arrow_backDocumentation
functionTechnical deep dive

How RankForge calculates authority flow

No black box. This page explains the model behind the authority scores in your report — what it measures, the principles it's built on, and the things we deliberately don'tdo. If you're a technical SEO deciding whether to trust the numbers, this is for you.

What “authority” means here

RankForge's authority score is internal-only. It does not use backlinks, domain rating, or any third-party metric. It answers one question: given the way your own pages link to each other, how much internal ranking equity does each page accumulate?

This is the same principle Google's original PageRank described — a page that receives links from many well-linked pages is treated as more important — applied strictly to the internal link graph we crawl from your site. It tells you where your site's own structure concentrates equity, and which pages are starved of it.

Step 1 — Building the internal link graph

We crawl your site and extract every internal link. Before scoring, the graph is cleaned so the math reflects real structure, not crawl noise:

  • check_circleInternal links only. Outbound links to other domains never enter the authority graph. They're audited separately (external link / authority-leak analysis), not scored here.
  • check_circleCanonicals are honored. When a page declares a rel="canonical" tag, we ingest it under the canonical URL and consolidate duplicates, so canonicalized variants don't create phantom nodes that split authority.
  • check_circleLocale variants are suppressed. /en/page and /page (and similar locale duplicates) collapse to one canonical node, so authority isn't diluted across translations of the same page.
  • check_circleDuplicate edges collapse. If page A links to page B five times, that's one edge — repeated links don't multiply the authority passed.
  • check_circleSelf-links are ignored. A page linking to itself passes no authority.

Step 2 — The PageRank-style algorithm

Each page starts with an equal share of authority (1 / N, where N is the number of pages). On every iteration, each page distributes its current score evenly across its unique internal outlinks, discounted by the damping factor. Formally, the score of page p on each pass is:

PR(p) = (1 − d)/N  +  d × ( Σ PR(q)/L(q) + D/N )

d = damping factor (0.85)

N = total pages in the graph

Σ PR(q)/L(q) = sum over every page q that links to p, of q's score divided by its number of unique internal outlinks L(q)

D = total score held by dangling pages (pages with no internal outlinks), redistributed evenly across the whole graph

That D/N term matters: pages with zero internal outlinks (dead ends) would otherwise leak authority out of the system each iteration. We catch that pooled score and redistribute it, so the totals stay conserved and dead-end-heavy sites are modeled correctly.

Damping factor (d)

0.85

The standard PageRank damping search engines themselves popularised — most of a page's score flows through its links, with a small random-restart baseline.

Convergence

iterates until stable

Authority is propagated across the graph repeatedly until the scores settle, then frozen.

Graph edges

unique internal links

Multiple links from A→B collapse to a single edge. Self-links are ignored.

Scale

relative 0–100

Scores are relative to the rest of your site, not an absolute SEO metric.

After the iterations converge, scores are normalized to sum to 1 and multiplied by 100. So a page's authority score is always relative to the rest of your site— it answers “how does this page compare to its neighbours,” not “what is this page's absolute SEO value.” The run is fully deterministic: the same crawl always produces the same scores.

What we deliberately don't do

Being precise about the model's boundaries matters more than overclaiming. Three things are worth stating plainly:

info

The PageRank pass weights every unique internal link equally

A body link and a footer link to the same page count the same in the core authority math. We do classify link location (body, header, footer) and link quality (editorial vs. boilerplate) — but those signals drive the recommendation engine and link-quality scoring, not the raw PageRank value. We'd rather tell you that than pretend the model is more granular than it is.

info

Internal nofollow links are not silently obeyed

Rather than dropping internal nofollow links from the graph, we keep them and flag them in a dedicated nofollow audit as authority leakage. On most sites, internal nofollow is an accidental mistake starving a page of equity — surfacing it is more useful than quietly honoring it.

info

We read raw HTML — we don't execute JavaScript

Links that only exist after client-side JavaScript runs are not in the crawl, exactly like a raw-HTML pass from a search engine's first wave. We detect when a site depends heavily on client-side rendering and surface that as a risk (and cap the headline score), rather than silently under-reporting your structure.

Step 3 — From authority to a specific link to add

Authority scores are the input, not the output. A recommendation is generated when a strong page and an under-supported, topically related page aren't yet linked. In short:

  • arrow_forward

    Strong sources. Content-rich, indexable pages that carry enough authority — and enough body text — to host a natural editorial link.

  • arrow_forward

    Under-supported targets. Topically related pages punching below the authority their content deserves, so a new link actually changes their standing.

  • arrow_forward

    Topical relevance. We only pair pages that are genuinely about related things, so every suggestion is contextually sensible — not authority-driven noise.

  • arrow_forward

    Exact placement. For each pair we find the in-body sentence on the source page where the topic is already mentioned and suggest descriptive anchor text — so you can act on it in seconds.

How topical clusters are detected

The relevance layer above relies on grouping your pages into topical clusters based on the language of their content, then treating the best-internally-supported page in each cluster as its pillar. Clustering is what lets us tell the difference between “these two pages are genuinely about the same thing” and “these two pages just happen to share some common words” — so the links we recommend reinforce real topical relationships rather than coincidental keyword overlap.

See it run on your own site

The scores in this document are computed live for every audit. Run one on your site — free for the first 100 pages — and see exactly which internal links to add.