A new disavow file replaces the existing file submitted for that Search Console property. It is not an extra batch appended to a blacklist Google safely maintains somewhere else. Upload a partial list and omitted entries leave the submitted state. That fact turns a small text file into production configuration.
Before anybody uploads one, the team must establish that disavow is justified, confirm the exact property, preserve the current state, review every retained, added and removed entry and approve the complete replacement. Google describes its Disavow Links tool as an advanced feature which can harm Search performance when used incorrectly. Its guidance says most sites do not need it and sets a narrow threshold around a considerable number of spammy, artificial or low-quality links together with a manual action or likely manual action.
TL;DR
- A disavow upload is a complete replacement, so treat it as a release with a preserved previous state.
- A backlink export is discovery data. It cannot authorise a production change.
- Prove the link-spam pattern with provenance, not a toxicity score or an ugly domain.
- Use the narrowest scope that covers the evidence, and never promise a ranking recovery.
An ugly domain, a low third-party score or an unexplained traffic decline does not meet that threshold by itself. If somebody hands me a file and says “it only needs uploading”, the upload stops. The missing decision history is now the first problem.
The export is discovery data, not a release candidate
A backlink tool can discover links and classify them under its own model. It cannot authorise a production change. Use explicit dispositions for candidates:
| Decision | Use it when |
|---|---|
| Stop | Evidence consists only of aesthetics, automated risk scores, scraping or correlation with a traffic change |
| Investigate | A plausible scheme exists but provenance, property scope or scale remains unclear |
| Request removal or qualification | A policy-problematic placement is controlled by a reachable publisher or supplier |
| Disavow URL | Evidence concerns one exact source page and does not justify condemning the wider domain |
| Disavow domain | The evidenced manipulative pattern spans the relevant domain or subdomain boundary |
| Retain existing entry | A previous justified decision remains part of the desired complete state |
| Remove existing entry | New reviewed evidence reverses the old decision |
| Escalate or hold | Manual-action, inherited-file or property uncertainty makes a safe change impossible |
The domain count is deliberately absent from the decision rule. Ten evidenced paid-network domains may matter more than ten thousand scraper URLs the business never commissioned. Volume without provenance is a bigger unknown, not stronger evidence.
Prove the policy problem before choosing the remedy
The useful question is not whether a source “looks spammy”. It is whether the placement belongs to a pattern created primarily to manipulate rankings. Google’s spam policies describe link-spam patterns including paid links intended to pass ranking credit, excessive exchanges, automated links, low-quality directory links, unqualified advertorial links, distributed widgets or footers with optimised anchors and low-value content created for link manipulation. Properly qualified advertising and sponsorship are treated differently.
Strong evidence
- Contracts or invoices for followed links sold for ranking effects
- A known automated or network campaign commissioned for the site
- Repeated controlled commercial anchor text
- Publisher correspondence confirming paid placement
- A Search Console manual action concerning unnatural links
- An inherited campaign whose supplier, destinations and method can be reconstructed
- Documented removal attempts for artificial placements which remain live
Weak on its own
- A toxicity or authority score
- An unfamiliar country, language or top-level domain
- A cheap-looking page
- A sudden scraper cluster
- A link appearing after a ranking decline
- Exact-match anchor text without evidence of a controlled pattern
- A cleanup vendor recommending its own paid service
Weak evidence can open an investigation. It cannot close one. The companion diagnostic question is whether the link pattern presents a real current risk. This article starts only when that threshold has been addressed. It does not use the upload procedure to launder an uncertain diagnosis.
A manual action changes the evidence burden
If Search Console reports an unnatural-links manual action, work from the notice and its affected scope. Google’s Manual Actions report guidance says manual actions follow human review against spam policies. For unnatural links to a site, its guidance includes reviewing the policy, investigating link data, seeking removal or appropriate qualification, disavowing links that could not be removed and documenting the work in a reconsideration request.
That creates two connected but different records: the machine-readable disavow file, and the human-readable cleanup evidence supporting the remediation and any reconsideration request. Do not force the second record into # comments. Comments can orient a reviewer, but Google ignores them for disavow processing. The evidence ledger must preserve contacts, responses, provenance, decisions and approval outside the upload format.
