Tiers two through five

Lesson 12 of 50Module 3 · Reading the link graph6 min read

By Bill Hartzer · Hartzer Consulting · 25+ years in search, domains and expert witness work

The method most people are taught, and why it is the slow one

The conventional way to read upstream is by hand. Open your backlink report, pick a referring domain, run it through the tool as if it were the target, look at its referring domains, pick one of those, repeat. Five lookups gets you to tier 5.

It works, and I taught it for years, but it has a flaw that took me a long time to take seriously: a hand-walked chain tells you about one path. You picked the domains. If you picked the interesting-looking ones you have confirmed your own suspicion, and if you picked at random you have a sample of one per tier. Either way you are describing a thread and calling it a fabric.

Majestic's Link Graph tiers the whole profile natively. You load a domain, you get a force-directed picture of its graph with a tier slider under it, and moving the slider re-selects every node at that depth at once. Tier 2 is not one domain's referrers — it is all of them. That is a different question with a much better answer, and it is why this lesson is built around the slider rather than around the chain walk.

A detail worth knowing: you can filter a backlink report first and then build the graph from the filtered set. Filter to follow links above a Trust Flow threshold, build the graph from that, and you are looking at the structure behind the links that actually matter rather than behind everything a scraper ever produced.

The one number to read at every tier

When you select a tier, the footer gives you three figures: domains selected, links between them, and links per domain. That third number is the most diagnostic single figure in this module, and almost nobody reads it.

It measures how densely the layer is wired. A layer of independent sites that happen to link onward produces a low, irregular ratio. A layer built to link to itself produces a high one. A layer of purchases made one at a time produces exactly 1.0.

Here are two complete tier walks, captured on the same day:

The clean profile runs between 1.35 and 1.96 at every depth. It wobbles, it does not spike, and there is no tier where the population is noticeably wired to itself. That irregular-but-flat pattern is what accumulation looks like.

Now read the other column.

What exactly 1.0 means, and what 4.5 means

Tier 1 of the manufactured profile is 30 domains sending exactly 30 links. Every referring domain sends precisely one link, and not one of them sends two.

Think about how a real link happens. Somebody references you in a post; six months later they reference you again; the archive page carries both; the category page carries the excerpt; the sidebar carries the recent-posts widget. Real linking is lumpy, because real relationships produce more than one link. A perfectly flat 1.0 across an entire tier is what buying links one at a time looks like — one placement, one invoice, one link, next site. It is the most compact evidence of purchase I know of, and it is a footer statistic.

Then tier 3 of that same profile: 8 domains, 36 links, 4.5 links per domain — more than three times the density of anything in the clean profile, on the smallest population in the table. Eight sites are not sending 36 links outward at random. They are linking to each other. In the graph rendering you can see it directly, as two clusters with arrowheads at both ends of most edges. That is a link wheel: a small set of properties cross-linked to concentrate value before pushing it at the target.

The shape is worth naming. A manufactured profile is often thin at tier 1 and dense underneath, because the money was spent on the layer that feeds the layer you can see. An earned profile is the reverse — a wide, irregular tier 1 and no particular structure below it.

What else appears only at depth

Density is the headline, but four other things surface once you start moving the slider:

• Topical incoherence. The deep tiers of the manufactured profile mix a Dutch airport taxi service, NFL merchandise, a pharmacy spam site, stand mixers, dolls and acrylic products. Unrelated subjects, one wiring pattern.
• Metric inversion. At depth, Citation Flow above Trust Flow becomes the rule rather than the exception — domains scoring 30 for trust and 71 for citation, or 4 and 39. Volume without trust, which is what a network manufactures.
• Disconnection. Rendered as a picture, a healthy tier 3 is largely satellite clusters that happen to be three hops out. A manufactured one is a connected mesh. The eye catches that faster than any table.
• Cheap infrastructure. Junk registration clusters — long strings of .site, .click, .monster, .sbs, .website domains, and names built from SEO vocabulary — concentrate in deep tiers because that is where they are cheapest to place.

None of these is conclusive alone. Together, in one layer, they are.

You do not control your tier 2

Now the part that changes how you should react to everything above.

In my own site's tier 2 there is a run of domains — hotfri.com, mantaw.com, zomatt.com, zlutag.com, dotpim.com, yelpad.com, ylutag.com — with Trust Flow between 33 and 36 and Citation Flow between 48 and 50. Nonsense five-and-six-letter names, metrics identical to the point. That is one link network, sitting two hops from a site where no link was ever bought.

I did not put it there. I cannot remove it. Nobody asked me. Somebody pointed a network at sites that happen to link to me, for their own reasons, and my graph inherited it.

Three consequences follow, and they are the practical payoff of this lesson:

• A clean tier 1 proves nothing about your tier 2, and a dirty tier 2 proves nothing about you. Depth is diagnostic about how a profile was assembled; it is not a moral assessment of the site at tier 0.
• Do not disavow at depth. The disavow file operates on links pointing at you. There is no mechanism for repudiating a link between two other people's sites, and no reason to want one. If a vendor's audit recommends action on your tier 2, the vendor does not understand the tool.
• Judge what you chose. Ask of your tier 1: did I place this, buy it, ask for it, or earn it? Everything from tier 2 outward is context for reading that answer, not a finding about your conduct.

The same network, incidentally, appears in the manufactured domain's tier 4. Same domains, same metrics, two unrelated sites, different depths. Lesson 3.4 takes that apart properly.

How to run the walk, and when to stop

The routine takes about ten minutes per domain:

1. Load the domain in Site Explorer and open the Link Graph on the Fresh index.
2. Record the all-tier footer: domains, links, links per domain.
3. Move the slider to tier 1 and record the same three figures. Repeat for tiers 2, 3 and 4.
4. Put the four readings in one table beside a comparable competitor, because the numbers only mean something relative to something.
5. At each tier, scan the domain panel for runs of near-identical Trust Flow and Citation Flow pairs, and for naming and TLD clusters.
6. Export the tiers that looked wrong. The export button carries the current selection.

Stop at tier 4 unless something specific pulls you further. Beyond that the population is enormous, mostly disconnected, and shared with half the web — you are no longer describing this profile. And read every tier count as approximate: the interface itself sometimes reports 681 domains in the footer and 682 on the export button. Use these figures as shape, not as accountancy.