Where it can be wrong
| Limit | What happens | What to do |
|---|---|---|
| Keywords can’t tell topic from intent | An encyclopaedia article about casinos can match the gambling words, a false positive. | Keyword-only matches are marked Needs review, not Not safe. Look at them. |
| Blocklists are domain-level and lag behind | A new site, or even a large operator, may not be listed. In our testing a well-known gambling operator wasn’t on the list. | Treat “No known issues” as “none found”, not as proof. |
| AI judgements can be wrong | The model can misjudge a borderline page in either direction. | Findings judged by the model say so in the detail panel. |
| One report per destination | Each destination URL is checked once per scan and reported against the first page it was found on, even if it appears on many pages. | Search your pages for the anchor text to find the other places. |
What it cannot see
| Limit | What happens | What to do |
|---|---|---|
| Inline styles only | A link hidden by a stylesheet class is not detected as hidden. | Check suspicious findings by viewing the page source. |
| Scripts are read, not run | Links a page’s own inline scripts write, in plain or encoded text, are found. Links built by scripts loaded from other files, or assembled while the page runs, are not. See script-written links. | Treat a clean scan as “none found in the HTML and inline scripts”. |
| Cloaking | On the homepage and about 10% of pages, Link Signals also requests the page as a search crawler and as a visitor from Google, and flags links only those views show. Pages outside the sample are not compared, and code that verifies a crawler’s network address is not fooled. See cloaked links. | Don’t treat a clean scan as proof a compromised site is clean. |
| Domain registration coverage | Registration is read from the domain’s registry where it offers a public lookup. Some extensions don’t, and for those nothing is flagged on registration grounds. Who owns a domain, and a change of owner without re-registration, aren’t visible. | Treat a clean result as “nothing found”, not as proof of ownership. |
| Sites that block the crawler | They are reported as Couldn’t check, not as clean, along with what is known about the domain: how long it has been registered and that it is on no blocklist we use. Owners can also opt out. | Open them yourself. |
| Sitemap coverage | Only pages listed in the sitemap are scanned. If the sitemap is bigger than your plan reads, a sample spread across the whole site is read, not just the first pages. | See sitemap discovery. |
| Your own domain, certificate and HTTPS | The domain’s dates come from its registry and the certificate is the one your site presents on the standard secure port. Some domain extensions publish no dates, and a certificate on another port isn’t seen. | Read a missing date as unknown, not as fine. |
| Link markers (rel) | nofollow, sponsored and ugc are read from the page’s HTML. A marker added by JavaScript after the page loads, or sent only in a header, isn’t seen. | Check the live page if it matters. |
| Re-checks | A re-check reads only the pages your flagged links are on. It won’t find new problems elsewhere, and a site that won’t answer is left as it was. | Run a full scan, or use monitoring, for the rest. |
| External links only | Links within your own domain are not evaluated. | — |
How to read a report
Start with Not safe and anything with a Hidden badge. Then work through Needs review. A confirmed blocklist match is the strongest signal in the report; a keyword-only match is the weakest. The verdicts page explains each level.
Worked examples of reading a result carefully
A keyword match on an article
A local newspaper’s site is linked from your community page. Its homepage mentions a casino planning application. The keyword rules pick up “casino”, and the result is Needs review with the matched words listed. This is the false positive the keyword rules are most likely to produce. Open the page, see that it is a news story, and ignore the finding if you are content with it.
A clean scan on a site that was hacked
Injected spam is sometimes written to appear only to search crawlers, or loaded by a script from a separate file. Link Signals compares a sample of pages as a crawler would see them, and reads inline scripts, but it cannot cover every trick. A clean report is a good sign and not proof. If you have other reasons to suspect a compromise, such as odd search results for your own site, use the scan as one input and also check your files and your search console. See after a WordPress hack.
A large brand reported as Couldn’t check
Many well-known sites refuse automated requests. Link Signals retries once with a browser-style user agent, and does not go further than that to get past a block. The result is Couldn’t check with whatever is known about the domain. That is the honest answer, because the page was never read.
Which domain extensions give us expiry dates
Expiry dates come from each registry’s public lookup. Registries differ in what they publish, so expiry warnings work for some extensions and not for others. This is what we found when we checked a spread of real domains on 13 October 2026.
| Extensions checked | What the registry publishes | What it means for you |
|---|---|---|
| .com, .org, .co.uk, .fr, .ca, .cz, .in, .tv, .cc, .xyz | Registration and expiry dates | Expiry warnings, the expiring list and the monitoring alerts all work. |
| .nl, .org.au, .pl, .br, and gov.uk | A lookup, but no expiry date | A domain that does not exist is still found. One about to expire is not, because there is no date to read. |
| .io, .co, .me, .it, .ch, .at, .jp, .eu | No public lookup we can use | No registry data at all. Nothing is flagged on registration grounds, and the link is judged on its page alone. |
What the report is built for
Link Signals is designed for triage. It aims to put the few links worth a person’s time at the top of a long list, with the evidence beside each one, so that nobody has to open hundreds of links by hand. It is not a malware scanner for your server, a penetration test, or a replacement for a security plugin or hosting-level protection. Treat it as the part of your routine that watches where your pages point.
Reducing the chance of surprises
- Add your sitemap to monitoring so that a new link is caught within days and not at your next manual scan.
- Keep your sitemap dates accurate, so changed pages are read promptly. See checking sitemap dates.
- Read the reason and the page a finding was found on before you decide, not just the verdict.
- Search your site’s source for the anchor text of a finding, as each destination is reported once against the first page it was found on.
- Re-check after fixing, so the report reflects the live site. See re-checking fixed links.
Questions
How often is Link Signals wrong?
There is no single figure that would be honest. The answer depends on the site: a small 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 for yourself.
Does it change anything on my site?
No. It reads public pages and reports on them. It never edits, logs in to, or writes to your site.
Why was a link I know is bad not flagged?
It may not be on a list, the page may have changed since, or it may be shown only in a way that scans can’t see. See the tables above, and report it as feedback if you think a gap could be closed.
Are blocked sites a problem I should fix?
No. A site that blocks automated requests is reported as Couldn’t check, and it is not a fault with your link.
How this counts in your LinkSignal
LinkSignal inherits every limit described on this page. It counts only what a scan recorded, so a page that was not read, a link that could not be checked or a registry that publishes no date adds nothing to it, and a clean score is not proof that a site is clean. A scan that stopped early shows a provisional score for what it read. how the score is worked out gives the rules, and the LinkSignal overview lists the rows. LinkSignal is Link Signals’ own metric, worked out only from what this tool finds, and it is not connected to any search engine or third-party score.
Run a scan, then judge the findings with these limits in mind.
