Backlink Audit Checklist: Find, Judge and Fix Your Links

A backlink audit is a scheduled inventory of every external link pointing at your site: where it comes from, whether it is still live, whether it passes value, and whether you would be comfortable if a search engineer looked at it. It is not a one-off cleanup project, and its main output is usually not a disavow file.

Most audits produce three lists. Links you have lost and can plausibly get back. Domains worth going back to, because a site that linked once will often link again. And a small pile of genuinely bad links, which for most sites is smaller than expected.

The value sits in the first two. Recovering a link from a site that already knows you is far cheaper than earning a new one from a stranger, and the audit is what surfaces those opportunities. Treat cleanup as a by-product. If you go in expecting to find a toxic-link problem, you will find one whether or not it exists, and spend a quarter fixing something that was never broken.

No tool sees the whole web. Every backlink index is a crawl with its own budget, refresh cycle and rules about what counts as a link, so two tools will hand you two different totals for the same site. That is expected, not a defect, and it is a reason to work from more than one export.

Pull from at least these three:

• Google Search Console — free, and the only source telling you what Google itself associates with your site. Export the linking sites and the top linked pages.
• A commercial backlink index — for anchor text, page-level detail, discovery dates and competitor profiles.
• Your own records — every link you or an agency built, with the date and the placement URL. If this does not exist, start it today; it is the file you will miss most in a year.

Merge the exports and deduplicate by linking domain. Then fix the questions you are answering before you look at anything: which links have I lost, which domains are worth another pitch, which links misrepresent us, and is there anything here that would embarrass us. An audit without those four questions turns into an afternoon of scrolling.

Ten links from one site are approximately one link. Collapse the export to unique linking domains first, or a single sitewide footer link will make a thin profile look busy.

Then judge each domain on evidence you can see, rather than on one vendor’s score:

Open twenty of them by hand. Vendor metrics are useful for sorting thousands of rows into an order and useless as a verdict on any single site. Five minutes of looking at a page beats any score, and after twenty domains you will have calibrated your own judgement well enough to move quickly through the rest.

Domain quality settled, look at the placement. A link on a good site can still be worth nothing:

• rel attribute — nofollow, sponsored and ugc all tell search engines not to pass ranking value in the normal way. Worth having for traffic and discovery; not worth paying for as a ranking link.
• Is the linking page indexed? A page blocked in robots.txt, carrying noindex, or canonicalised elsewhere passes little or nothing.
• Does your URL still resolve? Links pointing at pages you deleted or moved should be redirected properly, or repointed at the source.
• Redirect chains — a link arriving through three hops is weaker than a direct one and easier to break later.
• Position on the page — a sentence in the body beats a sidebar or a footer.
• Protocol and host — links to the non-canonical variant still work through your redirects, but the redirect is worth verifying.

This is the part that reliably finds free wins: links that already exist and point at a 404. Fixing those costs one redirect each and requires no outreach at all, which makes them the first thing to action in any audit.

Links disappear constantly and quietly. A site redesigns, an article gets pruned, a CMS migration drops a paragraph, an editor swaps your link for a competitor’s. Nobody emails you about it.

Every audit should therefore compare against the previous one and produce a lost-link list. For each entry, work out which of these happened:

1. The linking page is gone entirely — check an archive, then pitch a replacement resource to the site.
2. The page is live but your link was removed — a short, factual email to the author works more often than expected, especially with a reason the link still helps their reader.
3. Your target URL moved and now returns 404 — fix it on your side; no outreach needed.
4. The link became nofollow — usually a sitewide policy change, and usually not worth arguing about.

Doing this by hand stops scaling past a few hundred placements, which is why Growmindr’s backlink module re-crawls every tracked placement and stamps it requested, promised, live or lost with a date. Whatever tool you use, the requirement is the same: a dated status per link, checked on a schedule, so a removal surfaces within days rather than at the next quarterly audit.

Anchor text is a ranking signal and an over-optimisation signal at the same time. Export every anchor pointing at your site and group it into four buckets: branded, naked URLs, generic phrases such as ‘read more’ or ‘this guide’, and exact-match commercial terms.

A profile built naturally is dominated by the first three. Real editors link with your brand name or the article title, because that is what reads well inside a sentence. A profile where a meaningful share of anchors are the exact phrase you want to rank for did not arrive by accident, and it looks that way to a search engine too.

There is no safe ratio to memorise, and anyone quoting one to a decimal place is guessing. What is reliable is the comparison: pull the anchor distribution for the three sites currently outranking you and compare the shapes. If yours is visibly more commercial than theirs, that is your finding, and the fix is to keep building links you do not control the anchor text of.

Read the anchors as text as well. Anchors mentioning products you do not sell, or in a language you do not operate in, mean a hacked linking page or a spray of junk links. Both are worth knowing about; neither is common.

The disavow tool tells Google to ignore specified links. It is a genuine fix for a narrow problem and a popular way to create a slow, invisible one.

Use it when you have a manual action for unnatural links in Search Console, or when you know links were bought or built in bulk at some point in the site’s history and you can identify them. In those cases, disavow at domain level, keep the file short, and submit it alongside a reconsideration request where one applies.

Do not use it because a tool coloured some rows red. Automated toxicity scores are heuristics, not judgements, and search engines already ignore ordinary link spam without your help — it arrives at every site that ranks for anything. Disavowing a good link is silent and slow to detect, because nothing in any interface will tell you that you switched off something that was working.

The test: can you explain, in one sentence and without citing a tool’s score, why this specific link exists and why it is a problem? If not, leave it alone. For most small sites the correct disavow file is empty, and the hours are better spent on the lost-link list.

An annual audit presents a year of problems at once. A monthly pass finds each one while it is still fixable, and takes far less time in total.

• Monthly — re-crawl tracked placements for live or lost status, and act on removals within the week, while the editor still remembers the article.
• Monthly — check link target URLs for 404s and redirect chains. This is the cheapest recurring win on the list.
• Quarterly — full export from Search Console and your index of choice, diffed against last quarter, with new referring domains reviewed for anything you did not expect.
• Quarterly — recheck the anchor distribution against the same set of competitors, so you are watching a trend rather than a snapshot.
• Yearly — the judgement pass: open the top fifty referring domains and ask whether you would still want each one.

Keep the results in one file, dated. An audit compounds only if you can compare it against the last one, and most of the useful findings in link work are differences rather than absolutes. A profile that lost fourteen links this quarter is telling you something that a profile with 312 links is not.