Link building metrics that matter, and the vanity ones to drop

A link report can look excellent and describe a quarter that did almost nothing. Picture 412 new backlinks in one month. That sounds like a win until you notice they came from six domains, most of them sitewide footer credits on partner sites that have nothing to do with your product. The count went up. The case for a ranking improvement did not.

Choosing the right link building metrics is most of the work of measuring link building at all. The numbers teams reach for first (total backlinks, third-party authority scores, links built this month) are easy to pull and easy to inflate. The ones that predict results are less flattering and take more effort. Here is how to tell them apart, what belongs on a dashboard, and how to connect that dashboard to rankings without fooling yourself.

The link building metrics most reports get wrong

Vanity metrics share one trait: they rise with outreach volume whether or not any of it works. That’s fine as a record of activity. It becomes a problem when it’s the scorecard.

Third-party authority scores are useful for triage, the way a credit score is useful for a lender’s first pass. Use them to shortlist targets. Keep them off the headline of your report, because they describe a vendor’s index as much as they describe your profile.

Count referring domains, not backlinks

A referring domain is a distinct site that links to you. It’s the simplest of the link building metrics that carries real information, and it’s the one most reports bury.

The reason is editorial. Each new site that independently decides to link to you is a separate judgment by a separate person. Ten links from one site are usually one decision repeated ten times: one editor, one relationship, one article that happened to link. Count it once.

Deduplication takes one rule and one exception. The rule is to group by registrable domain, the name you’d register with a registrar, so blog.example.co.uk and shop.example.co.uk collapse into one. The exception is hosted platforms, where thousands of unrelated people publish under their own subdomains. There, the registrable domain is the platform itself, and grouping by it merges unrelated publishers into a single “site.” Group those by full hostname instead.

To measure, export inbound links from your crawler or link index, group them, and record each domain’s first-seen date. Net new referring domains for a month are the domains first seen in that month. One caveat matters. First-seen means when your tool found the link, not when the link went live. A link that’s two years old but was first crawled last week looks new. For placements you built yourself, record the placement date in your own log and trust that over the crawler’s timestamp.

Run the 412-link month again with this lens. Six referring domains is a small month. Thirty-eight relevant links across thirty-eight separate sites is a much bigger one, even though its raw count is less than a tenth of the first.

Relevance share: the metric most reports skip

This is the number I’d put first on any report, and it’s the one you can’t read off a tool. We made the mechanical case in relevance beats volume. Here it is as a number you can trend.

Score each placement on a three-point scale:

• 2, on target: the site covers your subject, and the linking page is about it or would send a reader to you naturally.
• 1, adjacent: the site sits in a neighbouring niche, or your subject is a minor part of the page.
• 0, off target: the site is unrelated, or the page exists to hold links rather than to be read.

Relevance share is the proportion of sampled placements that score 2. Sample thirty placements at random each month and score them yourself, or have two people score the same set and compare notes. Thirty samples is enough to see a direction, but not a precise number. At a true rate near 50 percent, a sample of thirty could plausibly land anywhere from about 32 to 68 percent. Sixty samples narrows that to roughly 13 points either way. So judge the trend across quarters, not the wobble between months.

Automated topical classifiers can pre-score at volume, and they’re worth using. They still miss the placement that looks relevant from its URL and isn’t, so keep a human sample to audit them.

Placement type, follow status and anchors

Three more attributes belong in the log. Each one is a diagnostic, not a pass-fail test.

Placement type. An in-content editorial link sits beside text that gives it context, which is why a reader would click it. A footer, a sidebar, or a link inside an author bio template is still a link, but it’s the kind of placement that’s cheap to produce at scale and often removed by an editor later. Track editorial share next to relevance share. The figure that matters most is the overlap: placements that are both editorial and on target.

