Toxic backlinks are increasing, and AI is the reason
We’ve doubled the frequency of our backlink audits across most client accounts this year. Not because our process changed, but because the volume of toxic backlinks landing on client sites has gone up sharply, and the cause is generative AI making spam link-building fast and cheap.
Building a spam network used to take real effort: registering domains, writing content, designing layouts, placing links by hand. AI has removed most of that friction.
Mass domain creation
Attackers can spin up hundreds of lookalike or low-quality domains, with plausible copy and cloned layouts, in hours rather than weeks. Research from brand-protection firm Abion recorded a 140% surge in malicious domain registrations in early 2025, alongside AI-written phishing emails achieving a 54% click-through rate against 12% for human-written ones. Different threat to backlinks specifically, but it shows how far the cost of producing convincing fakes has dropped.
Private blog networks, industrialised
PBNs (clusters of low-quality sites built to cross-link and inflate authority) used to be labour-intensive to run. Scripts and language models now build and refresh them with far less manual work, which is why toxicity scores on client accounts are climbing even when nobody at the client end has changed anything.
Spam that reads like a person wrote it
Forum and comment spam used to be obvious: crude, keyword-stuffed, easy to filter. AI bots now write conversational replies on forums, Reddit threads and Q&A sites that are harder to flag, and increasingly aimed at shaping what AI tools say about a brand, not just at manipulating traditional rankings.
Why this hits harder in AI search
Google has a formal process for distancing a site from toxic backlinks: the disavow file. AI search tools like ChatGPT, Perplexity and Google’s AI Overviews don’t have an equivalent. If a coordinated backlink campaign confuses which source is authoritative on a topic, there’s no upload-and-wait fix. The tools are making split-second judgements about source credibility with none of the anti-spam infrastructure Google has built up over two decades.
| Traditional Google spam | AI search citation confusion | |
|---|---|---|
| Formal appeal process | Disavow file via Search Console | No standard mechanism yet |
| How it’s usually caught | Manual action notices, ranking drops | No dashboard; has to be caught by checking what AI tools actually say |
| Recovery once addressed | Typically within a few months | Untested at scale, AI search itself is still new |
Why construction, property and manufacturing sites carry more of this risk
Buyers researching a local contractor, a housebuilder or an industry specialist increasingly start that search inside ChatGPT or an AI Overview rather than a list of blue links. Trust in that first answer matters more than trust in a page-two Google result ever did.
Sites in this sector also tend to share a few traits that make them easier to crowd out or impersonate:
- Thin content beyond core service pages
- Little trade press or news coverage
- Generic company names that are hard to tell apart from competitors
- Few authoritative sites linking in, outside of niche trade directories
The sector also has a defence most industries don’t: accreditations. A PCA membership, an FMB or NHBC accreditation, a case study with a named site and date, all do double duty. They are E-E-A-T signals Google already rewards, and they’re the kind of independent, verifiable signal that’s much harder for a lookalike site to fake.
Defending a site against this
- Audit more often. Quarterly is too slow for most sites now. Monthly, or fortnightly for anything in a competitive niche, catches a spike before it compounds.
- Don’t treat a high toxicity score as an automatic disavow list. Semrush’s own guidance on toxic links is blunt about this: disavowing unnecessarily can do more harm than good. Try removal requests to the referring site first, and reserve disavow files for cases with a Google manual action or clear evidence of past manipulation.
- Watch what AI tools say about the brand, not just the backlink profile. If ChatGPT or an AI Overview starts citing a competitor, a directory listing, or a low-quality clone site as the authority on a service the client offers, that’s worth catching early, and it won’t show up in a standard backlink audit.
- Lean on accreditations and named case studies. They’re slower to fake than a backlink, and they give both Google and AI tools something concrete to verify a brand against.
- Build the review into a schedule with a clear escalation point, rather than an occasional deep dive when something looks off. The sites getting hurt by this are the ones nobody’s watching.
FAQs
Are toxic backlinks something we caused, or can they land on a site without us doing anything?
Almost always the latter now. AI has made it cheap enough to target a domain with toxic backlinks regardless of what the site owner has done, which is a big part of why audit frequency matters more than it used to.
Should we disavow every toxic backlink our audit finds?
No. Request removal from the referring site first. Disavowing is for confirmed manual actions or clear evidence of past manipulative link-building, not a routine response to a toxicity score.
How often should a construction or property site check its backlink profile now?
Monthly as a baseline, fortnightly if the site sits in a competitive or high-value niche.