Purchased-link audit findings
What the link metrics hid.
High DR, reported traffic and a familiar publication name can make a placement look reassuring. The underlying pages, queries and backlink evidence can tell a different story.
This brief brings together findings from two agency audits. It shows what failed review, the evidence behind selected cases and why human judgement remains part of the decision.
Evidence dates: 20 and 27 August 2026. Historical snapshots are dated separately. Published synthesis: 10 October 2026.
Two audits. Different units. A clear counting rule.
The wider review project concerns 2,800 links built across May, July and September 2026. The rates below are calculated from the two documented audit cohorts shown here. They are not measured results for all 2,800 links.
Audit A covers 424 unique domains. Audit B covers 21 purchased placements across four suppliers. Client identities are omitted from this public brief.
| Measure | Audit A | Audit B |
|---|---|---|
| Unit counted | Unique domains | Purchased placements |
| Cohort size | 424 | 21 |
| Did not pass the published QA definition | 160 | 11 |
| QA failure rate | 37.7% | 52.4% |
| Classified as manipulation | 51 | 10 |
| Share classified as manipulation | 12.0% | 47.6% |
45.1% - Simple average of the two QA failure rates.
29.8% - Simple average of the two shares classified as manipulation.
Each audit has equal weight in these averages. They are not pooled rates: Audit A counts domains and Audit B counts placements, and the cohort sizes differ substantially. Calculations use the unrounded fractions.
Manipulation is included within QA failures. The percentages must not be added together.
Audit B's 11 failures comprise ten manipulation classifications and one private blog network (PBN) classification. One additional placement on a real publication was flagged for a geographic mismatch and excluded from that failure total. We preserve the original counting rule here.
Finding 01 · Domain-name traffic
40,100 estimated monthly visits. Top queries dominated by the domain name.
mygreenbucks.net | Audit B | 20 August 2026
The placement cost $300. The retained Top Pages report showed approximately 40,100 estimated monthly organic visits, with leading queries built around variants of the site's own name.
Examples included “my greenbucks.net”, “mygreenbucks.net kenneth” and “www mygreen bucks .net”. The report displayed a monthly search-volume estimate of 10,000 for “my greenbucks.net”.
The audit classified the site as manipulation. The useful evidence is the concentration of domain-name queries: the traffic headline did not establish an audience interested in the subject of the purchased article.
mygreenbucks.net, Top Pages, 20 August 2026. The highlighted Top keyword column shows domain-name variants across the leading pages. Traffic and search volume are tool estimates.
Source: Ahrefs, retained audit capture. Client names and explicit terms redacted by us.
A related pattern: vibromedia.com
The homepage, disclaimer and About page were recorded as ranking first for “vibromedia.com”. The Contact page was credited with 121 estimated monthly visits, also against the domain query. The audit classified this placement as manipulation.
A reason to investigate the audience behind a reported traffic figure. Branded searches alone do not prove automated visits, invented queries or the identity of an operator.
Finding 02 · Score vs. evidence
The score was prominent. The supporting evidence needed a closer look.
blessingsfusion.com | Audit B | 20 August 2026
The retained overview displayed DR 46, seven organic keywords and 60 estimated monthly organic visits. With “Best links: Only” selected, the backlink panel showed two referring domains and four backlinks.
The placement cost $205 and concerned personalised medicine in mental health care. The audit recorded celebrity net-worth and biography queries among the pages carrying traffic and classified the placement as manipulation.
The screenshot supports a finding about the selected backlink view. It does not establish that only two domains linked to the site across every view or dataset.
blessingsfusion.com overview, retained from the 20 August 2026 audit. “Best links: Only” is visible. The highlighted panel shows DR 46, two referring domains and four backlinks in that selected view, alongside 60 estimated monthly organic visits.
Source: Ahrefs, retained audit capture. Client names and explicit terms redacted by us.
A second example: grosswheel.com
Audit A recorded DR 51. Under a dofollow, in-content filter with referring-domain traffic set to 100+, one qualifying referring domain remained. The worksheet also recorded leading pages ranking for “grosswheel.com” and “grosswheel.com home page”.
The final review classified the domain as manipulation. The single qualifying result is a filtered finding, not the count of its entire backlink profile.
Reading a score together with its underlying evidence and stating the filters used. These snapshots do not directly prove that someone purchased DR.
Finding 03 · Audience and brand context
A history site. Adult-performer traffic. A used-car placement.
historicallives.com | Audit B | Historical snapshot: 5 March 2026
The purchased article concerned maintaining a used car. The audit's historical snapshot told a different audience story: one page ranking for an adult-performer query accounted for 58.6% of the site's estimated traffic. Further adult-performer queries appeared among its leading pages.
