site.

10. The public site (and thus all of this) only matters when everjust.public_website == '1'. If / bounces to /odoo the site is app-only — SEO/sitemap/JSON-LD render for nobody. That gate is an ir.config_parameter (confirm:true to write) and flipping it is platform-ops, not SEO work ([[everjust-platform]]). On the connectdomain tenant it's '1' (public site live).

11. You act with YOUR role, everything scoped by website_id. SEO writes need admin / website-designer (else AccessError), run as the connected user, and are audit-logged. Odoo 19 note: the users' group field is group_ids, not groups_id — check your role via group_ids if you introspect. One website per tenant (id 1 here) — confirm you're on the right tenant DB before a backfill pass (see [[everjust-agent-mcp]] / [[everjust-platform]]).

See also

• [[everjust-website]] — edit page CONTENT/layout (Tailwind-QWeb arch, COW, publish/menu/ redirect). Its SEO recipe #6 is a subset; this skill is the full discoverability layer.
• [[everjust-website-snippets]] — the on-brand s_cd_* snippets + html_builder options and the "wrap-don't-rewrite" method for the marketing ports.
• [[everjust-agent-mcp]] — connect to the tenant; the generic search/get/create/ update/call/describe_model + website_* tool surface and the confirm gates.
• [[everjust-platform]] — COW, the /odoo debrand, everjust.public_website, re-sweep on -u, structural-model confirm gates, never self-escalate, Odoo-19 group_ids.
• [[godaddy-api]] — registrar DNS (domain-property GSC verification, the site's own domain), a separate system from these fields.