--- name: modu-ai/moai-ref-seo source: https://app.decimal.ai/s/modu-ai-moai-ref-seo@1/SKILL.md source_sha256: 51b30bcebc5a --- # Search Visibility Reference ## Target Agents - `manager-develop` — applies these rules while implementing pages, routes, and serialized metadata output - `/moai review` — pre-ship indexing review surface; equivalently available as a per-spawn `Agent(general-purpose)` with web-output review instructions ## Core Principle A page is indexed on what a machine can fetch and parse, not on what a person sees once everything has finished loading. Two failures dominate everything below: the same content reachable at more than one address, and a claim asserted in serialized data that nothing on the page corroborates. Every rule here is a specialization of one of those two. Figures that engines publish and then quietly revise — truncation widths, ranking weights, crawl allowances — are deliberately absent. Where a limit matters, this reference gives the decision rule and the measurement to take, so the guidance outlives the figure. ## Document Semantics The structure a parser reads before it reaches any content signal. Cheapest class of defect to introduce, and the cheapest to detect. | Rule | How to check it | Failure it prevents | |---|---|---| | Exactly one `h1` per document | Count `h1` elements in the rendered output of each page | Competing topic signals leave the subject ambiguous | | No skipped heading level | Walk the heading sequence; each descent moves by at most one level | Section nesting a parser cannot reconstruct | | Every image carries `alt` text | Each image element has a non-empty `alt` attribute, or an empty one paired with an explicit decorative marker | Meaning locked inside a binary the parser cannot open | | Anchor text names its destination | Read each link's text with the surrounding sentence hidden, then ask where it goes | Instruction-shaped link text carries no signal about the target | | Fragment targets resolve | For each in-page fragment link, confirm a matching identifier exists in the same rendered output | Navigation that silently lands nowhere | Derive `alt` text from what the image depicts, never from its file name. Where an image genuinely adds nothing a reader needs, mark it decorative on purpose instead of inventing a description for it. ## Identity and Canonical Address One resource, one address; everything else redirects to it. | Decision | Rule | |---|---| | Which address is canonical | Pick one form per resource and declare it. The declared value must equal the address actually served | | Trailing-slash variants | Choose one form and redirect the other permanently. Serving both splits one resource across two addresses | | Case, query parameters, tracking suffixes | Normalize before serving. A parameter that does not change the response must not mint a second address | | Parameterized routes | Build the declared value from the resolved parameters, never from a fixed string shared across the whole route | | Retiring an address | Redirect permanently from the old form and update the declaration in the same change | A declaration pointing at an address that redirects, errors, or serves different content is worse than no declaration at all: it actively steers the reader away from the page it appears on. ## Per-Page Metadata | Field | Rule | Recurring defect | |---|---|---| | `title` | Unique per page, composed from a page-specific part plus a stable identifier for the site | The scaffold's default survives to production on every route | | meta description | Unique per page, written for a person reading a result snippet | One description copied everywhere, or prose that describes nothing | | Indexing directive | Set a default for public pages, then narrow per page class. Authenticated areas and internal tools are excluded deliberately, not merely left unlinked | A private surface gets indexed because nothing declared otherwise | | Social preview fields | Present and absolute. A relative address in a field a third party fetches resolves against the wrong host | A share card that renders blank off-site | Length budgets are a decision rule, not a constant: draft the text, observe where the consuming surface truncates it, then cut to what you observed. Placeholder detection belongs in the pre-ship check — scan the emitted output for the scaffold's own default strings and fail when one survives. ## Structured Data Serialized JSON-LD states what a page *is*. Its one hard constraint is correspondence with the visible page. | Decision | Rule | |---|---| | Which types to emit | Derive them from the page's actual role — an organization, a product, a piece of software, a place, an article. Emit the site-level type everywhere and the page-specific type only where it applies | | Required fields | Each type has a small mandatory core. Emit that core completely rather than emitting several types partially | | Addresses inside the data | Absolute, and resolving. Relative values are discarded without a warning | | Several entities on one page | Emit one block holding all of them, give each a stable identifier, and cross-reference by identifier instead of nesting duplicate copies | | Placement | Keep each type on the page it describes. A type repeated site-wide dilutes rather than reinforces | **Mirroring rule** — every claim in the serialized data needs a visible counterpart on the same page. A name, a description, a rating, or a question-and-answer pair that exists only in the markup is an assertion the reader cannot check, and it is now scored as a negative signal rather than a neutral one. ## Entity Consistency Structured data identifies