Domain Rating Versus Domain Authority

Domain Rating is Ahrefs’ 0–100 score for a domain’s inbound link profile. Domain Authority is Moz’s 0–100 score built to predict how a site ranks. They look interchangeable because they share a range and a vibe, and they are not: different crawls, different formulas, different design goals.

The short version is that you can compare a site to another site inside one tool, and you cannot compare a number in one tool to a number in another.

What each vendor documents

Ahrefs documents DR as a link-only score. It is computed from Ahrefs’ index of the link graph, it reflects the strength of a domain’s backlink profile, and Ahrefs describes the scale as logarithmic. It does not take traffic, content, or brand into account by design.

Moz documents DA differently. As of this writing Moz describes it as a machine-learning score trained to predict how likely a domain is to rank in search results, computed over Moz’s own link index. The framing matters: DR is a description of a link profile, DA is a prediction about ranking, fit against a training set.

Those are different objects. One is a summary statistic of a graph. The other is a model output. A model can, in principle, move because the model was retrained even if the graph didn’t change at all.

Semrush’s Authority Score is a third design again — a composite, documented as blending link signals with organic traffic data and spam signals. Composites are harder to interpret precisely because a single number now has multiple ways of moving.

Why the numbers don’t line up

Three mechanisms, and they stack.

Different indexes. There is no canonical copy of the web’s link graph. Each vendor has a crawler with its own frontier, its own recrawl priorities, and its own budget. A link that one crawler has seen and another hasn’t contributes to one score and not the other. This is the same mechanism behind why two tools report different backlink counts, and it propagates upward into every derived score.

Different formulas. Both are closed. Nobody outside the vendors knows the damping factors, the treatment of link attributes, how sitewide links are discounted, or the base of the log. Published descriptions are descriptions, not specifications.

Different calibration. Each score is normalised against its own population. “Roughly the top of our index sits at 100” is a decision about scaling, and two vendors making that decision independently will place the middle of the distribution in different places.

The comparison that is actually valid

Same tool, same day, relative position.

“We’re DR 41; the three domains outranking us for our main term are DR 58, 62 and 71” is a defensible sentence. Every number in it came from one index on one date, and the claim is ordinal.

“We’re DR 41 and they’re DA 44, so we’re behind” is not a defensible sentence. It compares two different measurements of two different datasets and then subtracts them.

If your reporting has to survive someone opening a different tool, the fix isn’t picking the “right” metric. It’s labelling: metric name, vendor, pull date, and the comparison set. That habit is most of what makes a report hold up — more on the rest in building a link report you can defend.

Both are ordinal, which breaks common arithmetic

Because these scales are compressed at the top, several operations people perform on them routinely produce numbers that mean nothing.

• Percentage change. “DA up 12%” is not a quantity. Report the endpoints: 38 to 43.
• Averaging across a link set. The mean of DR 15 and DR 85 is not “an average link of DR 50.” Report a distribution or a median plus a count of the top band.
• Differences as distances. The gap from 30 to 40 and the gap from 70 to 80 are both “ten” and are not comparable amounts of work.
• Cross-vendor deltas. Subtracting a DA from a DR produces a number with no referent at all.

A worked example, hypothetical: suppose you audit two prospect lists. List A averages DR 45 across 40 domains. List B averages DR 45 across 40 domains, but 3 of them are DR 90 and the rest are DR 20. Same average, completely different lists. The average told you nothing the median and a histogram wouldn’t have told you better.

Which one to use

Whichever one you will use consistently, for the job it’s suited to.

Use a link-only score (DR, or Moz’s link-based metrics) when the question is about link position: prospect triage, sorting a competitor set, spotting an outlier in a profile.

Use a ranking-prediction score (DA, as documented) when the question is “does this site plausibly rank,” accepting that you’re consuming a model’s guess and the model’s training set is not visible to you.

Use neither as a quality judgement. No 0–100 score in any of these tools has read the site. A high score means a favourable link position, which correlates with quality in the wild because sites people cite tend to deserve citing — but the correlation is a fact about the world, not a property of the metric.

What nobody outside the vendors knows

The formulas. Also: how often they are recalibrated, how much of a given score’s movement in a quarter is site change versus index growth versus model retraining, and how either score relates to anything Google actually computes.

That last one deserves emphasis. Neither DR nor DA is a Google metric, an input to Google, or a reconstruction of one. Google has said for years that it does not expose a domain-level authority score. These are vendor estimates of a graph Google also crawls, and the honest description of the relationship is “correlated, by construction, to an unknown degree.”

What to say when someone asks which is right

Neither. They’re measuring different things over different data with unpublished formulas, and their disagreement is expected rather than diagnostic. Pick one, label it, use it for relative comparison, and don’t do arithmetic on it that its scale won’t support.

If you want to know why the underlying counts diverge before the scores are even computed, start with what a link index actually contains. If you want the fuller treatment of the log-scale problem, see what Domain Rating actually measures.