Short answer: the impact of internal linking must be measured by discovery, owner coverage, wrong-target rate, path completion and crawlability before being linked to ranking or AI citations. Google recommends crawlable HTML links and contextual anchor text, but does not publish a universal topical authority score.

Baselines

Export the graph before the intervention:

  • URL;
  • types;
  • owner;
  • status;
  • canonical;
  • incoming links;
  • outgoing links;
  • anchor text;
  • click depth;
  • language;
  • product;
  • lifecycle status.

Population

Define priority pages: product, feature, docs, integrations, comparisons, pricing, research. Don't use the entire site as the denominator for each metric.

Metric 1: orphan rate

Numerator: eligible pages without incoming contextual links. Denominator: eligible pages.

Metric 2: near-orphan rate

Pages with a single weak link or just from a generic hub.

Metric 3: wrong-target rate

Links to 404, noindex, unnecessary redirect, wrong product or wrong language.

Metric 4: owner coverage

Priority pages and intents have a clear owner.

Metric 5: path completion

In a defined task, can the user get from explainer to docs or product detail without a dead end?

Metric 6: anchor quality

Use rubric: descriptive, generic, misleading. Do not create exact-match score.

It measures internal links that go through redirects. After migration, they can signal debt.

Links to deprecated pages or old versions.

Metric 9: Search observation

Track indexing, queries and landing pages. Don't attribute ranking changes to internal linking without a confounder log.

Metric 10: AI source observation

In the query set, note if the owner pages are cited and if the information is correct.

Observation window

Graph metrics can be measured immediately. Search/AI need longer periods and repeated observations.

False-attribution risks

  • content rewrites;
  • redirects;
  • product launch;
  • document migration;
  • backlinks;
  • campaign traffic;
  • seasonality;
  • Search updates;
  • AI model changes.

The denominators

Orphan rate uses eligible pages. Wrong-target rate uses rated links. Path completion uses tested tasks. Citation rate uses eligible observations.

Do not mix.

Before/after

Keep the same set of pages when possible. If the site migrates, version the baseline.

Comparison group

You can apply new rules to a subset of clusters and compare with similar ones, if there are no critical defects that need to be fixed immediately.

How do you treat Search lift

If treated pages increase, also check rewrites, backlinks or product demand. It reports association, not automatic causation.

How do you handle null result

If orphans decrease and path completion improves without Search lift, the intervention is still operationally useful.

How do you report

A mature dashboard separates graph health',user path', Search',AI source observation'.

Acceptance criteria

The measurement is auditable when:

  1. graph snapshot is versioned;
  2. the population is clear;
  3. the denominators are clear;
  4. the owners are documented;
  5. the change log is kept;
  6. external outcomes are separate;
  7. confounders are noted;
  8. raw links can be re-audited;
  9. metrics have explicit thresholds;
  10. there is no invented score presented as a standard.

How do you measure overconcentration

Some sites send a very high proportion of links to the homepage or a few hubs. It measures the distribution of incoming links across page types, but doesn't make uniformity a goal. Some pages need to receive more links; what is important is whether the graph supports the priority tasks.

How do you handle automatic modules

Separate editorial links from those generated by related-content modules. If a module produces the majority of the graph, measure the quality of recommendations on a sample and the rates of wrong-product/wrong-language targets. This metric can identify problems that the raw link count hides.

How to measure after consolidations

After merge and redirects, track if the links have been updated to the final destination or just go through the redirect. A valid redirect keeps navigation functional, but can hide operational debt and old recommendation rules.

How do you report uncertainty

If the Search lift appears in a major release or campaign interval, it marks the NOT_PROVEN assignment. It is not a failure of the experiment; it is a data limitation. A credible report preserves this state rather than forcing a causal story.

How do you treat click depth without dogma

There is no magic number of clicks for any page. It uses click depth as a signal for priority pages and checks that the path is logical. A rarely used page can be deeper without breaking, while a critical setup page may require direct access from feature or docs.

Separate global navigation from contextual links. A URL can have many incoming links in the header and still be poorly integrated editorially. For topic relationships, measure links in relevant content and modules separately.

Stop criterion

Close intervention when priority pages are discoverable, wrong targets are rare and new releases do not reintroduce orphans. Search lift is not a prerequisite for a healthier graph.

Final note

Keep the same population on replay.

Claim ledger

  • FACT/EVIDENCE: Google recommends crawlable links and contextual anchor text.
  • PRACTITIONER GUIDANCE: internal-link impact must be measured by graph health and user paths.
  • INFERENCE: Clearer graphs can help discovery and contextual understanding.
  • NOT PROVEN: a universal topical authority score or direct effect on AI citations.

Conclusion

In B2B SaaS, internal linking produces measurable results before ranking: less isolated pages, correct destinations and better paths. This is solid evidence. Search and AI are outcomes that must be interpreted with change log and limitations.

Sources reviewed

Related insights