Backlink Checker: Why Two Tools Disagree on the Same Site

Quick answer: A backlink checker is a third-party web crawler plus its own index, not a view into Google. Because vendors make different choices about crawl breadth, recrawl frequency, link expiry and deduplication, two tools can report totals that differ by an order of magnitude for the same site. Both can be correct.

The most common question we are asked about link tools is not how to use one. It is why the numbers do not match. A business owner runs three checkers on their own domain in one sitting, gets 68, 412 and 1,900, and reasonably concludes that at least two of them are broken. None of them is. They are three different instruments measuring three slightly different things, and the gap between them is not an error margin but a consequence of design decisions made years earlier in someone else’s engineering meeting. This post is about the category rather than any product: how these indexes are actually built, the seven specific mechanisms that produce disagreement, and what to do with numbers you cannot reconcile. It does not cover the free tier in detail or how to prospect for new links, which are separate subjects. If you want the service framing, that sits on our SEO services page.

What a Backlink Checker Actually Is

Three components, and understanding the shape of them removes most of the mystery.

A crawler. Software that fetches pages continuously, follows the links it finds, and adds newly discovered addresses to a queue. Every vendor runs its own, under its own user agent, at its own rate, from its own seed list.

An index. A database of link records: source URL, target URL, anchor text, rel attributes, position, first seen, last seen. This is the product. The crawler exists to fill it.

A layer of derived metrics. Proprietary authority scores, spam scores, traffic estimates and topic classifications, all computed from the index. These are opinions, not observations.

Nobody has Google’s index and nobody is close. That is not a criticism of the tools, it is arithmetic. Google crawls at a scale no commercial vendor funds, discards most of what it finds, and does not publish the link graph. Every third-party number you have ever seen is a sample of the open web assembled by a crawler with a budget. The correct mental model is a survey, not a census.

The practical consequence is that a link tool answers “what did my crawler see recently” and you are usually asking “what does Google count”. Those questions have related but different answers, and no amount of switching tools closes the gap. Reconciling the two is one of the first things we do on a new account, and it sits inside the diagnostic rather than the reporting, which is why it lives with our SEO audit and consulting work.

The Seven Mechanisms That Produce Disagreement

Each of these alone can shift a total by a large multiple. Together they explain essentially every discrepancy you will meet.

One: crawl coverage. Crawlers prioritise differently. One vendor may weight recency and news, another may weight breadth across low-authority domains, another may aggressively skip sites it has classified as low value. A link sitting on a small local blog may be in one index and absent from two others simply because two crawlers never got there.

Two: recrawl frequency. A link added to a page last Tuesday exists only in the indexes that have refetched that page since. Popular pages get refetched constantly; a static member-directory page on an association site might be revisited rarely. In a market with as many low-traffic institutional pages as Singapore has, this single mechanism produces a lot of the divergence we observe.

Three: expiry policy. When a link disappears, how long does it stay in the report? Some indexes retain removed links for a long window and flag them as lost. Others drop them on the next confirmed fetch. A site that has lost a few hundred links over two years can therefore show wildly different totals depending purely on retention rules.

Four: deduplication. This is the biggest single cause of order-of-magnitude gaps. If a partner site puts your logo in its footer and has 6,000 pages, that is either one link or 6,000 links depending on the vendor’s collapsing rule. Some collapse to one row per source domain, some per template, some not at all.

Five: host and protocol handling. Whether the tool treats www and non-www, http and https, and subdomains as one entity or several. Ask for the domain and you may get the host. Ask for the host and you may miss half the profile.

Six: redirect and canonical handling. If a link points at an old URL that now redirects to your current page, is that a link to the current page? Vendors answer differently, and for any site that has ever migrated, this alone can account for a large share of the difference.

Seven: rendering and spam filtering. Links inserted by JavaScript are visible only to crawlers that execute it, and many do not, or do so selectively. Separately, some indexes suppress domains they have classified as spam networks, so a profile with a dubious history looks much cleaner in one tool than another.

