Checking backlinks is the easiest way to feel productive and learn nothing. A backlink audit tool hands you tens of thousands of rows in four seconds, and almost none of them change what you do next week. The useful question is narrower than “how many links do I have”. It is: which link facts would alter a decision, and which are just decoration?
We’ve maintained export websites since 2017 — 30–40 client sites and three of our own — so we look at link data every week. The habit that made the difference was separating the tool’s opinion from the evidence underneath it. This page is the reading order we use, the numbers we ignore, and the point where a backlink audit turns into an action.
What a backlink tool is actually showing you
A link index is a crawl snapshot. Each vendor crawls at its own speed, keeps its own history, and decides for itself what counts as a link. That is why the same site can show clearly different referring-domain counts in two tools on the same afternoon. Neither number is wrong. They’re answers to slightly different questions.
Google Search Console is the only first-party view, and it is deliberately partial. The Links report shows top linking sites and a sample of linking pages, not the complete graph. Treat it as the ground truth for direction, not for volume. If Google lists a domain as linking to you, that link exists. If Google’s sample misses one, it may still exist.
Two fields cause the most confusion. “First seen” is the date the vendor’s crawler first recorded the link, not the date Google discovered it, and the gap can run to months. “Lost” usually means the crawler stopped finding the link, which happens when a page is edited, a footer rotates, or the crawler simply didn’t revisit. A lost-link alert is a prompt to check, not a fact.
The four numbers worth reading
Most dashboards lead with total backlinks, which is the least informative figure on the screen. These four carry more signal:
- Referring domains, not backlinks. One template with a sitewide footer can produce thousands of links from a single domain. Domains are the unit that reflects independent editorial decisions.
- Anchor distribution. The share of exact-match commercial anchors is the number that tells you whether the profile looks engineered.
- Link attributes. Follow, nofollow, sponsored and UGC. Google treats these as hints, so a nofollow link is not worthless, but a profile made only of nofollows behaves differently from one with editorial follows.
- Change over time. Links gained and lost per month says more about momentum than any cumulative total.
| Number | What it answers | What it hides |
|---|---|---|
| Total backlinks | Almost nothing on its own | Sitewide duplicates, dead pages |
| Referring domains | How many sites vouch for you | Quality, relevance, whether the page ranks |
| Exact-match anchor share | Whether the profile looks natural | Which links you actually control |
| Net new domains per month | Whether the site still earns links | Whether the new links matter |
A diagnostic order that survives real sites
The order matters, because link work is expensive and easy to do in the wrong place. We run five steps and stop when one produces a decision.
1. Confirm the page’s job first
Open Search Console, go to Performance, and filter to the page. Does it get impressions? If a money page has no impressions at all, links are not the first problem — indexing or internal links are. We’ve seen a product page with a strong external profile that Google had never crawled, because nothing on the site linked to it.
2. Read the Links report for pages that get seen but not clicked
Impressions without clicks can mean weak titles, the wrong query, or an answer box taking the top of the page. Links rarely fix any of those. This step exists to rule links out cheaply before you spend a month on outreach.
3. Compare referring domains for the query you want
Pick one buyer query, find the pages ranking in the top five, and count how many distinct domains link to each. That gap is the only competitor link number with a clear interpretation. Aggregate domain-authority scores across a whole site are not comparable between vendors and shouldn’t drive a budget.
4. Read anchors for the three pages that matter
Export the anchors pointing at your top commercial pages. If the same phrase supplies most of them, the profile is telling a story you may not want. One site we took over had most of its external anchors pointed at a single product phrase, and no branded anchors at all. That is a pattern, not luck.
5. Only now look at spam scores
Third-party “toxicity” scores are heuristics, not Google data. They’re a sorting tool for a human review, and they should never be the reason for a bulk disavow. Google’s own guidance on link spam is the policy that actually applies.
Anchor text: what over-optimization looks like in data
A natural profile is boring. It is mostly the brand name, the bare URL, a paragraph of surrounding text, and a minority of descriptive phrases. The exact commercial phrase usually appears least often, because most people link to a company, not to a keyword.
You can’t rewrite anchors on sites you don’t own, so this check is mainly a brake. It tells you to stop building exact-match links when the share is already high, and it tells you when to use a broader mix in guest content. The anchors you fully control are internal ones, and those are covered below.
The disavow decision
Google’s position is that most sites never need the disavow tool, because its systems ignore spam links on their own. In practice there are two situations where we’d use it: a manual action that names unnatural links, or a sudden, obvious pattern of paid or network links pointed at one page at the same time. Both are visible in the data before you touch anything.
The reason to hesitate is asymmetry. Disavowing is slow to reverse and can remove links that were helping. Before submitting a file, we screenshot the pattern and note why each domain is in it. If the explanation is “the score was red”, it doesn’t go in.
Where the leverage usually sits: internal links
External links take months. Internal links can ship in an afternoon, and they’re the only part of the graph you own outright. In August 2026 we ran a pass that built internal links from already-indexed pages to pages Google had not indexed. Within four days the number of indexed pages had risen sharply, with no other change deployed in that window. The full method, including the traps, is in the 63-versus-0 indexing test.
Two rules came out of that work. Link from pages that are already indexed and topically close to the target, and use the target page’s real title as the anchor rather than a keyword you’d like to rank for. The second rule keeps you honest about what the page is actually about.
Reading a single link: four attributes, one decision
Link counts compress four separate questions into one number, and that’s the root of most bad decisions. When we look at one link, we score it on four attributes.