If no manual action exists, do not infer one from a traffic chart. Visibility can change because of demand, technical faults, migrations, content quality, competition or algorithmic reassessment. The absence of a manual action does not prove every link is harmless, but it removes the strongest direct trigger and raises the burden for claiming one is likely.
Search Console does not export the backlink universe
The Search Console Links report can support discovery. It does not output Google’s verdict on which links should be disavowed. Google’s Links report documentation describes an incomplete view with grouped canonicals, combined duplicates and interface limits; reported links can also include historical discoveries. Keep those boundaries attached.
Use exports to identify sources for investigation, compare patterns with known campaign dates and preserve a dated view. Do not upload Latest links or sample exports wholesale. A third-party crawler can widen discovery, but its finding proves only that the crawler observed a link under its conditions, not that Google counted, trusted or distrusted it. This is a recurring SEO failure: a discovery surface becomes a decision surface because the CSV already looks like work.
Build a ledger richer than the text file
The review system needs more information than the upload format can contain.
| Field | Record |
|---|---|
| Candidate source | Exact linking URL and relevant domain |
| Target | Page on the reviewed site receiving the link |
| Discovery source | Search Console, supplier file, crawler, manual action or manual discovery |
| Observation | Date, live state and preserved evidence |
| Placement provenance | Paid, exchanged, automated, hacked, editorial, user-generated or unresolved |
| Control history | Who commissioned or managed it, where known |
| Policy rationale | Exact link-spam pattern supported by the evidence |
| Removal action | Contact, response, removal, qualification or reason contact was not viable |
| Scope decision | Keep, URL, domain, investigate, remove existing or escalate |
| Existing-file state | Present, proposed addition, proposed withdrawal or unknown |
| Reviewer | Accountable reviewer and date |
| Limitation | Missing evidence or unresolved conflict |
The existing-file field is crucial. A source absent from the new investigation may remain in the currently submitted file because it came from an earlier campaign. An old entry may deserve removal, but that reversal needs a reason rather than silent disappearance. Ten thousand processed domains are not impressive if nobody can explain the first active line.
URL scope and domain scope carry different risks
A URL entry asks Google to disregard links from one exact source page. Use it when the evidence concerns an isolated placement and the wider domain contains other material the investigation does not condemn.
https://publisher.example/archive/paid-widget-review.htmlThe entry is the source page linking to the reviewed website, not the target page on the website. Preserve the exact path; paths may be case-sensitive and parameters may distinguish pages. Mechanical “cleanup” can produce a tidy file containing altered sources. A domain entry uses the domain: prefix:
domain:paid-network.exampleDomain scope is efficient and broad. Ask whether the problematic pattern genuinely spans the source, whether a network or sitewide arrangement is evidenced, whether independent editorial links may coexist and whether a subdomain or root domain is the defensible boundary. Do not use domain scope as punishment for a publisher or website the business dislikes. The file is not a reputation blacklist. It asks Google to disregard link signals from the submitted source when assessing the selected property.
My preference is the narrowest scope that fully covers the evidenced pattern. Precision is not timid when the alternative can exclude legitimate references.
Been handed a “toxic links” export and told it just needs uploading? We reconstruct the decision record before anything reaches Search Console.
Preserve the existing submission before editing
Before preparing a candidate, identify the exact Search Console property and establish whether a disavow list already exists. Preserve:
If the current file cannot be obtained or reconstructed, stop before uploading a partial “new bad links” list. Missing production state is not administrative untidiness. It is an unknown configuration with asymmetric risk: replacement may withdraw earlier justified entries or retain harmful overreach nobody understands.
The upload candidate must represent the complete desired state. The change set explains how it differs from the previous state. That same distinction makes rollback intelligible: rollback is another approved complete replacement, not “put the old file back and hope”.