Follow status. Google has treated rel="nofollow" as a hint rather than a strict directive since 2020, and it also recognises rel="sponsored" and rel="ugc". So a nofollow link isn’t worthless. It can still send readers, put your brand in front of an editor, and get the page crawled. Don’t chase 100 percent dofollow as a goal. If nearly every placement you earn is dofollow and the placements look paid, that’s a risk of its own, not a win.

Anchor type. Label each anchor as branded, naked URL, generic (“this guide”), partial match or exact match. The metric to watch is concentration, not any single anchor. If one exact-match phrase is a growing slice of new anchors, you’re drifting toward the over-optimized pattern that looks engineered at scale. Set a direction for this metric rather than a target percentage. The risk lives in the trend.

Indexed share, survival and referral visits: where value leaks

Every placement passes through a funnel before it counts. The graphic below shows a worked example with illustrative numbers, not benchmarks. Each stage drops some links out, and each drop has a different cause.

Start with never indexed. The page the link sits on is not in Google’s index, so the link isn’t something Google can use. Check it in Search Console’s URL Inspection for a sample. Bulk inspection through the URL Inspection API works, but it’s quota-limited, so sample rather than checking everything every day. A site: search is a rough check, not a measurement. Watch for “Crawled - currently not indexed” in Search Console’s page indexing report. It means Google found the page and chose not to keep it. That usually points at the page’s value, not at crawl access, which makes it a content problem rather than a plumbing one.

Indexed share is the indexed placements divided by the sampled placements. Check each placement at two, four and eight weeks after it goes live, because indexing is often slow and sometimes never happens.

Next is lost by day 90. The editor rewrote the article, the link was removed, the page was redirected or deleted, or the site migrated. Survival is the number of placements still live at 90 days divided by the number that have had 90 days to be checked. Recheck monthly: confirm the page responds, the link is still there, and its rel attribute hasn’t changed. Nobody tells you when a link disappears, so the recheck is the only way to know.

Then there’s off target or non-editorial, the relevance leak covered above. A placement can be indexed, live and dofollow and still be the wrong kind of link.

If your links are indexed, live and on target, and rankings still don’t move, the cause is usually somewhere else. Why your new backlinks aren’t moving rankings is the place to look next.

Referral visits as a ranking-free check

Referral visits are the one measure that doesn’t depend on how Google interprets a link. If a real reader arrived through a placement, that link is working on a scoreboard that needs no model at all. Pull sessions by source from the referral report in GA4 and record a trailing 90-day figure for each placement.

Expect small numbers. Twenty visits a month from a relevant guide is worth more than zero, whatever a third-party authority score says about the site. Treat the figure as a floor. Some referral traffic arrives without a referrer and shows up as direct, so the true number is higher than what the report can prove.

Tying link building metrics to rankings

This is where most reports fall apart, because link effects are slow, confounded and noisy. You can still get a defensible answer by building three habits into the process.

Map every placement to a target URL and a primary keyword. If links point at your homepage while you’re trying to rank a product page, a ranking chart for the product page will tell you very little. Your log should answer, for each placement, which page it’s supposed to help and which query that page is supposed to win.

Start the clock at first indexing. A placement can’t count until Google indexes the page it sits on, so the placement date is the wrong origin for your timeline. Annotate each target URL’s chart with the first-indexed date of every link pointing at it.

Compare against control pages. Pick pages with a similar template, a similar starting position and similar traffic that received no new links in the same window. Then compare the change in clicks and rank for the linked pages against the controls. This isn’t proof of causation. It does cancel out the shared noise you’d otherwise credit to links: seasonality, a core update, a sitewide decline. If linked pages move and controls don’t, something worth investigating happened. If both move together, the market probably moved.

Two measurement details matter here. Pull clicks and impressions per URL from Search Console each week. Track the primary keyword’s rank in a rank tracker rather than relying on Search Console’s average position, which is weighted by impressions across every query a page appears for and shifts whenever the query mix changes. And treat 60 to 90 days from first indexing as the minimum window you’d trust. Shorter windows mostly measure noise.

