Short answer: for publishers, internal linking must be treated as editorial infrastructure. Google recommends crawlable links and contextual anchor text, but does not publish a "topical authority score". The healthy playbook defines intent owners, connects news to explainers, cleans up redirects and recommendation modules, and stops expansion when the graph no longer adds useful paths.

Precondition 1: content types

Separate breaking news, analysis, evergreen explainer, review, guide, category and tag. Each type has a different role.

If all pages are treated as "article", the linking becomes arbitrary.

Precondition 2: canonical intent

For recurring questions, designate owners. The news item describes the event. The explainer owns the stable context. The guide can hold the practical task.

This taxonomy needs to be versioned.

What do you change on the page

When an article mentions a complex concept, it links to the canonical explainer, not the newest article on the topic.

After the reader understands the context, it provides the next step: methodology, timeline, comparison, or primary source.

3. Anchor text

Describe the destination. It avoids both "click here" and repetitive exact-match.

Update redirects and retired pages. A valid 301 does not mean that the internal link is optimal.

What are you changing in the site?

5. Hubs with an editorial role

A hub doesn't have to be just a list of articles. Explain the subject structure and links to the real owners.

6. Controlled taxonomies

Automatically created tag pages can compete with categories or explainers. Decide which taxonomies are indexable and what their role is.

7. Recommendation modules

Use similarity as a signal for candidate links, not as final authority. An editor or ownership rule must be able to exclude redundant results.

8. Language/region routing

If there is a local variant, don't automatically send the user to EN just because that URL has more backlinks.

What do you change in distribution

9. Homepage and Sections

For big events, link the news to the stable explainer and vice versa. This creates a path between recency and context.

10. Newsletter

When the external distribution links to a snapshot article, it also includes a path to the canonical resource if the topic asks for context.

11. Social

Not every post has to submit to the same hub. It links to the page corresponding to the promoted task.

Distribution is not only traffic, but also a choice of the owner.

Operational workflow

  1. export the graph;
  2. classifies URLs by type and intent;
  3. detects orphans, redirects and wrong targets;
  4. define owners for priority topics;
  5. generate candidate links;
  6. review editorially;
  7. apply small batches;
  8. re-crawl;
  9. measure pathways;
  10. keep the change manifest.

QA at scale

Automate:

  • 404 targets;
  • redirect chains;
  • noindex targets;
  • canonical mismatch;
  • wrong-language targets;
  • orphan/near-orphan URLs;
  • anchor repetition.

Human review decides relevance and next step.

Metrics

Track orphan rate, wrong-target rate, redirect-link rate, path completion and contextual clicks. For Search, track indexing and queries. For AI, source citations on canonical owners can be noted separately.

Do not compress them into a single authority score.

Prioritization

P1: Link to 404/noindex or wrong language on important pages. P2: orphan owners and collisions. P3: anchor quality and recommendation noise. P4: editorial opportunities.

Rollback

Keep the manifest of each batch of links. If a pattern causes a regression, it retracts the changes of that batch without touching other editorial changes.

Don't do massive graph resets without evidence.

Acceptance criteria

A cluster passes the gate when:

  1. the intent owners are clear;
  2. new articles have incoming context;
  3. explainers are not near-orphan;
  4. important links are crawlable;
  5. old redirects are cleaned;
  6. recommendation modules do not predominantly send to duplicates;
  7. the languages/regions are correct;
  8. the manifest is versioned;
  9. QA can be rerun;
  10. new links continue the reader's task.

Stop criterion

It stops expanding when canonical pages have clear paths, new recommendations become predominantly redundant, and the graph remains stable between crawls. Link density is not objective.

How do you deal with recurring topics in news cycles

For recurring topics, keep an evergreen explainer and link new news to it when the context remains relevant. The explainer can refer back to the timeline of the most important events, but it doesn't need to be rewritten for every minor tidbit.

This model reduces duplication and provides a stable point for readers arriving from distribution.

How to check recommendation systems

Periodically sample automated modules and measure the proportion of redundant, stale, or wrong-role targets. If the algorithm repeats the same errors, fix the selection rule instead of manually correcting each page.

Stop condition

When the canonical owners have enough incoming links for discovery, wrong targets are rare and new suggestions are predominantly redundant, switch the graph to monitoring.

Re-audit threshold

Resume full audit after migration, taxonomy redesign or recommendation engine change. Otherwise, use incremental monitoring for 404, redirects, language mismatch and canonical owners who have lost incoming links.

Claim ledger

  • FACT/EVIDENCE: Google recommends crawlable HTML links and contextual anchor text.
  • FACT/EVIDENCE: links help with discovery and provide context.
  • PRACTITIONER GUIDANCE: publisher internal linking must be linked to content types and canonical intent.
  • NOT PROVEN: a universal topical authority score derived from the number of internal links.

Conclusion

Internal linking at publishers is a discipline of editing and architecture. When news, explainers, and guides have clear roles, the graph becomes more useful without being artificially dense. The playbook should optimize the routes, not the number of links.

Sources reviewed

Related insights