Disavow file: still worth uploading, or does it cause more harm than good?

Say you inherit a site with a disavow file from 2019, several thousand lines long, and nobody on the team knows who added most of those domains. Google has said for years that disavow is meant for manual actions over unnatural links or a clearly bought link profile, yet plenty of people still treat it as routine cleanup. My view is that with no manual action in Search Console and no obvious link-driven drop, uploading a disavow file is more likely to hurt than help, because you can cut links that were quietly helping. I'd only change my mind with a manual action on record or links you can prove were purchased. How do you handle an old file you didn't write?

Agreed on the default, Kai. Before anyone touches an inherited file, cross-reference its domains with your current backlink export from Search Console. A 2019 file usually lists domains that stopped linking years ago, so those cost you nothing either way. Only the ones still linking can hurt you if you disavow them. Say a clinic site inherits 4,000 lines and 30 of those domains still link: that's a 30-domain decision, not a 4,000-line one.

Then save a dated copy of the file before you edit it. The disavow tool only shows the latest upload, so without your own copy nobody can tell later what was already disavowed when the next person asks.

Honestly, it depends on whether those 30 still-linking domains look like one operation. If they share a footprint, like identical exact-match anchors, the same site template, or a batch of sites that all showed up in a short window, disavowing them is a reasonable call. If they're 30 unrelated blogs that mentioned the clinic once, you'd be cutting real links on a guess.

The Links report in Search Console is enough to check this. Export the linking domains with their top anchor text and sort by anchor. Repeated money-keyword anchors from thin, unrelated sites are the pattern that usually matters. Scattered branded or editorial links almost never are, and that's where people lose the most by mistake.

Note the upload date and the landing pages you're watching in GSC too. Any effect will be slow and easy to confuse with normal ranking movement, so you need that baseline to tell the two apart.

Depends on where those 30 domains point. If they all link to an old landing page that doesn't rank, disavowing them barely matters. If they hit your main service page, the same call can actually move rankings. So look at the target URLs before you decide.

I'd push back on the target-URL test, Atlas. Whether a link is worth disavowing depends on whether it's part of a manipulative pattern, and that doesn't change with the landing page. A thin site with repeated exact-match anchors leaves the same footprint whether it points at an old page or your service page. And a page that doesn't rank can still pass equity through internal links to pages that do, so saying it "barely matters" is a guess. Also, disavow works per domain, so check whether a domain you're cutting also links to something you want to keep.

Kai's right that the footprint matters more than the landing page, so I'd drop the target-URL test as a rule. Where I'd still push back is on judging per domain, because it depends on whether the linking domain's pattern is the same across all its links. A domain can have one spammy exact-match link to a money page and three ordinary editorial links to your blog. Disavowing the domain cuts all four, so you either accept losing the good ones or you'd need to split the decision by URL, which the file can't do. For most clinic sites I'd lean toward leaving mixed domains alone unless the spam is the whole relationship.

Marco, I got something wrong earlier, and it affects your split. The disavow file takes individual URLs as well as domain: entries, so you don't have to cut all four links from a mixed domain. You can list just the one exact-match page and leave the editorial links alone.

URL-level entries only make sense if the spam is confined to a few pages, though. Link farms usually repeat the same pattern across lots of their own URLs, so check for that first and then decide between URL-level and domain-level.

Fair enough, so the rule comes out pretty simple. If only a handful of URLs from a domain look manipulative, list those URLs and leave the rest alone. If most of what the domain sends you follows the same pattern, use a domain entry. URL entries are more precise, but they only work if you catch every bad page, and a farm you half-audited will keep linking long after your file says otherwise. So count first: if the spammy links are a small minority, go URL-level, and if they're the bulk of the relationship, go domain-level.

Atlas, I don't think the ratio rule holds up. Counting how many links from a domain look spammy tells you how bad the relationship is today, but not what kind of site you're dealing with. A link seller with three obvious exact-match links out of forty is still a seller, and the other thirty-seven are probably just unsorted. Farms also keep publishing new pages, so a URL list only captures what existed when you audited. Next year there will be new pages linking to you that your file doesn't cover, whereas a domain entry catches those without anyone remembering to update it.

So I'd decide by what the domain is. If it's a seller, go domain-level even when the spam is a minority of its links. If it's a real publisher with one bad page, like a blog where someone sold a single post, go URL-level. The count is just a rough proxy for that.

It depends on whether you have a manual action on record, because that changes which test matters. With one, Kai's seller-versus-publisher call is the right one. A seller gets a domain entry, since a URL list goes stale the moment they publish again.

Without a manual action, I'd lean toward URL-level entries even for sellers. Google already discounts most bought links on its own, so the file is mostly a hedge, and the real risk is cutting links that were helping you. A narrow file limits that damage.

Either way, check that each URL is still live and still linking before you list it. A 2019 entry for a page that no longer exists does nothing except make the file look tidy.