The report recorded approximately 9,700 estimated monthly visits at that historical point. By the August audit, traffic had fallen to near zero. The placement was classified as manipulation.
The historical peak and the later audit are different observations. The concern is both the audience associated with the domain and the deterioration between those dates.
historicallives.com, historical Top Pages snapshot dated 5 March 2026, as documented in the August audit. The leading adult-performer query page represented 58.6% of estimated site traffic. This is a historical share, not an October measurement.
Source: Ahrefs, retained audit capture. Client names and explicit terms redacted by us.
Adult-related queries also appeared in Audit A
The worksheet for thebulletintime.com recorded leading queries including “sweetie fox height”, “mia khalifa net worth” and “stacy cruz porn”. The human reviewer classified the domain as Bad, with the note “useless kws”.
Reviewing the actual audience and brand context rather than relying on a publication's category label. It does not establish the effect of a placement on rankings or prove that every reader came through these queries.
Finding 04 · Repeated exposure
The same domain was flagged in both audits.
ceocolumn.com | Audits A and B
In Audit B, ceocolumn.com was a $140 placement. The 20 August record displayed DR 81 and described ranking queries centred on celebrity net worth, salaries and marriages, despite the publication's business positioning. The placement was classified as manipulation.
The domain appeared again in Audit A. The 27 August record displayed DR 80, and the human reviewer also classified it as manipulation.
| Audit | Evidence date | Recorded DR | Final classification |
|---|---|---|---|
| B | 20 August 2026 | 81 | Manipulation |
| A | 27 August 2026 | 80 | Manipulation |
The observations were recorded on different dates. Metrics from the two reports should retain their original scope rather than being combined into a single current profile.
Repeated exposure to the same domain across two buyers. The overlap alone does not establish common supplier ownership or a coordinated network.
A real publication can still be the wrong placement.
Audit B flagged constrofacilitator.com as a geographic mismatch: the publication's audience was described as predominantly Indian, while the article concerned choosing a plumbing, heating and cooling company in Albuquerque, New Mexico.
That finding was kept separate from the 11 QA failures. The original report included no geographic screenshot for this example, so it remains a recorded assessment rather than a visual exhibit here.
The same distinction matters for specialist sectors. A casino-payments article on a grammar and language blog raises a question about editorial fit. Casino or crypto content is not, by itself, evidence of manipulation. The publication, audience and placement need to be assessed together.
The review also kept genuine publications.
Audit A accepted 264 of its 424 domains under the published review definition. Human review reversed 115 automated rejections.
That matters because genuine publications can cover many subjects, have uneven traffic or trigger automated rules. A review needs a way to recognise those cases and explain why they pass.
In Audit B, accepted placements included dailyiowan.com, citybiz.co and lawfuel.com. Acceptance was an audit decision at the recorded date; these examples do not imply current Pharos availability or a sourcing agreement with those publishers.
- dailyiowan.com Accepted Audit decision at the recorded date
- citybiz.co Accepted Audit decision at the recorded date
- lawfuel.com Accepted Audit decision at the recorded date
Screening identifies questions. Review records the decision.
The audits used publication checks, traffic and query evidence, backlink analysis and human review. Findings were recorded against the available data and snapshots at the time.
The cases above were selected to illustrate clear failure patterns. They are not a random or representative sample of the wider project.
Three distinctions guide the interpretation:
- Observed data
- The pages, queries, estimates and filtered views retained in the evidence.
- Audit judgement
- The reviewer's classification and reasoning, including decisions to accept a publication.
- Unmeasured outcomes
- Ranking effects, actual visitor behaviour, financial losses across the wider project and who caused a suspicious pattern. These audits do not establish those outcomes.
Traffic figures are third-party estimates. Filters, dates and report scopes matter. The evidence here describes the recorded snapshots; it does not certify the domains' current condition.
Calculation note
Average QA failure rate:
[(160 / 424) + (11 / 21)] / 2 = 45.1%, rounded to one decimal place.
Average share classified as manipulation:
[(51 / 424) + (10 / 21)] / 2 = 29.8%, rounded to one decimal place.
Source note
This synthesis uses two retained agency audit reports prepared under TES, with QA by Kami's team, the associated screening and review worksheets, and dated Ahrefs captures. The wider 2,800-link scope and build months were supplied by the project owner; the displayed rates are derived only from the two documented cohorts.
What is behind the metrics in your latest batch?
Send up to 200 placement URLs for a free due diligence review. We'll return a written assessment of what passes, what needs attention and the reasons behind those decisions.
No purchase required. Scope and next steps confirmed by email.