Conventional wisdom says to pick the tool with the largest index. We would argue the opposite priority. Size determines how much of the long tail you can see, which matters for forensic work on an inherited profile. Consistency determines whether your own trend line means anything, which matters every single month. For an SME the second is worth more, and the practical rule is to choose once and then stop shopping.

The Metrics Problem, Which Is Worse Than the Count Problem

Disagreeing totals are at least visibly disagreeing. Derived metrics disagree invisibly, because each is expressed on a tidy 0 to 100 scale that invites comparison it cannot bear.

Every authority score is proprietary and computed only from that vendor’s index. If the index is partial, the score is a summary of a partial view. Two tools giving the same domain 34 and 61 are not contradicting each other; they are reporting on different graphs with different formulas.

The scales are non-linear. Vendors generally describe them as logarithmic, which means the effort to move from the low twenties to the low thirties is not remotely the effort to move from sixty to seventy. Treating the number as linear progress produces plans that are wrong by a large factor.

None of them is a Google signal. Google has not published a public per-domain authority figure for many years. When a proposal promises to raise a score, it is promising movement on a third-party vendor’s estimate of its own crawl, and the honest version of that promise is much smaller than it sounds.

Spam and toxicity scores are classifications, not verdicts. They are useful for sorting a long list. They are not evidence that any search engine has taken a view, and acting on them aggressively, particularly through disavow files, does more harm than good in most cases we have reviewed.

The useful discipline is narrow: use one vendor’s metric, use it only to rank prospects against each other inside that one tool, and never let it become a reported KPI. What gets reported should be referring domains, the quality of the sources in plain language, and the commercial outcome. That last point is why our reporting on accounts like the medical SEO results case study leads with enquiry volume rather than link scores.

How to Normalise Data You Cannot Reconcile

You will never make two tools agree. You can make their output usable.

Four rules make this operational. Standardise on one tool for trend and never switch mid-programme. Report referring domains at domain level, not link totals. Freeze the settings, meaning domain versus host, all links versus followed only, and live versus all history, and write down which you chose. Reconcile against Search Console quarterly to catch the case where your tool and Google have genuinely diverged, which happens most often after a migration.

One more habit worth building: when a number moves sharply, check the tool before you check the site. A jump of several thousand links overnight is far more likely to be a vendor changing a dedupe rule or a crawler discovering a sitewide footer than anything happening in your market.

The Three Questions a Checker Answers Well

Being precise about the tool’s competence makes it far more useful.

Does this specific link exist, and what does it look like? Excellent. Source, anchor, rel attribute, position, first seen. This is the tool at its best and it is why the first thing we do with any claimed placement is look it up rather than take the report on trust.

Which reachable domains link to my competitors and not to me? Good, with the caveat that you are seeing one crawler’s view of their profile. For prospecting in a small market this is genuinely valuable, because the same reachable domains recur.

Is the overall direction of my profile up or down over a year? Adequate, within one tool. Attrition is real and largely invisible without this.

And the three it answers badly: how much is this link worth, which nobody can answer; why did my rankings move, because a link graph cannot see relevance, intent or the dozens of other inputs; and is this link hurting me, which is a classification dressed as a diagnosis.

The Singapore Complication Nobody Mentions

Two structural features of this market make tool disagreement both worse and less consequential than the vendor documentation implies.

Worse, because the local link graph is thin and unevenly crawled. A large share of the legitimate linking domains available here are institutional: statutory board pages, association member lists, professional registers, polytechnic and university pages, chambers of commerce, and a small set of trade publications. Those pages change rarely, attract little traffic and sit in a sparse neighbourhood, which is exactly the profile a commercial crawler deprioritises. We have repeatedly found association and directory links present in one index and absent from two others, and confirmed by hand that the link was live the whole time. If your profile is weighted towards institutional sources, expect bigger gaps between tools than a global comparison post would predict.

