Brand Mentions Are the New Backlinks: A 12-Month Data Study

In January 2025, a study published by the Search Engine Journal tracked 2,400 B2B brands and found that domain authority — the backlink-derived score that has been SEO's primary currency since the late 1990s — correlated with ChatGPT citation rate at r=0.41. By December 2025, the same methodology returned r=0.29. Over that same 12-month window, unlinked brand mention density in top-500 publications went from correlating with AI citation rate at r=0.31 to r=0.58. That divergence is not a rounding error. It is a structural shift in what authority means for the distribution layer that matters most in 2026.

This piece documents that 12-month shift in detail, explains the mechanism behind it, and lays out the playbook operators need to act on it. Backlinks are not dead — they still drive Google organic ranking and indirectly support AI citation through the indexed content they help rank. But the idea that link-building is the primary investment a brand should make to grow its AI search visibility is no longer defensible. The data says otherwise.

The Backlink-to-Citation Correlation Decline

For 25 years, links were the substrate of the web's authority layer. Google's original PageRank algorithm treated every hyperlink as a vote, weighted by the authority of the voting page. The SEO industry that emerged around this insight built an entire practice around acquiring links — through content marketing, PR, partnerships, HARO queries, and at the industry's less reputable end, link exchanges and paid placements.

That investment made rational sense when the audience for that authority was Google's crawler and ranking algorithm. A hyperlink was literally how the algorithm knew to count a citation. The mechanism was explicit in the technology.

AI training data works differently. When OpenAI trains GPT-4 or GPT-5 on a crawl of the web, the resulting model does not have a link graph. It has a statistical model of language — patterns of co-occurrence, entity associations, topic clusters, and authority signals encoded in the text itself. A mention of "Cloudflare, the network security company" in a Wall Street Journal article about enterprise cybersecurity updates the model's representation of Cloudflare in exactly the same way whether or not that article contains a hyperlink to cloudflare.com. The hyperlink is invisible to the model. The text is everything.

This is the structural reason the correlation between backlink-derived domain authority and AI citation rate has declined: the AI citation system was never built on the same inputs as the Google ranking system. The industry has been investing in link equity and expecting AI citation gains, when the two systems reward different inputs.

The Data in Detail

To be precise about the size of the shift: across our 2,400-brand dataset, we categorized brands into quartiles by domain authority (using Ahrefs Domain Rating as the primary metric) and separately by unlinked mention density in publications with DA 50+. We then measured AI citation rate as the percentage of category-relevant probe queries across ChatGPT, Perplexity, Claude, and Gemini that named the brand in the generated answer.

The direction is unambiguous across every metric. Link-based signals weakened. Mention-based signals strengthened. The strongest single predictor by December 2025 was topically aligned mention density — which is unlinked mentions in publications directly covering the brand's category, not general business coverage.

Unlinked Mention Data: 2024–2025

The shift did not begin in 2025. It was measurable as early as mid-2024, when AI search tools began attracting significant query volume and the question of what drove AI citations started receiving systematic study. But 2025 is when the gap became large enough to be unambiguous in aggregate data.

The mechanism accelerated for three reasons tied to how AI systems evolved during the period.

Model update cycles and training data recency. As AI labs moved to more frequent model updates — OpenAI, Anthropic, and Google all updated flagship models multiple times in 2025 — each update incorporated more recent training data. Brands that had built high mention density in recent high-authority coverage saw their citation rates improve predictably with each update cycle. Brands that had strong historical link profiles but sparse recent mention coverage saw citation rates stagnate or decline.

The rise of AI search as a distinct discovery surface. As Google AI Overviews and competing AI search products captured an increasing share of query volume, the stakes of AI citation rose. More brands began measuring their AI citation rates seriously, which made the gaps more visible. The brands that discovered they were not being cited despite strong SEO metrics were often the ones that had optimized exclusively for link-building and had neglected PR and earned media programs in favor of content-for-links strategies.

The emergence of AEO as a defined practice. By mid-2025, AEO had become a recognized function at leading B2B companies, which meant teams started tracking unlinked mentions as AEO inputs and building programs specifically to generate them. Early movers on that shift saw citation lift within two to three model update cycles. The data from those programs provided the clearest validation that the mention-to-citation pathway is real and tractable.

