Site Architecture Checklist: Build for Crawl, UX and AI Search

What site architecture controls

Site architecture is the system connecting a website’s pages, URLs, menus, directories, taxonomies and internal links. Information architecture focuses on how people understand and navigate information. Technical architecture adds crawl paths, status codes, canonicals, rendering and indexation controls. Effective SEO work requires both.

Google says it primarily discovers pages through links. It recommends crawlable HTML links, descriptive anchor text and a route from a known page to every page that matters. XML sitemaps can supplement discovery, particularly on large or complex sites, but they do not repair an isolated page or communicate its importance as effectively as useful internal links.

Architecture is not a standalone ranking factor. Its value comes from making content discoverable, clarifying page relationships, distributing internal signals and helping users complete tasks. A shallow structure can still fail if it creates irrelevant menus, duplicates or hundreds of undifferentiated links.

Terms to know

• Crawl depth: The number of link steps between an entry point and a URL.
• Orphan page: A page with no discovered internal inlinks.
• Hub page: A navigational or editorial page connecting a topic and its supporting resources.
• Canonical URL: The preferred representative for duplicate or substantially similar pages.
• Internal PageRank: A practical description of authority distributed through internal links, not a metric exposed by Google.

Choose the right architecture model

Most mature sites use a hybrid architecture. The useful question is not which diagram looks cleanest, but which model gives users predictable paths while keeping valuable inventory discoverable and duplicate inventory controlled.

Do not treat three clicks from the homepage as an absolute rule. Depth should reflect importance and expected demand. A legal archive may reasonably sit deeper than a revenue category, while a valuable long-tail page can remain accessible through its category, related items and contextual content.

The site architecture checklist

1. Inventory all known URLs. Combine crawler data, analytics, Search Console, XML sitemaps, content systems and server logs.
2. Assign a purpose. Label each page by topic, search intent, audience, funnel role, owner and conversion value.
3. Choose one preferred URL. Standardize protocol, host, casing, trailing slash behavior and parameter handling.
4. Build a user-centered taxonomy. Use labels people recognize, not internal department terminology.
5. Map hubs and destinations. Give every important page a logical parent or hub without preventing relevant cross-links.
6. Verify crawlable links. Use normal HTML anchors with usable destinations, including in rendered mobile experiences.
7. Write descriptive anchors. Explain the destination naturally instead of repeating generic phrases such as click here.
8. Find orphans and weakly linked pages. Repair valuable pages, merge duplicates and retire inventory with no continuing purpose.
9. Control filters and search pages. Decide which combinations deserve stable, indexable landing pages and which should remain crawl-controlled or non-indexable.
10. Align canonicals, redirects and sitemaps. Sitemap URLs should be canonical, indexable and return successful responses.
11. Test templates. Check menus, breadcrumbs, pagination, related modules and JavaScript rendering across devices.
12. Establish monitoring. Track depth, inlinks, indexation, crawl activity, traffic, conversions and architecture-related errors over time.

Design an internal-link graph, not a collection of menus

Global navigation establishes broad hierarchy, but contextual links explain specific relationships. Create a page graph containing each URL’s inlinks, outlinks, depth, status, canonical target, indexability, topic, organic demand and business value. This reveals where architecture differs from the intended sitemap.

Prioritize links with a value-gap rule: a page deserves attention when it has high user or business value, strong topical relevance and fewer useful inlinks than comparable pages. Link to it from authoritative, relevant pages where the destination genuinely advances the reader’s task. Homepage links are not the only solution.

Hub-and-spoke structures work when hubs help users compare, choose or learn. Spokes should link back to the hub and laterally to relevant follow-up material. Cross-topic links are appropriate when entities or tasks overlap. Rigid silos can suppress useful connections and create dead ends.

Use varied but accurate anchors, and avoid sitewide keyword-heavy links inserted only to manipulate signals. The 2025 HTTP Archive dataset found medians of 43 same-site links on desktop pages and 39 on mobile, with 90th percentiles of 174 and 161. These are descriptive benchmarks, not recommended targets. Link count should follow utility, template complexity and inventory.

Control crawling, rendering and indexation

Architecture problems often appear as indexation problems. Start by separating discovery, crawling, rendering, canonical selection and indexing. A URL can appear in a sitemap yet remain weakly connected, duplicate, blocked from rendering or judged unsuitable for indexing.

• Facets: Permit only combinations with distinct demand, useful inventory and maintainable content. Prevent effectively infinite sorting, calendar or filter spaces.
• Canonicals: Use them consistently, but do not expect a canonical hint to compensate for contradictory internal links and sitemap entries.
• Redirects: Send retired URLs to a close replacement. Do not redirect unrelated pages to a category or homepage merely to preserve signals.
• JavaScript: Confirm important links and content exist in rendered HTML. Avoid interactions that require clicks, scrolling or client state before a crawler can obtain a destination.
• Pagination: Give component pages crawlable URLs and links. Do not rely exclusively on infinite scroll.

Google says focused crawl-budget work is primarily relevant to very large or rapidly changing sites, including sites around 1 million or more unique pages, sites with roughly 10,000 or more rapidly changing pages, or sites with substantial discovered but currently not indexed inventory. Smaller sites should usually fix broken paths, duplicates and content quality before pursuing elaborate crawl-budget tactics.