Relevance of the linking page. A page about industrial procurement linking to a page about custom metal parts is worth more than a general business blog doing the same thing, even if the second site is larger. Topical closeness is the part of a link you can judge without any tool score.
Placement. A link inside the body of an article is an editorial choice. A link in a footer, a widget, or a “partners” block is a template. The second kind is easy to get in bulk and easy for Google to discount, which is why sitewide links flatter the total and change nothing.
Does the target page deserve it? Check whether the linked URL is the page that answers the query. Links pointed at a homepage when the ranking page is a product page are common in export sites, and they’re wasted discovery signals for the page that needs them.
The neighborhood. One link from a page that also links to forty unrelated commercial sites reads differently from one link on a page with three outbound citations. You’re judging the company the page keeps, not just the page.
Backlinks are mostly invisible to AI answers
Answer engines quote sentences, not link graphs. When an AI Overview or an assistant names a company, it is drawing on text it can read and sources it can cite, and the number of external links behind a page is not part of that surface. A link can help a page rank well enough to be read in the first place, so the two are connected. They are not the same measurement.
That distinction matters for how you spend time. If the goal is to be mentioned inside an answer, the measurable inputs are the text on the page, the entities it names, and whether third-party sources describe you consistently. We counted this on our own properties: across 72 questions in a clean environment, AI answers named 18 companies, and our sites were cited three times. Each of those three citations came from a “how to choose a supplier” list, not from a recommendation. The coverage side of the same picture is in our measurement of 500 Chinese queries, where 80.9% returned an AI Overview.
Keep the instruments separate. A link audit answers “who links here”. It does not answer “what does an answer engine say about us”, and no backlink tool will ever fill that gap.
What link data cannot tell you
A backlink audit cannot establish causation. Two sites with similar profiles can rank a hundred positions apart, because links are one input among content, intent match, site structure, and brand demand. The tool’s composite authority score is proprietary, changes formula without notice, and shouldn’t cross a vendor boundary.
It also can’t see most of the places where brands now get mentioned. A forum answer or an AI Overview citation may carry no link at all. If you’re measuring visibility inside answer engines, a link report is the wrong instrument; that measurement needs its own method.
A one-page backlink audit output
The deliverable we find useful is not a spreadsheet. It is a short decision list. Each row names a finding, the action it triggers, who owns the action, and how long it should take.
| Finding | Decision it changes | Owner |
|---|---|---|
| Zero referring domains to a money page that ranks on page two | Start targeted outreach for that one page | Content or founder |
| One exact-match anchor above 50% of the profile | Stop building that anchor; diversify | Content |
| A linked URL now returns 404 | Redirect the old URL to the live page | Developer |
| Sitewide footer links from an unrelated network | Review for a disavow decision | SEO lead |
| Competitor gap on one buyer query | Build a target list of 20 relevant sites | Outreach |
If a row has no owner or no action, delete it. A backlink audit that produces 60 rows and no decisions is a report about the tool, not about the site.
How often to re-run it
Monthly is enough for the first-party Links report, because link profiles move slowly. A competitor gap review makes sense once a quarter, and a fresh gap set for a new page or a new market. Lost-link alerts are worth a look when they arrive, but only to check whether the linking page still exists. Rebuilding the same report every week is a way to avoid harder work.
The numbers on our own sites set the expectation. Over 16 months, our own sites recorded 543,305 impressions and 7,460 clicks, a 1.37% CTR at an average position of 15.8. Those totals say nothing about the quality of any single link, and no link report would have predicted them. What changed the site was fixing which page answered which query, and making the path from an indexed page to an unindexed one shorter.
The checks that decide anything are also the ones you can re-run yourself. Start with the pages that already get impressions, read the anchors aimed at them, and count how many independent domains are behind those links. Then look at the page’s job again. Most backlink work is really page-assignment work wearing a link costume.
For the wider checklist we apply to a site before touching links, see the 12 checks we run on every export site. The studies behind them are indexed on the research index.
Questions about checking backlinks
Is the total number of backlinks a useful KPI?
No. A single sitewide footer can produce thousands of links from one domain, so the total rewards templates rather than editorial choices. Track referring domains and net new domains per month instead, and read a page’s job before you read its link count.
Which backlink tool should I trust?
None of them completely. Each vendor crawls at its own speed and keeps its own history, so counts differ for the same site on the same day. Google Search Console is the only first-party view and it is a sample, not the full graph. Use it for direction and treat vendor totals as estimates.
Should I disavow every link with a low spam score?
No. Third-party toxicity scores are heuristics, not Google data. Google’s own guidance is that most sites never need the disavow tool. We would only use it for a manual action naming unnatural links, or a sudden obvious pattern aimed at one page.
Do nofollow links help rankings?
Google treats link attributes as hints, so a nofollow link can still be seen and can still drive qualified visitors. It will not pass the same signal as an editorial follow link, but it is not worthless either. Judge the linking page first, the attribute second.
Who wrote this and how to check it. Written by Liu Yu, founder of Tianwenseo (Henan Tianwenseo Technology Co., Ltd., registered 3 May 2017; Zhengzhou, China; UTC+8). Tianwenseo has built and maintained WordPress export websites and run Google SEO for B2B manufacturers since 2017, and currently maintains 30–40 client sites plus its own research domains. Every figure on this page is measured on a site we run, or linked to the source it came from. We don’t promise rankings or a number of enquiries. Contact: WhatsApp +86 135 2681 6415.