Less consequential, because the absolute numbers are small. The total pool of realistic linking domains for a Singapore SME is countable. Once you have collapsed duplicates, a typical local service business is working with a profile in the tens, and a market leader in a local niche is often in the low hundreds rather than the thousands. When the whole profile is 40 domains, you can verify it manually in an afternoon, and manual verification outranks any index. That option simply does not exist for a business with 12,000 referring domains.

The percentage trap follows directly from the small numbers. Moving from 31 to 38 referring domains is a 23 per cent increase, and it will look spectacular in a report. It is seven links. Conversely, losing four directory listings when a local aggregator shuts down reads as a 13 per cent decline and means almost nothing. Report absolute counts and name the sources in a market this size; percentages mislead in both directions. This is one reason our pricing separates diagnostic work from ongoing work, because the diagnostic is where absolute reality gets established.

The most common cause of a dramatic disagreement is a single sitewide placement. A distributor, partner or supplier site that carries your logo and a product link in its footer or page template can generate hundreds or thousands of backlinks in one tool, because every page counts, while another tool samples only a fraction of those pages. At domain level, both tools usually agree closely, because that site is one referring domain however many pages it has. The disagreement is real, the underlying reality is not in dispute, and the only number worth reporting is the referring domain count.

Choosing and Configuring One, Without the Shopping Trip

Match the tool to the job rather than the marketing. Forensic work on an inherited profile needs index depth and history. Ongoing monitoring for a small site needs consistency and a clean export. Multi-market work needs regional coverage. These pull in different directions and no single product wins all three.

Test on a domain you already understand. Run the candidates against your own site, where you know what you earned, and see which one finds the placements you can personally verify. That is a far better test than a published index-size claim.

Check the crawler is not being blocked. Some hosts and security layers block third-party SEO bots by default, which silently distorts your own data and your competitors’.

Write the configuration down. Domain or host, all or followed, live or historical. Six months later nobody remembers, and an unexplained step change in the chart is usually someone having toggled a setting.

Budget honestly. For most Singapore SMEs one seat on one mid-tier product, or a free tier plus Search Console, is the correct answer, and the money saved is better spent on the work. Where a business also sells online, the feed and product-page questions usually outrank the link questions, which is the argument on our e-commerce SEO page.

In regulated sectors the constraint is different again, because the sources worth having are the ones with editorial and compliance standards, and there are not many of them. That is set out on our finance SEO page.

Field notes: In our medical case study, the GP clinic in Toa Payoh started with no directory listings of any kind. The off-page work was a named list of sources rather than a link total: consistent NAP data submitted to 40+ Singapore healthcare and general business directories, and four conflicting old listings with outdated phone numbers resolved. Because the work is a list of specific sources, each one can be opened and checked by hand rather than taken from whatever an index happens to record. Over the 6-month programme, which also covered technical fixes, a Google Business Profile rebuild and medical content, Domain Authority moved from 8 to 19 and monthly organic enquiries from 2 to 19. Domain Authority is itself a vendor metric: useful as a direction of travel, and a different tool would show a different number for the same site on the same day. The figure to trust is the list of sources you can verify yourself, which is why we report domains and named sources rather than raw link counts.

Our Take

A backlink checker is a survey instrument with a house style. Learning the house style is more valuable than learning the interface, because it tells you which of its numbers to trust and which to ignore.

The trade-off worth resolving rather than leaving open: depth and consistency are in tension, and you have to decide which your situation needs. If you inherited a site with an unclear history, buy depth for one quarter and do the forensics properly. If you are building a profile from a small base in a small market, consistency is worth more, and the cheapest tool you will stick with beats the best one you will abandon.

Most agencies report total backlinks because the number is larger and the chart is steeper. We would report referring domains, name the sources in words, and put the commercial metric next to them, because in a market this size a profile of 40 named domains is a description a business owner can actually check. If the tools disagree and the number matters, verify it by hand. That is a realistic option here and it is the one advantage of working in a small market. Which sources are worth pursuing varies enormously by sector, which is why we scope this per vertical across our industry SEO work.