an entity. Consistency is what makes that identity hold across independent sources. | Surface | Requirement | |---|---| | Canonical name | One spelling, one casing, one suffix — the same in the serialized data, the page copy, and the headings | | Owned profile links | Restricted to accounts this entity itself administers, each confirmed to resolve rather than bounce to a sign-in wall | | Contact details | Name, address, and contact number identical across the serialized data, the visible page, and any third-party listings | | Omission over approximation | A field with no accurate value is left out. A partial or invented one fragments the identity it was meant to establish | Formatting differences that read as trivial to a person are distinct values to a matcher. Fix one format per field and hold it everywhere the entity appears. ## Host-Derived Crawl Artifacts `robots.txt` and `sitemap.xml` are output the site serves, not files carried between environments. | Artifact | Rule | |---|---| | `robots.txt` | Derive the absolute addresses it contains from the incoming request's host, so every environment serves a correct description of itself | | `sitemap.xml` | Enumerate the addresses actually served. An entry for a removed page spends crawl attempts and signals staleness | | Staying in sync | Adding or removing a route updates the enumeration in the same change. A hand-maintained list drifts within one iteration | | Per-entry hints | Freshness and weighting hints attached to entries are advisory and widely discounted. Correctness of the address set matters far more than the hints on it | Hardcoding one environment's host into either artifact is the defect that keeps recurring, because it passes review while being correct in exactly one place. ## Delivery Chokepoints | Concern | Rule | |---|---| | Response headers | Apply them at a single wrapping point every response passes through, error and redirect responses included. A header set only on the success path is missing exactly when a reader is most likely to be misled | | Rendering mode | Treat "what does a reader receive in the first response?" as a question the project answers explicitly. Content that appears only after client-side work is content a non-executing reader may never see | | Redirect chains | One hop to the canonical address. Each extra hop is another opportunity to lose the reader | | Third-party origins | Connection hints for origins the page depends on belong ahead of the resources that use them. Measure the effect on this project's own pages rather than assuming one | ## Common Rationalizations | Rationalization | Reality | |---|---| | "The framework already sets sensible metadata defaults" | A default is identical on every route by construction. Uniqueness is the entire purpose of the field, so it cannot be inherited. | | "Both address forms serve the same page, so it makes no difference" | Two addresses serving one resource are two resources to a crawler. Which one survives is guessed unless a declaration and a redirect settle it deliberately. | | "Serialized data may say more than the page shows — it is extra context" | Data with no visible counterpart is an unverifiable claim, and it is scored as one. | | "Heading structure and `alt` text belong to accessibility, not to indexing" | They are the parse structure a machine uses to segment and describe a page. Sharing a mechanism with accessibility is not a reason to defer them. | | "The address enumeration can be updated in a follow-up change" | A list maintained apart from the routes drifts immediately and then silently. Deriving it from the routes is the only version that stays true. | | "The staging host in the artifact is fine, we swap it at deploy" | A hardcoded host is right in one environment and wrong in every other, including the one that gets indexed. | ## Red Flags - Two or more `h1` elements on one page, or a heading sequence that skips a level - A canonical declaration that does not match the address actually serving the page - The same `title` or meta description reused across more than one route - An image with no `alt` attribute, or `alt` text that repeats the file name - Link text that reads as an instruction instead of naming a destination - An in-page fragment link with no matching identifier in the rendered output - `robots.txt` or `sitemap.xml` carrying a hardcoded host instead of one derived from the request - Serialized JSON-LD asserting a name, description, or offer with no visible counterpart - A relative address in a field a third party fetches — a share-preview image, a structured-data logo - Response headers applied on the success path only, leaving redirect and error responses uncovered ## Verification - [ ] Every rendered page has exactly one `h1` and a heading sequence with no skipped level — report the counts observed, not the intent - [ ] Each route's canonical declaration was fetched and returned that same page with no redirect in between - [ ] `title` and meta description were collected across all routes; no duplicates and no scaffold default strings remain - [ ] Every image element has an `alt` attribute, and decorative images are marked as such on purpose - [ ] Every in-page fragment link was resolved against the rendered output of the page carrying it - [ ] `robots.txt` and `sitemap.xml` were fetched from the deployed host and every address in them resolves — quote the address set observed - [ ] Serialized JSON-LD parses, its addresses are absolute, and each claim it makes has a visible counterpart quoted from the page - [ ] Response headers were observed on a success response, on a redirect response, and on an error response