Reading the Metrics on a Link Vendor's Screenshot

A vendor pitching a placement will usually lead with one number: an authority score in the eighties or nineties, cropped from a tool’s dashboard. That number is a domain-level estimate, and the thing you’d be buying is a page. Almost everything that determines whether the placement is worth anything lives in columns the screenshot doesn’t include.

This post is about reading such a screenshot as a data artefact — what the number can support, what it can’t, and which fields you’d need to see before the claim means anything. It is not about whether to buy the placement; that’s a strategy question and it belongs on a different site.

What a domain-level authority score can and can’t support

Every major link index publishes a 0–100 domain-level authority estimate. The names differ, the underlying graphs differ, and the exact computation is proprietary in each case — so anyone stating the formula as fact is guessing. What they share is structure: they’re logarithmic-ish estimates derived from the link graph the vendor has crawled, and they describe a whole host, not a URL.

Three consequences follow directly.

The score is not comparable across tools. A “92” in one index and a “92” in another are two different estimates of two different graphs. Domain Rating Versus Domain Authority covers why the two don’t convert.

Domain-level scores say nothing about the page. A high-authority host can serve a page with no inbound links, no internal links pointing at it, and no traffic. Page-level metrics exist precisely because of this gap — see Page-Level Authority Metrics — and they are the ones missing from almost every pitch.

Scores respond to link volume, and link volume can be manufactured. This is the part worth being blunt about. A score built from a crawled link graph will move if the graph moves, regardless of why it moved. A host with a large number of low-quality inbound links can carry a high score, which is exactly why the score is a poor proxy for editorial quality.

So a domain-level score in the nineties supports one claim: this host has accumulated a lot of inbound links according to one vendor’s crawl. It does not support “this is a strong site,” and it certainly doesn’t support “a link here will do something.”

The columns that aren’t in the screenshot

If you’re evaluating the data rather than the pitch, these are the fields that carry information — and their absence is itself a signal, because a vendor with good numbers shows them.

Organic traffic estimate for the host, and for the specific page. A host with a high authority score and a traffic estimate at or near zero is the classic shape of a link network: the graph is dense, but nobody arrives from search. Read the estimate with the caveats in Reading Organic Traffic Estimates — it’s modelled, not measured — but “essentially zero” and “substantial” are far enough apart that the model’s error bars don’t rescue the first case.

Referring domains, not backlinks. A total-backlinks figure counts links; a referring-domains figure counts sources. The ratio between them is one of the fastest reads available on a profile, and the two get conflated constantly: Referring Domains Versus Total Backlinks and What Counts as One Referring Domain cover why. A purely illustrative example of the shape you’re looking for: hypothetically, 200,000 backlinks from 300 referring domains is a very different profile from 200,000 backlinks from 40,000 referring domains, at the same authority score.

Whether the page that would carry the link is indexed. A link on an unindexed page is a link a search engine may never have processed. This is checkable, cheaply, and it’s covered in Whether the Linking Page Is Indexed.

Outbound link count on the page. A page carrying a large number of external links is distributing whatever it has across all of them, and a page whose outbound links are overwhelmingly commercial reads differently from one with a handful of contextual references. See Links on Page and Outbound Link Counts.

Anchor-text distribution for the host. A profile whose anchors are dominated by exact-match commercial phrases has a distribution no organic profile produces. Anchor Text Distribution as a Dataset covers how to read one.

The link attribute that would be applied. nofollow, sponsored and ugc change what the link is claimed to be. Vendors rarely volunteer this. Tools report it inconsistently, too: How Tools Count Nofollow, UGC and Sponsored Links.

Date fields. When the host’s links were first seen tells you whether the profile accumulated over years or over three weeks. What First Seen and Last Seen Mean explains what those timestamps do and don’t record.

Verifying a screenshot at all

A screenshot is an unverifiable artefact. It can be from a different domain, a different date, a subdomain rather than the host, or a URL-prefix scope rather than a domain scope — and cropping hides all of that. Before treating the number as data:

1. Get the domain in text, not in an image. Without the hostname you can’t check anything.
2. Pull the metrics yourself from whichever index you already pay for. If your number and theirs differ substantially, the interesting question is scope: host vs subdomain vs URL prefix produces wildly different figures for the same property.
3. Look at the host’s own pages. Not a metric — just read three of them. Topic coherence, whether the writing is machine-generated, whether the outbound links make editorial sense.
4. Check the page, not the domain. Page-level authority, indexation, outbound count.
5. Check the archive. A host whose history shows a topic change and a wholesale content replacement is a rebuilt expired domain, and its authority score is inherited from a site that no longer exists.

What to write down

If you’re keeping records — and you should be, because these evaluations are the input to any later claim about whether links did anything — capture the values and the date and vendor for each. Metrics move on recrawl, and a figure with no timestamp can’t be compared to anything later. Building a Link Report You Can Defend covers the record-keeping; When a Metric Moves but Nothing Changed covers why the same host reads differently next month.

The summary judgement: a cropped authority score is the least informative number in the entire dataset, which is presumably why it’s the one that gets screenshotted. Every field that would let you evaluate the placement is available and absent.