Link graph audits and prioritization
Audit the link graph and choose the next repair.
Read the directed relationships, check the destination and repair the missing reader route before adding more links.
Free learning from Semantec SEO.
Set the graph boundary before counting links.
An internal link graph is a directed model: pages are nodes and a link from one page to another is an edge. Direction matters. A link from an overview to a method does not prove that the method returns to the overview. An adjacency matrix records those directions in rows and columns, making missing routes easier to inspect than a long unsorted link list.
Define the approved URL inventory, crawl date, hostnames, page types and inclusion rules first. Decide if the audit covers the entire site, one topic cluster or a proposed release. Keep navigation, footer and contextual links distinguishable. Ten repeated menu links do not necessarily provide ten different reading paths. Decide if fragments, redirects, language versions, pagination and non-HTML resources belong in the model.
Normalize URLs carefully. Record the observed URL, final response and intended canonical owner rather than silently merging distinct pages because their titles resemble one another. Exclude a deleted page from the active graph only after recording its replacement or removal decision. Label an incomplete crawl as partial: a page with no incoming edge in that sample is not proven to be orphaned across the whole website.
Separate observed edges from routes the reader needs.
Keep two layers. The observed layer records links found in the acquired pages. The proposed layer records required, optional, blocked or unresolved routes. A zero in the observed matrix means no edge was found under the chosen scope; it does not mean a link should be added. The decision needs a reader task, an approved destination and a sensible point in the source text.
Ask what the reader needs next: a definition before a procedure, evidence after a claim, a comparison before choosing, or a return to the topic overview. A useful sibling link advances that task. Shared vocabulary alone is weak justification. A glossary explanation can remain free of an early sales jump when the reader has not reached that decision.
Check incoming routes, onward routes and recovery paths separately. An isolated detail page may need a hub introduction; a detail page already reached through its hub may still need a return route. A group of pages linking only to each other can form a closed loop. Those patterns are review signals whose consequences depend on the wider graph and visible context.
Read this fictional five-page matrix.
The example is a fictional learning cluster, not a crawl of Semantec SEO or a measured experiment. H is the overview, M the method, E a worked example, C a comparison and S a troubleshooting page. Rows are sources; columns are destinations. A 1 means a contextual link exists in the supplied fictional snapshot, 0 means none, and a dash excludes self-links. Header, footer and external links are outside this example.
| Source → destination | H overview | M method | E example | C comparison | S support |
|---|---|---|---|---|---|
| H overview | — | 1 | 0 | 1 | 0 |
| M method | 1 | — | 1 | 0 | 0 |
| E example | 0 | 0 | — | 1 | 0 |
| C comparison | 1 | 1 | 0 | — | 0 |
| S support | 0 | 1 | 0 | 0 | — |
Read H→M as 1: the overview links to the method. M→H is also 1, but that return edge is a separate observation. The S column contains only zeros, so no included page links to support. S does link out to M; an outgoing link does not solve its missing incoming route. The E page has an incoming route from M, so H→E being zero is not automatically a defect.
The snapshot has eight directed edges. That count describes this invented graph only. It says nothing about link prominence, destination quality, index status or user success. The matrix also cannot show if the overview’s link to the method is buried in an irrelevant sentence; the source passage still needs inspection.
Check the destination before proposing the repair.
The strongest proposed route can still fail if its destination is unavailable, misleading or owned by the wrong page. Inspect the intended URL and final destination, response state, redirects, canonical signals and access conditions. Confirm that the promised answer is present, current and readable. A successful HTTP response alone does not prove the page fulfills the anchor.
For a route intended for public search discovery, check crawl access and indexing instructions separately. Record a noindex instruction rather than treating the page as an ordinary indexable learning destination. A restricted support page may still be useful to signed-in customers, but its access expectation must be clear. A canonical pointing elsewhere calls for an ownership check before adding more references.
Google’s crawlable link documentation explains the normal anchor-with-href pattern and descriptive anchor context. Use a real link whose wording tells the reader what the destination supplies. That technical guidance does not prescribe this matrix, a universal link count or a ranking gain.
Repair the missing support route in dependency order.
In the fictional brief, readers who cannot complete the method need troubleshooting. The proposed first repair is M→S, placed after the method’s failure checks, with the anchor “Troubleshoot the method’s failed checks.” This addresses a specific task and makes S reachable from the existing H→M route. It is preferable to adding S to every page just to raise its incoming count.
First confirm S is the approved support owner and contains the relevant checks. Resolve any conflict with a duplicate support URL. Then check response, access, visible answer and final URL. Only after those dependencies pass should the editor insert M→S, check the rendered anchor and destination, and record the changed edge. If S is incomplete, repair S before sending readers there.
| Field | Decision | Evidence or next check |
|---|---|---|
| Observed gap | No included incoming route to S | All five observed source rows checked |
| Reader consequence | Method failure has no support continuation | Fictional brief requires troubleshooting |
| Proposed change | M→S after failure checks | Anchor promises the checks S contains |
| Dependency | S owner, content and access must pass | Do not publish link before destination repair |
| Acceptance | M→S becomes 1; H→M→S path exists | Reacquire source and follow final URL |
| Unmeasured effect | Possible easier recovery | No user test, crawl change or ranking gain claimed |
A later review may add S→H to restore orientation. Keep it as a distinct proposal with its own placement reason. The original S→M route already offers a return to the method, so the graph alone cannot prove an extra overview link is necessary.
Prioritize consequences and dependencies, then effort.
Start with links that send readers to a wrong, unsafe, restricted or unavailable destination. Next resolve blocked dependencies: unclear page ownership or missing destination content prevents a responsible insertion. Then address missing required routes to useful pages, recovery failures and excessive detours. Treat optional cross-links and wording improvements after those task blockers unless their reader consequence is more serious.
Record the reason, affected audience, evidence confidence, repair owner, dependency and expected effort. If the crawl is partial, uncertainty may justify acquisition before editing. A page with many incoming links can still deserve urgent repair when every anchor promises the wrong answer. A low link count alone does not establish priority.
MIRENA’s routing records can organize source, target, reason, anchor direction, owner and state. Numeric depth limits, link minimums or anchor ratios used in a particular plan are local review rules. They should not be presented as universal search requirements. A protected destination or unresolved duplicate remains held even if a priority score is high.
Measure the repair without inventing its effects.
After an authorized change, record what can be observed: the source now contains the intended href, the final destination matches, the edge exists in a fresh acquisition and the required path is reachable under the stated scope. Preserve the before and after matrix, acquisition dates, URL changes and remaining exceptions. Those checks establish the repair itself.
Reader tests, consented navigation analytics, support outcomes, crawl observations and search reporting answer different questions. Choose a relevant measure before attributing an effect. More clicks to support might mean better discovery or a newly confusing method. Search movement can reflect other changes and does not isolate this link as the cause.
For cluster design, return to the internal linking hub. The semantic SEO learning overview connects graph work with page ownership and content review. Reopen the audit when a destination changes, a page moves, a cluster expands or fresh evidence contradicts the route assumption.