None of that is a tooling decision, which is the wider point. It is a decision about what you are willing to report and defend, and how we think about that is on our about page rather than in a rate card.

We cross-check at least two backlink checkers before reporting a number to a client, because in our experience no single crawler’s index matches Google’s view of the web, and presenting one tool’s count as ground truth sets the wrong expectation from the start. Our team normalises the data to referring domains specifically because that figure drifts less between tools than raw link counts do.

Frequently Asked Questions

Which backlink checker is the most accurate?

There is no defensible answer, because accuracy against Google’s link graph cannot be measured from outside it. What can be compared is coverage against links you have personally verified. Run the candidates on your own domain, where you know which placements exist, and see which finds them. In our experience the ranking that produces varies by market and by the type of source you rely on, and it differs for a Singapore business weighted towards institutional links.

Why does Google Search Console show different links again?

Because it is not a crawler report at all. Search Console shows a sample of what Google recorded, deduplicated and capped, with its own reporting lag and no proprietary metrics attached. It is the closest thing to authoritative for the question of what Google has seen, and it is the weakest tool for competitor work because it only covers domains you own. Use both and expect them to differ.

Should I report total backlinks or referring domains?

Referring domains, without exception. Total links is inflated by sitewide templates and is the easiest number in SEO to make look good without doing anything. Referring domains is closer to the thing that has value, is far more stable between tools, and is much harder to game accidentally. If a report leads with total backlinks, ask for the domain-level figure alongside it.

My backlinks dropped by thousands overnight. What happened?

Check the tool before the site. The overwhelmingly likely explanations are a vendor changing a deduplication or expiry rule, a settings toggle between domain and host, or a single sitewide link source being removed or reclassified. Genuine loss of thousands of distinct referring domains in one day is almost unheard of outside a major site migration. Confirm against referring domains and against Search Console before treating it as an event.

Are the authority scores comparable between tools?

No, and they are not meant to be. Each is computed from that vendor’s own index using its own formula, and each is on a non-linear scale. A domain scoring 40 in one product and 22 in another is not a contradiction. Use whichever you have chosen to rank prospects against each other within that one tool, and drop the number entirely from anything client-facing.

How often should I check my backlink profile?

Monthly for a small site, and quarterly is defensible if the profile is not growing fast. The useful cadence is tied to your placement list rather than to the dashboard: check that everything you earned still exists, then glance at the totals. During the ninety days after a site migration, check weekly, because that is when redirect handling breaks links in volume and when the tools will disagree most.

Do these tools see links on pages that need JavaScript?

Sometimes, and inconsistently. Crawlers that execute JavaScript do so selectively because it is expensive, so a link inserted by a script may appear in one index, appear late in another, and never appear in a third. If you have earned a placement and it is not showing up anywhere, view the page source rather than the rendered page and check whether the link is in the HTML at all. That single check resolves a surprising share of missing-link questions.

Can a backlink checker tell me why my rankings moved?

No, and treating it as though it can is one of the more expensive habits in the field. A link index has no visibility into query intent, page-level relevance, competitor content changes, technical health, user behaviour or the dozens of other inputs involved. It can tell you your link profile changed, which is one candidate explanation among many. Correlating a ranking drop with a link report and stopping there produces confident wrong answers.

Is it worth paying for two tools?

Rarely for an SME, and only for a defined period. The case for it is forensic: one quarter of overlapping access while you diagnose an inherited profile, then cancel one. Running two indefinitely produces two trend lines that disagree, two sets of scores and an ongoing argument about which to believe, which costs more attention than it returns. Pick one for the long run.

Should I be worried that a tool shows spam links I did not build?

Usually not. Scraper sites, auto-generated directories and content farms copy links indiscriminately and every site of any age accumulates them. Search engines discount this category routinely. The situations that warrant attention are a sudden large volume appearing in a short window, which often indicates a security problem on a linking site or your own, and a documented history of paid link building, which is a cleanup project rather than a monitoring one.