Build topical architecture around complete user journeys

Start with entities, problems and tasks, then map query fanout: definitions, comparisons, prerequisites, costs, alternatives, implementation steps and troubleshooting questions. A service page may connect to a comparison, case study, pricing explanation, technical guide and relevant location page. The links should represent meaningful relationships, not a quota.

Use content consolidation when several pages compete for the same intent without serving distinct needs. Merge the strongest material, redirect retired URLs and update internal links to the surviving page. For decaying content, review query loss, outdated claims, broken pathways and changed intent before adding words.

Architecture can also create natural link demand. Original datasets, statistics pages, comparison assets, calculators and expert contribution programs deserve prominent hub placement and durable URLs. Support these assets with digital PR, link-intersect research and outreach to publishers that already reference the subject. Reclaim accurate unlinked brand mentions when a link would help readers verify the source.

Architecture for AI Overviews, Copilot and ChatGPT

Google states that supporting links in AI Overviews and AI Mode must be indexed and eligible to appear in ordinary Search. Crawlable content and internal findability therefore remain foundational. Architecture cannot guarantee selection, but it can make the right passage easier to discover and place in context.

Create pages with answer-first passages, explicit entity relationships, sourced numerical facts, definitions, comparisons and ordered procedures. Connect supporting evidence to the page that synthesizes it. Keep citations near the claims they support and ensure visible content agrees with structured data.

Measure two different outcomes: citation selection, whether a system chooses the page as a source, and citation absorption, whether the answer uses the page’s facts or framing. A 2026 GEO research framework examined 602 prompts, 21,143 search-layer citations, 18,151 fetched pages and 72 page-level features using this distinction. Other emerging research suggests conventional Google visibility can predict AI citation better than page-only features, but results vary by platform and intent. These findings are useful directions, not guarantees.

Diagnose architecture failures with evidence

Do not infer an architecture problem from rankings alone. Compare crawl data, internal graphs, server logs, index coverage, analytics and page purpose. Use the following triage framework.

Core KPIs include the percentage of valuable URLs indexed, orphan count, median depth by page class, internal inlinks to priority pages, noncanonical internal links, successful crawler hits, crawl spent on unwanted patterns, organic entrances, assisted conversions and task completion. Compare by template and business segment, not only sitewide averages.

Implementation sequence and migration safeguards

1. Baseline: Export current URLs, links, rankings, traffic, conversions, canonicals, sitemaps and log activity.
2. Classify: Mark each URL keep, improve, merge, redirect, noindex or remove, with an owner and reason.
3. Prototype: Test taxonomy, labels, menus and user paths before changing production URLs.
4. Map: Define old-to-new redirects, canonical destinations, breadcrumb paths and internal-link updates.
5. Stage: Crawl rendered staging pages while preventing the environment from entering search indexes.
6. Launch: Apply direct redirects, update internal links and publish clean canonical sitemaps.
7. Validate: Test critical templates, top landing pages and representative long-tail pages immediately.
8. Monitor: Review errors, logs, indexation, rankings and conversions daily at first, then weekly as conditions stabilize.

Avoid changing domains, URL paths, navigation, templates and content simultaneously unless necessary. Smaller releases make attribution and rollback easier. Keep redirects active long enough for users, crawlers and external links, and do not create chains when a direct destination is available.

Proven findings, practitioner consensus and uncertainty

Supported by official guidance

Google documents that links are central to discovery, important pages should be reachable from findable pages, anchors should be relevant, and logical organization can improve understanding. Google also analyzes site link structure for sitelinks. Indexed, search-eligible pages are required for supporting links in its AI search features.

Practitioner consensus

Experienced practitioners commonly favor descriptive URLs, useful hubs, contextual links and orphan remediation. Community discussions also warn against rigid silo rules. Reddit reports describe ranking or crawl improvements after internal-link restructuring without new backlinks, but these are uncontrolled anecdotes and cannot establish causation.

Still uncertain

No universal depth, internal-link count or hub size guarantees rankings. Early graph-embedding research, including WebKnoGraph, may improve recommendation workflows but is not proof of ranking impact. Cross-platform AI citation research is evolving, and a structure that supports one answer system may perform differently by query, platform and intent.

Risk and reward: Automated link suggestions can improve coverage on large sites, but unsupervised deployment can create irrelevant anchors and excessive template links. Controlled testing is safer: release changes to a defined page group, record the graph change, monitor crawling and performance, and retain a comparable control group where practical.

Selecting tools or an architecture partner

A small site can begin with a crawler, Search Console and analytics. Larger database-driven sites may need log-file processing, data warehouse joins, automated URL-pattern classification and graph visualization. Evaluate tools by rendered crawling, canonical analysis, parameter handling, custom extraction, API access, scheduling and export quality rather than by the size of a generic feature list.

When hiring a consultant or agency, ask for a URL-level diagnostic, an implementation sequence and measurable acceptance criteria. Strong recommendations should distinguish navigation, indexation and content problems, show the affected page classes, and explain how developers will verify the fix. Be cautious when a proposal promises rankings from a flat structure, recommends linking every page from the homepage, or treats XML sitemaps as a substitute for internal links.

The final deliverable should be implementable: taxonomy rules, page-type specifications, redirect maps, canonical rules, facet policies, link-module requirements, QA tests, dashboards and ownership. Architecture produces durable value only when editorial, design, product and engineering teams follow the same rules.