Make the syntax boring and the decisions difficult
Google’s documented format requires one URL or domain per line, domain: for domain entries, supported text encoding, a .txt file and compliance with published size and line limits. Optional comments begin with # and are ignored during processing. Check the current official guidance before release because product requirements can change. A compact file could look like:
#Property: https://www.example.com/
#Approved change: DIS-014
domain:paid-network.example
https://publisher.example/archive/paid-widget-review.htmlBefore review, remove spreadsheet debris such as quotation marks, comma-separated columns, headings, formulas and target-site URLs. Deduplicate exact entries. Check whether URL entries are already covered by approved domain entries. Never widen URL decisions to domain scope merely to shorten the file. The syntax should be boring. All controversy belongs in the evidence ledger and change review, where another person can challenge it.
Parser acceptance is not safety approval
Validate four layers.
Use a second reviewer for exactly one question, and make it the one that matters:
The important question is not “Does this parse?” It is “Which legitimate link could this line cause us to ask Google to ignore, and what evidence justifies that risk?”
Treat upload and reversal as controlled releases
The whole process, from justification to observation, is a release:
- Establish that disavow is justified
The narrow threshold: a considerable number of spammy, artificial or low-quality links together with a manual action or likely manual action. Not an ugly domain or a traffic chart.
- Confirm the exact property
Protocol, host, property type, active account and observed permission in the live tool.
- Preserve the current state
Previous file, hash, counts, submission date, approver and a screenshot. If it cannot be recovered, stop or escalate.
- Review the complete change set
Every retained, added and removed entry maps to a ledger decision with provenance, policy rationale and scope.
- Approve the complete replacement
The candidate is the whole desired state, checked at file, set, evidence and property level by a second reviewer.
- Upload, record and observe
Preserve the receipt, time, account, file hash and counts. If the tool errors, confirm the previous state remains. Success is a submission, not a recovery.
Google says incorporation can take time while it recrawls and reprocesses pages. Links may continue to appear in reports. Their presence does not prove the file failed. Removing one entry also requires a complete replacement without that entry. Cancelling all disavowals is another consequential decision. Before reversal, record why the entry existed, what new evidence changes the decision, which boundary is affected, who approved it and what will be observed afterwards.
Do not promise that adding or removing entries will restore a ranking, traffic or customer-acquisition result. Search systems and link graphs change while the file is processed. Repeatedly swapping files is not a clean experiment.
An inherited file is a governance problem first
An undocumented inherited file may contain careful remediation, indiscriminate agency output or both. Deleting it is not automatically safer than retaining it. Re-uploading it unchanged is not automatically responsible. Preserve the state, identify former owners and suppliers, recover campaign records, sample active URL/domain decisions and separate defensible entries from unresolved ones. Escalate high-impact uncertainty instead of guessing.
The honest decision may be hold. That is appropriate when removal could reintroduce an evidenced risk and retention could continue excluding legitimate signals, but the organisation lacks the records to choose. A new SEO agency should not overwrite the file to demonstrate activity. Its first useful deliverable is the reconstructed decision record.
Make the owner approve the complete state
Before release, the accountable owner should be able to confirm:
If one answer is no, the next action is not Upload. It is the unresolved line above it. The account owner’s job is not to perform a specialist’s click. It is to refuse a production change until another person can reconstruct what the complete file asks Google to disregard, why every consequential line belongs there and what changed from the state it replaced.
Good to know
FAQs
No. A new file replaces the existing file for that Search Console property. Any entry omitted from the new file leaves the submitted state. That is why the candidate must be the complete desired state and the previous file must be preserved before editing.
Google says most sites do not need it and describes a narrow threshold: a considerable number of spammy, artificial or low-quality links together with a manual action or a likely one. A toxicity score, an ugly domain or a traffic decline does not meet that threshold on its own.
Use the narrowest scope that fully covers the evidenced pattern. A URL entry suits an isolated placement on a domain that otherwise carries legitimate material; a domain entry needs evidence that the manipulative pattern spans the whole domain or subdomain.
Treat it as a governance problem first. Preserve it, recover former owners and campaign records, sample the active decisions and separate defensible entries from unresolved ones. Holding is a legitimate decision when the records to choose safely do not exist.
Next step
Inherited a disavow file or a toxic-links export?
We reconstruct the decision record, preserve the current state and take the complete replacement through file, set, evidence and property checks before anything is uploaded.
