How to Audit Your Backlinks Properly
Reporting is not fixing. An audit that ends in a spreadsheet was not an audit. It ends in a ranked list of actions with somebody's name against each one. Failing that, it produced nothing at all.
What An Audit Is For
An audit establishes what you have before anybody decides anything. It is a one off exercise producing a position: this is the profile, this is its shape, these are the things worth doing about it, in this order.
A monthly report is a different instrument entirely. It tracks movement against a baseline that already exists. It answers whether things went up or down. It assumes somebody has already decided what to watch, which is exactly what an audit is for.
Confusing the two produces the commonest failure in this work. A business asks for an audit, receives a document containing every link it has with a score beside each one, then finds it has learned nothing it can act on. That is a report of the current state dressed as an audit. It stops at the point where the useful part begins.
The test is what comes out of it. If the output is data, no audit happened. If the output is a ranked list of actions with owners and reasons attached, one did. Everything below is written towards that output. The full series sits in the complete guide to backlink building.
Why The Data Sources Disagree
Run the same website through three different tools and you will get three different link totals. None of them is broken. Understanding why saves a great deal of wasted argument.
No tool has access to Google's index. Each vendor operates its own crawler, discovers pages independently and maintains its own picture of the web. Those pictures are enormous and none is complete, so each tool is reporting a different sample of the same underlying reality.
The definitions differ too, which accounts for more of the gap than most people assume. Whether subdomains count as separate referring domains. Whether a redirected link is attributed to the old address or the new one. How long a link that has vanished stays in the report before being marked lost. Whether links from pages the crawler cannot reach are counted at all. Two tools using different answers to those questions will disagree even where their crawl coverage is identical.
What follows practically is a rule worth adopting. Choose one source as your system of record and take every trend measurement from it, so that comparisons over time are like for like. Use other sources to find things the first one missed, which is what they are good for. Never compare a figure from one against a figure from another and conclude something has changed, because you have measured two different things.
Establishing The Baseline
This is the step most often skipped and the one that makes everything afterwards interpretable. It takes an hour and it cannot be recovered retrospectively.
What to record. Total links and total referring domains. The distribution of anchor text, meaning which phrases appear and how often. Which pages on your site are receiving links, ranked by how many. The proportion carrying attributes against those that do not. The sources you judge most relevant to your field. Your visibility on search terms that do not contain your business name.
And the date. Stated explicitly, because the whole point of a baseline is that a future version of you can compare against it.
Export it rather than trusting the tool to hold it. Vendor indexes are refreshed and recalculated, sometimes retroactively, so a figure you read in March may not be the figure the same tool shows for March when you look again in September. An exported file is a record. A dashboard is a live view.
The reason this matters is not administrative. Every decision made later in the audit is a change. A change can only be judged against a starting position. A business that begins removing and pursuing links without a baseline has converted a reversible experiment into a permanent mystery, which is the point made at greater length in our guide to toxic backlinks.
Assessing Quality Rather Than Counting
Assessing every link individually is impossible on any profile of size and unnecessary on all of them. The objective is to characterise a profile rather than to adjudicate each link. Characterising is what sampling is for.
How to sample. Take three slices rather than a random handful. The strongest sources by apparent authority, because those are the links doing most of the work. A random selection from the middle, because that is where the truth about a profile usually lives. A slice of whatever a tool has flagged, to find out what the flagging actually consists of.
Four questions per link. Is the page real, meaning does it exist for a reason beyond hosting links. Is it read, meaning would anybody arrive there by any route. Was the link editorial, meaning did somebody choose to make it. Is the site in a related field, meaning does the reference say anything about what you do.
Then generalise. If forty links from the flagged slice are all scrapers and aggregators, you have learned something about the whole flagged category without opening the other three hundred. If the strongest sources turn out to be unrelated to your sector, you have learned why they are not producing anything.
What sampling avoids is the trap of treating a score as an assessment. A tool figure is a vendor's opinion, discussed in our guide to Domain Authority. Four minutes of looking at an actual page tells you more than any number attached to its domain.
Looking For Patterns
Individual links tell you about themselves. The shape of a profile tells you about its history. On an audit that history is frequently the thing you most need to establish.
A profile that grew looks irregular. References arrive at no particular rate, because they arrive when people happen to notice things. The wording varies enormously and much of it is the business name, a bare web address or a phrase like this guide. The sources have little to do with each other beyond a shared interest in the subject. Attributes are mixed, because different publishers apply different policies.
A profile that was assembled looks tidier. Arrivals cluster into periods when somebody was working. Commercially useful phrasing recurs. Sources resemble one another and frequently point at many of the same other destinations.
Neither description is a verdict on any individual link. This block is deliberately descriptive rather than diagnostic. The reasons those shapes emerge, plus why they cannot be designed around, are set out in how Google detects unnatural backlink patterns.