OptiLink’s tracking covers the placement side, including live status, dofollow checks and indexing progress, so the placement log doesn’t have to be maintained entirely by hand. You can see what it covers in the features section.

Build the dashboard step by step

A spreadsheet with a few COUNTIFS formulas and a basic chart tool is enough. You don’t need a data warehouse. What you need is discipline about what goes into the log.

1. Start a placement log. One row per placement. Include the placement ID, root domain, full host, target URL, primary keyword, date placed, date first indexed, placement type, follow status, relevance score, anchor type, last checked date, live status, and sessions over the last 90 days.
2. Add a sample flag. Mark the thirty rows scored in each month’s relevance sample, so the sampled figures can be reproduced and audited later.
3. Build a summary tab with formulas. Calculate unique referring domains (distinct root domains), net new this month (first seen in the month), relevance share (the share of sampled rows scoring 2), editorial share, indexed share over the sampled rows, and live-at-90-days.
4. Add a weekly tab for each target URL. Record Search Console clicks and impressions, the primary keyword’s rank, and an event column holding the first-indexed dates of links pointing at that URL.
5. Chart two things. Plot new referring domains per month as bars. Plot target-page clicks and keyword rank as lines, with linked and control pages shown separately.
6. Set the cadence. Check indexing weekly for placements under 30 days old. Run live and follow checks and the relevance sample monthly. Review linked against control pages quarterly.

For the top of the dashboard, use two rows of tiles. The first row holds referring domains (cumulative and net new), relevance share, editorial share, indexed share and live at 90 days. The second row holds referral visits, linked-versus-control clicks and primary-keyword rank. That’s eight numbers on one page. Any more and nobody reads past the first row.

Reading the dashboard without fooling yourself

A dashboard invites over-reading. A few rules keep you honest.

• Annotate everything. Log core updates, site migrations, content rewrites and your own link launches. Without annotations, every movement becomes a story written after the fact.
• Don’t read a single week. Rankings are noisy. Chart four-week rolling averages and judge the direction of the line, not the last point.
• Read the quality tiles before the quantity tiles. A month with more referring domains and a falling relevance share is a warning. It means outreach is drifting back toward volume.
• Watch survival. Strong placements that vanish by month four point to a leak in your process, such as a placement type that editors routinely remove. That’s a pattern to fix, not bad luck.

What to keep, and what to drop

Keep six things on the report: net new referring domains, relevance share, editorial share, indexed and survival rates, referral visits from each placement, and linked-versus-control rankings for target pages. Drop total backlinks as a headline, authority scores with no label, and “links built” as a stand-in for progress. Those numbers aren’t wrong. They just answer questions nobody needs answered.

The same logic sits behind OptiLink. The product finds relevant placements from millions of blogs, hand-places links where they fit the content, checks each one for dofollow status and pushes it to get indexed, and tracks it live. That sequence is the quality-first approach above, run as a process rather than a spreadsheet you maintain by hand. If you want to see your own placements measured this way, how OptiLink works walks through the steps.

Keep reading

Playbook

Anchor text optimization without tripping a filter

A hundred random links can do less than five relevant ones. Here's what makes a backlink pass real ranking value in 2026 — and what makes it worthless or worse.

Sofia Meyer September 24, 20262 min read

Playbook

Broken link building, step by step: find, fix, and pitch

A hundred random links can do less than five relevant ones. Here's what makes a backlink pass real ranking value in 2026 — and what makes it worthless or worse.

Sofia Meyer October 9, 20262 min read

Playbook

Competitor backlink analysis that finds real opportunities

A hundred random links can do less than five relevant ones. Here's what makes a backlink pass real ranking value in 2026 — and what makes it worthless or worse.

Sofia Meyer September 26, 20262 min read

Explainer

Dofollow vs nofollow in 2026: what still passes value

The link attributes changed, Google's treatment changed, and a lot of advice is out of date. Here's what dofollow, nofollow, sponsored and ugc actually do now.

The OptiLink Team October 10, 20262 min read