For the better part of two decades, link acquisition has been the most measurable activity in search marketing, which is why the arrival of AI-generated answers has raised an uncomfortable question inside many marketing departments: whether the link profile a company spent years building carries any weight when an answer is assembled by a language model rather than ranked by an algorithm. The evidence indicates that links still matter, though in a narrower and somewhat different way than most link-building programs assume, and the difference has direct consequences for how placement budgets should be evaluated.
Part of the confusion comes from treating AI search as a single system. Google’s AI Overviews, ChatGPT, Perplexity, and Gemini retrieve and select sources in different ways, and the published research shows the relationship between link metrics and AI visibility varying considerably across them. This article sets out what links continue to do, what the available studies do and do not establish, which characteristics of a link carry over into AI visibility, and how a company can assess its own profile on that basis.
What Backlinks Still Do in an AI Answer Pipeline
Answer engines that draw on a conventional search index inherit part of that index’s judgment about which pages deserve attention, and links remain one of the oldest inputs to that judgment. Google’s AI Overviews are assembled from pages that Google’s own systems have already crawled, indexed, and ranked, so a page with a weak link profile begins at a disadvantage before any model considers it as a source. The same logic applies, in varying degrees, to assistants that retrieve from other indexes: the retrieval step narrows the field of candidate sources, and only the candidates that survive it are available to be cited.
A second function is discovery. Crawlers find most pages by following links from pages they already know, which means a placement on a frequently crawled publisher is likely to be encountered sooner and revisited more often than a page that nothing points to. For a company whose most useful material sits on its own domain, inbound links from established sources are among the more dependable ways of ensuring that material is within reach of a retrieval system.
The third function is the one most often overlooked, and it is descriptive rather than structural. A link arrives attached to words: the anchor text, the sentence around it, and the page it appears on. When a publisher reports that a company has launched a product and links the company’s name to its website, the link and the mention are a single act, and the surrounding text supplies the kind of independent description a model draws on when asked what a company does. Ahrefs’ analysis of AI Overview brand visibility captured part of this overlap, since branded anchor text, a link-based signal, ranked second among the factors it examined, behind only branded web mentions.
What the Available Research Does and Does Not Show
The most widely cited analysis comes from Ahrefs, which studied 75,000 brands with a Domain Rating above 40 and measured, using Spearman correlation, how closely a range of signals tracked each brand’s visibility in Google’s AI Overviews. The signals most closely tied to a brand’s name and reputation produced the strongest results, and the measures of link quantity produced the weakest.
| Signal | Correlation with AI Overview brand visibility | What it measures |
|---|---|---|
| Branded web mentions | 0.664 | Independent references to the brand across the web |
| Branded anchors | 0.527 | Links whose anchor text contains the brand name |
| Branded search volume | 0.392 | Audience demand for the brand by name |
| Domain Rating | 0.326 | Site-level link authority |
| Referring domains | 0.295 | Number of distinct sites linking to the domain |
| Number of backlinks | 0.218 | Total volume of inbound links |
Source: Ahrefs, An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied), published May 2025. Figures are Spearman correlations.
Read at face value, the table shows that signals tied to a brand’s name outperform signals that measure link quantity, with raw backlink count near the bottom of the factors studied. Ahrefs itself cautions that correlation does not establish causation and that every factor it examined fell between moderate and very weak by conventional standards, so the figures describe a pattern rather than a mechanism. The study also measured brand visibility in AI Overviews, which is a different outcome from whether a particular page’s link was cited.
A separate analysis by SALT.agency, published in December 2025, examined 5,825 URLs from travel-industry sites and compared link metrics against citation counts on four platforms. The strength of the relationship differed by platform: Domain Rating showed a weak correlation (0.25) with citations in Google AI Overviews, while backlinks showed a moderate correlation (0.39) on ChatGPT and referring domains showed moderate correlations on Perplexity (0.42) and Gemini (0.41). The sample is confined to one industry and the authors acknowledge they could not cover every AI platform, which limits how far the results generalize, yet it reinforces a point practitioners should already take seriously: AI search is several systems with different retrieval and citation logic, and a single link-based score cannot summarize them.
Why Link Volume Stops Differentiating Past a Threshold
Two features of these studies, read together, point to a pattern that neither was designed to demonstrate. The Ahrefs sample was limited to domains with a Domain Rating above 40, so every brand in it already carried a meaningful link profile, and within that group link volume explained comparatively little of the variation in AI visibility. The SALT.agency data showed that no domain with a Domain Rating of 40 or lower appeared in the top quartile for total AI citations, while 18.5 percent of domains rated 80 or higher fell into the lowest quartile. These observations are consistent with links acting as a threshold for eligibility more than as a differentiator among eligible candidates, although the data cannot establish this, and the interpretation offered here is ours rather than the authors’.