The reason to do this on your own profile is usually less about risk than about knowledge. A business that has changed agency, bought a domain or inherited a site frequently has no idea what was done in its name three years ago. The shape is how you find out.
Lost Links And What To Do About Them
Every profile loses links continuously. Most of the losses do not matter. Sorting the two apart is a standard part of any audit.
Why they disappear. The page was deleted. The site closed. The article was rewritten and the reference dropped. A redesign discarded the link while keeping the text. An editor tidied up. The site moved platform and the old pages went with it. A paid placement lapsed when somebody stopped paying for it.
Which losses matter. A link from a relevant page on a site with readers, which was contributing standing and possibly sending visitors. That is worth attention. Everything else is attrition.
When recovery is worth attempting. Where the page still exists and the omission looks accidental, such as a redesign that dropped links throughout. Where you have a relationship with the publisher. Where the piece still references your business in the text and only the link went, which is the easiest case to raise and the most likely to be fixed.
When it is not. The site has gone. The removal was clearly deliberate. The publisher never responds to anything. Or the link vanished because a placement expired, in which case the loss has told you something useful about what that link was. Pursuing it would mean renewing an arrangement rather than recovering a reference.
One point on framing. A falling link count is not automatically bad news. A profile shedding junk while gaining relevant references is improving on every measure that matters, even as the headline number declines.
How Often To Do It
A fixed schedule, plus a defined list of events that trigger an unscheduled one. Quarterly suits most businesses. Monthly is reasonable while link work is actively running, because there is something new to look at.
The argument for a schedule is about the state of mind you audit in. A business that audits because rankings dropped last week is auditing in alarm. An audit conducted in alarm finds whatever confirms the alarm. Somebody looking for a link problem will find candidate link problems, because every profile contains hundreds of things that could be blamed. The scheduled version asks what is here. The reactive version asks what is wrong. It gets an answer whether or not anything is.
Events that justify an unscheduled audit. Taking on a website whose history you do not know, including any acquisition or domain purchase. Changing agency, in either direction. Receiving a manual action notice, which is the one case where urgency is genuinely warranted. A site migration, before and after. A visibility decline that persists beyond a few weeks and has no other explanation.
That last one carries a condition worth stating. Persisting beyond a few weeks, once the ordinary explanations have been checked. Most declines are algorithm updates, technical faults, seasonality or a competitor improving. Links are far down the list of likely causes and near the top of the list of suspected ones.
Turning Findings Into Actions
Findings sorted into four categories, ranked by impact against effort, each with an owner and a date. This is the deliverable.
Fix now. Things actively broken and entirely within your control. The highest return finding in most audits sits here and surprises people: links pointing at pages on your own site that no longer exist. Somebody referenced you. The page moved or was deleted. Everything that reference was contributing is now being discarded at your end. Nobody else has to agree to anything for this to be fixed.
Pursue. Lost links worth recovering. Relationships that should have produced a reference and have not. Memberships, supplier listings and sponsorships you already hold and have never claimed, which is usually a short and productive list.
Leave. Explicitly and in writing. Scrapers, aggregators, statistics pages, foreign language copies, anything flagged that you have sampled and understood. Recording the decision to leave something matters, because otherwise the same items get re-examined at every future audit.
Investigate. Things you cannot yet explain, held here rather than acted on.
The governing rule across all four: nothing is removed or disavowed on a tool score alone. Where removal genuinely is warranted, the process and its narrow scope are covered in our guide to the disavow tool.
Common Mistakes
Five. The first is the one nobody admits to.
Auditing to justify a decision already made. Somebody has already decided to disavow, to change agency or to buy a link package. The audit exists to support it. The findings will oblige, because a large profile contains evidence for almost any conclusion.
Counting rather than judging. Producing totals and averages while never opening an actual linking page. Every number in a link report is a summary of things nobody looked at.
Acting with no baseline. Covered in block three. It makes the result of your own work unknowable.
Disavowing on sight. Treating a flagged list as a work queue. The links were probably being discounted already, so the file achieves nothing while risking something.
Treating a competitor's profile as a target to match. You cannot see which of their links are counting. You cannot see what they paid. Matching a number tells you nothing about matching an outcome. Competitor profiles are useful for finding sources that exist in your field. They are a poor specification for what yours should look like.
An audit that ends
in decisions.
A recorded baseline, a sampled assessment rather than a scored list, plus findings sorted into fix, pursue, leave and investigate with owners against each. Frequently the recommendation is to leave the profile alone.
Included every month:
One fixed monthly rate. No setup fee. No price per link.
Twenty four guides.
One subject.
This page covers the audit. The rest of the series covers what a tool score is worth, where the disavow file fits, how patterns are recognised, what makes a link valuable and when links are not the right spend.
