Site Architecture Mistakes to Avoid

What site architecture includes

Site architecture is the organization of pages, URLs, navigation, taxonomies and internal links. It determines how users move through a site and how crawlers discover, interpret and prioritize its resources.

Common models include hierarchical structures for corporate sites, hub-and-spoke models for editorial libraries, database-driven structures for marketplaces, faceted systems for ecommerce and hybrids that combine these patterns. The correct model depends on inventory size, user tasks and how often pages change.

Important terms include crawl depth, the number of link steps from an entry page; orphan page, a URL with no discoverable internal link; hub page, a page organizing related resources; taxonomy, the classification system; and canonical URL, the preferred representative among duplicate or similar URLs.

Architecture is not a standalone ranking factor. Its value comes from improving discovery, indexation, topical relationships, internal signal distribution and usability.

The site architecture mistake matrix

This matrix is a triage tool, not a universal severity score. A deeply nested legal archive may be harmless, while a revenue page at the same depth can be an urgent problem.

Diagnostic framework: inventory, graph, classify and decide

1. Build a unified URL inventory

Combine crawl data, XML sitemaps, analytics landing pages, server logs where available, backlinks and indexation exports. Record status code, canonical target, indexability, template, topic, depth, inlinks, outlinks, organic traffic, conversions and business value.

2. Model the internal graph

Treat every URL as a node and each internal link as a directed edge. Find orphan pages, dead ends, hubs, excessive loops and priority pages with few relevant inlinks. A graph view is more revealing than a navigation screenshot because contextual body links also shape discovery.

3. Assign one action

• Keep and strengthen: unique page, valid intent and measurable value.
• Consolidate: overlapping pages compete for the same task or query family.
• Redirect: the page moved and has a clear replacement.
• Noindex: the page serves users but should not appear in search.
• Remove: the URL has no continuing user, search or operational purpose.

4. Prioritize by consequence

Score opportunities using business value, demand, current visibility, weak inlink count and implementation effort. Do not prioritize depth in isolation. A page one click from the homepage can still be poorly connected to its topic.

Mistake: confusing shallow architecture with good architecture

A shallow structure reduces link steps, but flattening every URL beneath the homepage can erase category meaning and overload navigation. The objective is a short, logical path, not the lowest possible depth number.

Start with user tasks and entities. A software company might organize resources around products, use cases, industries, integrations and support. Each durable concept can have a hub that defines the subject, links to important subtopics and receives links back from those pages. Cross-link between hubs when the relationship helps a user, rather than enforcing a rigid silo.

Map query fanout before publishing. A broad query often leads to definition, comparison, implementation, troubleshooting, cost and vendor-selection questions. Decide whether each need belongs on the hub, a supporting page or an existing page requiring expansion. This prevents thin spokes and duplicate intent.

Use descriptive, stable URLs where practical. Topical directories can communicate organization, but changing established URLs merely to add keywords usually creates more migration risk than benefit.

Mistake: treating menus and sitemaps as the entire internal-link strategy

Google says it primarily discovers pages through links, while sitemaps supplement discovery. Every important page should therefore be reachable from a findable page using a crawlable HTML link. Footer, menu and breadcrumb links help, but they cannot express every contextual relationship.

Add links where a reader would reasonably need the destination. Use anchors that concisely describe it, avoiding repeated generic phrases such as click here. Link from strong, relevant pages to important pages with weak internal support. Do not inject unrelated links simply because the source page receives traffic or backlinks.

Audit both incoming and outgoing links. Pages with no inlinks are orphans; pages with no useful onward path are dead ends. Broken links, redirecting internal links and chains waste crawl activity and create avoidable latency. Update links to final canonical destinations.

The 2025 HTTP Archive Web Almanac found median pages contained 43 same-site desktop links and 39 mobile links, with much higher counts at the 90th percentile. These are observational benchmarks, not recommended quotas. Relevance, placement and destination quality matter more than reaching a particular count.

Mistake: allowing duplicate and faceted URL growth

Filters, sorting, tracking parameters, internal search, calendars, tags and pagination can create enormous URL spaces. The mistake is not having facets. It is allowing every possible combination to become crawlable and indexable without evidence of distinct user demand.

Create a rules matrix for each URL pattern. Specify whether it may be linked internally, crawled, indexed, included in sitemaps and declared canonical. Valuable category combinations can become stable landing pages with unique inventory and useful content. Low-value combinations should not receive indexable links merely because the interface can generate them.

Canonical tags help identify preferred representatives, but they do not replace coherent linking and URL controls. Internally link to canonical URLs consistently. Avoid contradictory signals such as placing a noncanonical parameter URL in the sitemap while linking to it across navigation.