That reading is consistent with a situation familiar to most practitioners, in which two competitors with similar link profiles receive different treatment from AI assistants, one named when a user asks who leads a category and the other absent. In those comparisons the deciding factor tends to be the number of independent sources describing the company in running text, the consistency with which they name it, and whether those sources are ones answer engines already draw on, with referring-domain counts playing a smaller part. This is the territory covered in our analysis of publisher mentions, and it explains why the two signals are best evaluated together.
The practical consequence is that additional links beyond the threshold are likely to produce diminishing returns, while a placement that names and describes the company in editorial context continues to add value on the mention side. A link from a page that never states what the company does is a weaker asset in this environment than the same link embedded in a paragraph that does, which is one reason editorial placements in relevant publications tend to be a better measure of progress than link volume.
Which Link Characteristics Carry Over and Which Do Not
Not every attribute that mattered in a link-building program transfers to AI visibility with equal force. The characteristics most likely to carry over are those that make a link a credible, descriptive endorsement: topical relevance between the publisher and the company, an editorial context in which the company is named and characterized in running text, a publisher that is itself crawled and referenced frequently, and permanence, since a placement that is removed or de-indexed contributes nothing to a retrieval system.
Several attributes that dominated older evaluation frameworks carry less weight. Exact-match keyword anchors communicate little that a descriptive sentence does not communicate better, and branded anchors were the stronger correlate in the Ahrefs data. A single Domain Rating figure compresses site-level authority into one number that says little about whether a publisher covers a company’s category. Sitewide and footer links, directory listings, and placements on sites that publish indiscriminately offer little descriptive context, which is the property AI systems appear to rely on.
The follow attribute deserves separate mention. It governs how Google’s ranking systems treat a link, and sponsored placements should carry the appropriate rel attribute in line with Google’s published guidelines. It does not change whether the surrounding text names and describes the company, so a placement marked nofollow or sponsored can still contribute on the mention side even where it contributes little on the ranking side. Evaluating placements on that basis, rather than against a do-follow checklist, is closer to how answer engines appear to use the material they retrieve.
Auditing a Link Profile for AI Visibility
A conventional link audit sorts referring domains by an authority metric and screens for harmful links. An audit oriented toward AI visibility asks different questions of the same data, and most of them can be answered from a backlink export combined with a modest amount of manual review.
- Separate links that sit in running text naming and describing the company from links in footers, directories, and author boxes, and record the proportion of each.
- Review the anchor text distribution, looking for a healthy share of branded and descriptive anchors rather than a concentration of exact-match commercial phrases.
- Identify the publishers behind the strongest links and test whether they appear as sources when category questions are put to ChatGPT, Perplexity, and Google’s AI features, using the approach described in our guide to measuring AI search visibility. A publisher that never appears in those answers is a weaker prospect than its authority score implies.
- Check topical alignment between the linking pages and the company’s category, since a link from a respected but unrelated publication says little about what the company does.
- Confirm that linked placements remain live, indexed, and unaltered, and flag any where the company’s name has been removed or the page has dropped out of the index.
The result shows how much of the link profile also functions as independent description, which is the portion an answer engine can use, and it turns the next investment into a question about which relevant publishers are missing from the profile. A full AI website authority audit performs this analysis across link, mention, and entity signals together, but the five checks above can be run internally with a standard backlink tool and a few hours of manual review.
Where This Fits Into a Broader Authority Strategy
Links work best as one component of a connected set of signals. Schema markup and entity clarity determine whether a model can interpret who a company is, original research gives it something new to cite, and publisher mentions establish that independent sources agree on the description. Backlinks contribute at the first stage of that chain by helping the right pages be discovered and retrieved and, when carried by editorial placements, by supplying the descriptive text on which the later stages depend.
For companies deciding where to direct placement budget, the implication is to evaluate publishers on relevance and citation frequency before authority score, and to treat the linked mention, rather than the link alone, as the unit of value. Where an audit shows that relevant publishers are missing from the profile, closing that gap through editorial placements and PR distribution is the logical next step, provided each placement is judged by the criteria above.
A company’s link profile is difficult to judge in isolation, because its value depends on how it compares with the profiles of the competitors an AI assistant is choosing among. Visionary Financial’s AI Authority Audit evaluates backlink quality alongside entity consistency, structured data, content originality, and publisher signal in a single assessment, which shows whether a company’s links are reinforcing one consistent identity or sitting beside coverage that never mentions it. For a business investing in both link acquisition and media placements, that confirmation determines whether the two efforts compound or run in parallel.
