Link Building for AI Search: Why $500 Backlinks Fail

If you are weighing where to spend on visibility, spend it almost anywhere but a $500 backlink package. Link building for AI search still matters at the margins, but the specific unit that agencies sell most often, a guest post or link insert priced on the domain authority of the placement, buys you almost nothing an AI system will act on. Links still clear a baseline authority threshold. The high-DR package sold on top of that threshold is what Gaetano DiNardi, writing in Search Engine Land, calls fake authority: a number on a vendor’s spreadsheet that does not move whether ChatGPT, Perplexity, or Google’s AI Overviews decide to recommend your brand.

That distinction is the whole argument, so it is worth stating precisely before the list. This is a summary of DiNardi’s reporting, and the practical breakdown of it that follows.

Link building for AI search: what the $500 package actually buys

Domain Authority is a third-party score, not a Google or LLM signal. A vendor sells you a placement on a DR-70 site because the number is high, but as DiNardi argues, a high DA figure is often an illusion. It does not prove the site is legitimate, that the page drives any organic traffic, or that an AI system will ever treat that backlink as a source when deciding what to recommend. You are paying for a proxy that the buyer and seller both agreed to care about, on a page a language model may never retrieve.

The reason the model shrugged is structural. Answer engines are not counting your links the way early PageRank did. They assemble a picture of which entities are credible from many signals at once, and a paid link dropped into an irrelevant guest article is a weak one. The link exists. The credibility it is supposed to represent does not.

Links still clear a baseline, and that is the trap

Here is the part that keeps the guest-post market alive: links are not dead. Without some baseline of authority, you cannot clear the competitive SEO thresholds that get a page indexed and ranked at all, and ranking still gates AI visibility. The retrieval pool that AI Overviews draw from is heavily weighted toward pages that already rank. So a founder reasons, correctly, that links matter, and then incorrectly concludes that buying more of them helps.

The trap is that the baseline is a floor, not a dial. Once your site clears it, the next dollar of DR-priced links produces almost no marginal credibility. Ahrefs’ analysis of 863,000 keywords found that only 38% of AI Overview citations now come from pages ranking in the top 10, down from 76% (Search Engine Journal, 2026). AI systems are increasingly citing sources that rank poorly or not at all, which means the thing you bought the links to buy, top-10 ranking, is a shrinking share of what earns the citation.

What actually earns AI credibility instead

If the DR-priced link is fake authority, real authority is the set of signals an AI system can independently corroborate. These are the places to move the budget, drawn from DiNardi’s argument and the surrounding reporting:

• Unlinked brand mentions in the right places. Being named on Reddit, YouTube, and industry communities is growing into a signal on par with a traditional backlink, because it is corroboration a model can find without a paid placement. Reddit alone accounts for roughly a fifth of AI Overview citations by several 2026 counts.
• Real content that earns links as a byproduct. DiNardi’s point is that the best links are not rows in a database to be approved one by one; they are the natural result of publishing something worth citing. That reverses the spend: fund the asset, not the placement.
• Entity consistency. Consistent name, description, and facts across your site, your profiles, and third-party sources help a model resolve you to a single trusted entity rather than a fuzzy one.
• Verifiable authorship. Named authors with checkable credentials give an AI system a reason to treat your claims as expert rather than anonymous.
• Outbound citation. Pages that link out to primary sources read as participants in the web’s citation graph, which we broke down in our outbound linking strategy for AI search.

Notice that none of these is a thing you can buy for $500 from a vendor database. That is the point. They take work the DR package was sold to let you skip.

Fake authority vs real authority, in one test

If you want a single test to separate fake authority vs real authority, ask whether the signal survives being checked. A DR-70 link on an irrelevant page fails: click through and there is no audience, no topical fit, and no reason a model would retrieve it. A brand mention on a relevant subreddit passes: it is a real person referencing you in context that an AI system can read and corroborate. Fake authority is a number that only exists inside a tool. Real authority is corroboration that exists out on the open web where models actually look.

This is also why rank-tracking dashboards mislead here. A page can hold position one and still lose the AI citation, because being seen and being ranked have come apart, a gap we measured in AI search visibility versus organic rankings. Buying links optimizes the ordinal rank. It does not optimize the corroboration that decides the recommendation.

Choosing a GEO agency without buying fake authority

The market is repositioning fast, and plenty of link vendors have simply relabeled the same product as GEO agency LLM visibility work. The tell is the deliverable. If the proposal is priced per placement and reports on domain authority or DR of the sites secured, it is the old guest-post model in a new name, and it will produce the same fake authority. Ask instead what corroboration they will create: earned mentions in communities your buyers actually read, digital PR that lands in publications a model trusts, content assets that attract citations on their own merit. Those are harder to sell by the unit, which is exactly why they are worth more.

Brand credibility in AI recommendations is downstream of being genuinely, checkably good, and no line item shortcuts that. The right question for a GEO engagement is not how many links per month, but how many places a skeptical model could independently confirm you are what you claim to be.

Where to spend instead

Take the $500 you were going to spend on a link and spend it on being citable. That means the assets and signals above, and it means the engineering underneath them: consistent schema, clean structure, verifiable authorship, and generous outbound sourcing shipped as defaults rather than left to each writer. Much of that is a configuration decision made before the first word is written, a point we made in how context shapes AI agent output. If you are standing up a site and want real authority baked in from the first post rather than bolted on with purchased links, that is the layer to get right, and the rest of our guides on AI search and authority go deeper on the build patterns behind it.