Crawl-budget work is primarily important for very large or rapidly changing properties. Google highlights sites around one million or more unique pages, sites with at least 10,000 rapidly changing pages and sites with substantial discovered but not indexed inventory. Smaller sites should first fix accessibility, duplication and quality problems.

Mistake: overlooking JavaScript, mobile and migration paths

A control that looks like a link to a user may not provide a dependable crawl path. Use ordinary anchor elements with resolvable URLs for important navigation. Test rendered HTML, not only the source or visual interface. Confirm that desktop and mobile experiences expose the same essential destinations.

During redesigns, create an old-to-new URL map before launch. Redirect each retired URL directly to its closest relevant replacement, update internal links, preserve canonical intent and regenerate sitemaps. Redirecting every old page to the homepage obscures meaning and creates a poor user experience.

1. Crawl and export the current site.
2. Freeze the URL and redirect map.
3. Test status codes, canonicals, robots directives and rendered links in staging.
4. Launch redirects and updated internal links together.
5. Submit fresh sitemaps and inspect representative templates.
6. Monitor logs, indexation, rankings and conversion paths.

Retain redirects as long as users, search engines or external links may request the old URLs. Avoid chains by updating both redirects and internal references to final destinations.

Measure architecture with search and business KPIs

No single metric proves that an architecture is healthy. Use a scorecard that connects structural changes to crawler behavior, search outcomes and user value.

• Discovery: orphan count, valid inlinks, crawl depth and sitemap-only URLs.
• Crawl behavior: requests by directory, status code, parameter pattern and bot type.
• Indexation: canonical selection, indexed-to-submitted ratio and excluded URL patterns.
• Internal graph: inlinks to priority pages, broken edges, redirecting links and hub coverage.
• Search: impressions, query diversity, nonbrand clicks, rankings and sitelink appearance.
• Business: assisted conversions, product discovery, engagement and revenue by landing-page family.

Compare cohorts rather than sitewide averages. For example, measure strengthened product pages against similar unchanged pages. Annotate releases and allow for recrawling. Log-file analysis is especially valuable on large sites because it shows requested URLs, not merely what a crawler simulation can reach.

Use controlled title or intent tests separately from architecture changes when possible. Otherwise, a traffic increase cannot be attributed confidently to the new linking structure.

Architecture for AI Overviews, Copilot and ChatGPT

AI retrieval does not eliminate technical SEO. Google states that supporting links in AI Overviews and AI Mode must be indexed and eligible for normal Search. Crawlable content, clear internal paths and canonical discipline remain prerequisites for consideration.

Build pages that answer a distinct task and contain extractable definitions, comparisons, steps, caveats and numerical facts. Connect related entities through useful hubs and contextual links. This can help systems retrieve the right page for a rewritten query or follow-up question, but it does not guarantee citation.

Recent research supports separating citation selection, whether a system chooses a page, from citation absorption, whether the answer uses information from it. A 2026 framework analyzed 602 prompts, 21,143 search-layer citations, 18,151 fetched pages and 72 page-level features. Another multi-platform study found Google ranking could predict AI citation better than page-only features, while platform and intent differences remained important.

Architecture should also support earned authority. Publish original datasets, statistics resources, expert contributions and genuinely useful comparison assets. Use link-intersect analysis and reclaim accurate unlinked brand mentions. Emerging research suggests third-party authoritative sources can receive preference in generative search, although this is not a universal rule.

What is proven, consensus-based or uncertain

Proven through official documentation

Google discovers pages through links, recommends crawlable anchors and relevant anchor text, and advises making important pages reachable from findable pages. Logical organization, descriptive URLs, canonicalization and sitemaps can support understanding and discovery. AI search eligibility still depends on ordinary search eligibility.

Strong practitioner consensus

Experienced practitioners commonly favor hubs, contextual links, orphan remediation and logical shallow paths. They also warn against rigid silos that prevent useful cross-topic linking. Search Engine Land, Semrush and Backlinko broadly align on these practices.

Anecdotal observations

Reddit contributors report ranking or crawl improvements after internal-link restructuring without new backlinks. These reports are uncontrolled and cannot establish causation. They are most useful as hypotheses for measured tests.

Still uncertain

There is no universal ideal depth, link count or architecture template. Early graph-embedding research, including WebKnoGraph, may improve link recommendations, but it is not ranking proof. The effect of architecture on AI citations also varies by platform, intent and external authority.

When buying tools, prioritize complete crawl exports, rendered-page testing, log analysis, graph visualization and integrations over proprietary health scores. Commercial tool roundups can identify candidates, but validate them against a representative crawl before committing.