Accurate enough to put the few links worth a person’s time at the top of a long list, but not a final ruling. A blocklist match is confirmed. Anything judged only by keywords or by the model is marked Needs review so that a person decides.
No single accuracy figure would be honest. It depends on the site: a brochure site with ordinary external links will mostly see correct results, while a site that links to many large brands will see more Couldn’t check results. The detail panel shows how each verdict was reached, so you can judge each one.
How reliable each kind of evidence is
| Evidence | How far to trust it | Why |
|---|---|---|
| Blocklist match (Confirmed) | Strongest | The domain is on a public list. Lists are domain-level and can lag behind, and a domain that changed hands may be listed for what it hosted before. |
| AI-judged | Moderate | A model can misjudge a borderline page in either direction. The panel says when a model was used. |
| Keyword | Weakest | Words cannot tell a news story about casinos from a casino. These are always Needs review, never Not safe. |
| Registry data (expired, not registered, new) | Strong where published | Read from the domain’s registry. Some extensions publish no date, and for those nothing is flagged. |
What a clean report does not prove
- No known issues is not “safe”. A brand-new site on no list that avoids the words matched will pass.
- It will not find every injected link. It finds links in the HTML and inline scripts of pages in your sitemap. Links built by scripts loaded from other files, hidden by stylesheet classes or shown only to crawlers verified by network address can be missed.
- It will not find every dead link. It finds links in the pages it read that return errors or cannot be reached. Pages outside the sitemap are not read.
- A clean row is not proof that every link works. It means none were confirmed broken in what was checked, and pages can break at any time. That is the reason for monitoring.
- A match does not mean a site is dangerous to visit. Adult and gambling sites are often legal. The match means the content probably does not suit the kind of site linking to it.
Why a link you know is bad may not be flagged
It may not be on a list, the page may have changed since, or the link may be shown in a way a scan cannot see. Accuracy and limits sets these out in tables. If you think a gap could be closed, the roadmap guide explains how to report it.
When Link Signals gets it wrong
It can. Judgements about relevance are not exact, and a blocklist match can be out of date, which is why each finding carries its reason and its evidence. You can mark a finding reviewed or ignore it when you have looked and are content. Neither changes what Link Signals found.
Where to read more
- Accuracy and limits: the full list of what can go wrong.
- Blocklist sources: where the lists come from and what they leave out.
- False positive: what it is and how to treat one.
- Google’s spam policies: the search engine’s own description of hacked content and link spam.
Questions
Is a Confirmed finding ever wrong?
It is a confirmed match against the list, but lists are not perfect. A domain that changed hands may still be listed. Look at the destination yourself before acting.
Does a second scan make the results more accurate?
A second scan gives the same answer when nothing changed, so repeating it adds no accuracy. What improves coverage is monitoring, which catches new links, or a bigger plan, which reads more pages.
Should I trust a Needs review finding or not?
Treat it as a question, not a verdict. Open the reason, the page it was found on and the destination. If the link is deliberate and fine, mark it reviewed or ignore it.
Run a scan, then read the reason beside each finding.
