How a domain becomes suspect

The flag

Each measured domain carries one of three states. unverified: not yet measured, which is not a pass. ok: measured, and no test fired. suspect: at least one test fired; the response names which. Today 201,707 domains are measured and 24.7% of them read suspect.

The five tests, with thresholds

high_authority_no_traffic

Measured domain rank above 400 of 1000 with under 100 estimated monthly US organic visits.

high_authority_no_keywords

Measured domain rank above 300 with fewer than 50 ranking US keywords.

spam_score

The index's spam score on the domain's link profile above 50.

subdomain_without_own_footprint

A subdomain sold at its parent's rating, with neither 500 keywords and 500 visits of its own nor a rank of 200 and 300 referring domains of its own. A finance desk on a major portal's subdomain passes; a blog on a free host does not.

saturated_outbound

From the web graph: links out to 500 or more outside domains and at least five times as many as link in. A publisher that sells more placements than it earns links.

The tests overlap and are counted separately; they do not sum to the suspect share. Traffic and keyword tests are never applied as a blanket “zero and zero” rule, because the index is US English and a real site elsewhere reads as zero.

What it misses, what it over-catches

Misses: a farm that also buys a little traffic; a domain that was real when measured and has since been sold; a page-level problem on an otherwise clean site. Over-catches: young sites, sites ranking in other languages or markets, sites with real audiences that do not arrive from search. That is why the flag is a strong signal and a default exclusion, not a wall: a buyer can include suspect inventory and is told why it was flagged.

Restricted content

Gambling, adult and piracy publishers are marked from the sellers’ own acceptance labels and topic tags, and excluded from plans unless the brief names the vertical. They are not “suspect”; the two are separate fields, and the research page reports how they differ.

Placement Score v1

One labelled number per publisher, built only from measurements a seller cannot influence, and always shown with its components so it can be checked against them. Five components, each 0–100:

authority · weight 25

The claimed DR and our measured rank, averaged where both exist. On a subdomain the parent's DR is ignored.

visibility · weight 25

Measured monthly traffic and ranking keywords on log scales; one million visits or one hundred thousand keywords is 100.

quality · weight 30

Starts at 100 and loses points for each test that fired (40 to 50 each), spam over 30, outbound-to-inbound of two or more, and restricted content. Null until measured.

graph · weight 10

Inbound domains, PageRank and the share of any topic's leaders linked, from the web graph. Null on subdomains.

survival · weight 10

Live share of our own placements on the publisher, at 90 days where available. Null under three orders.

A component we have not measured drops out and the rest re-weight; confidence is how much of the weight had a measurement behind it. A suspect publisher is capped at 40 regardless. The score excludes observed effectiveness, relevance to a brief and supplier reliability; each is reported separately so a buyer sees which is which. Weights are stated, not fitted; they will be revised against observed outcomes and the version will change when they do.