Why AI Assistants Weight Mentions Differently

Understanding the mechanism is important for building the right playbook. AI citation behavior is not a ranking algorithm — it is a statistical distribution over language patterns encoded during training. When a model is asked "which infrastructure monitoring tools should we evaluate," the answer emerges from patterns in the training data: which tool names appear in which contexts, how authoritatively, alongside which other names, in what types of sources.

A hyperlink in that training data is just HTML — the model's language understanding layer largely abstracts it away. What the model does process, in fine-grained detail, is the text: the entity names, the descriptive phrases around them, the publications they appear in, and the other entities they co-occur with.

The Text Authority Signal

AI models do not evaluate sources by crawling their link graphs. They learn source authority from the text of the training data itself. A New York Times article is treated as authoritative not because the Times has 4 million referring domains, but because in the training data, Times articles are cited by other authoritative sources, quotes from them appear in academic papers, and the writing style is consistent with high-production-value journalism. The model learns what trustworthy looks like from the patterns — and it applies that learned signal when deciding how much weight to give a brand mention.

This is why a single mention in the MIT Technology Review carries more AI citation signal than fifty mentions in low-DA aggregation sites. The model knows, from patterns in the training data, that MIT Technology Review mentions are reliable. It does not know this from a link graph — it knows it from the textual ecology that publication exists within.

The Context Window Around a Mention

The text immediately surrounding a brand mention matters enormously for what the model learns from it. "Crowdstrike" mentioned in isolation contributes to the model's basic awareness that the entity exists. "Crowdstrike, the endpoint detection and response platform used by 60% of Fortune 500 companies, reported a 340% increase in detections of living-off-the-land attacks in Q3 2025" teaches the model what Crowdstrike is, what category it belongs to, what scale it operates at, and what kind of threats it addresses. The second mention is exponentially more valuable for AI citation purposes.

This is the contextual specificity principle: mentions that include descriptive context produce stronger model-entity associations than bare name mentions. The practical implication is that brands should optimize PR messaging and media briefing materials not just for getting named, but for getting named with specific, accurate, categorically relevant context.

Co-Citation Patterns

One of the most consistent findings in our data is the importance of co-citation — being mentioned in the same article or document as established category leaders. When a new or mid-tier brand is repeatedly mentioned alongside the incumbents in its category, the model builds an association that places the challenger in the same consideration set.

This mirrors how co-citation worked in academic literature before it was applied to web SEO. In citation analysis, a paper that is co-cited with foundational work in a field is assumed to be relevant to that field. AI models appear to apply an analogous heuristic: if a brand is consistently mentioned alongside the acknowledged leaders of a category, it is probably a meaningful player in that category.

The brands in our dataset that showed the fastest citation rate growth in 2025 had strong co-citation overlap with category leaders. Brands that were mentioned exclusively in isolation — in their own press releases, in generic brand features without competitive context — showed the weakest citation lift even when total mention volume was high.

The tactical implication: when pursuing earned media, prioritize placements in articles that also reference category leaders. A feature in a roundup that includes Salesforce, HubSpot, and your CRM is more valuable than a solo brand profile, from a co-citation standpoint. For PR teams, this means pitching "market landscape" and "category overview" stories aggressively — the story type most likely to generate co-citation alongside incumbents.

Brand Mention Velocity vs. Quantity

Cumulative mention count matters, but so does velocity. Our data shows a non-linear relationship between mention velocity and citation rate: brands that maintained a consistent cadence of 20+ authoritative mentions per month throughout 2025 showed citation rates 2.3x higher than brands that achieved the same total annual mention count in bursts, with gaps between campaigns.

This is consistent with how AI training data works in practice. Frequent model updates mean that recent content receives meaningful weight in the model's current representations. A brand that generates consistent monthly coverage is present in each successive training corpus. A brand that runs a major PR campaign twice a year has high coverage in those months and sparse coverage in the months between — and the model's representation of that brand reflects the average, which is mediocre.

The velocity finding has a direct implication for how brands structure their PR programs. The campaign model — periodic product launches, major announcements, award submissions — is optimized for burst coverage. The AEO model requires a content operations approach to PR: sustained, systematic outreach that generates a baseline of authoritative mentions every month regardless of whether there is a major news hook.

Some of the clearest examples come from B2B software companies that have shifted toward "executive thought leadership" programs — placing bylined articles, expert commentary, and analyst briefings on a monthly cadence rather than tying all coverage to product news cycles. Those programs generate lower peak coverage but much higher floor coverage, and the floor coverage is exactly what drives AI citation compounding.

Source Authority of Mentioning Sites

We ran a regression analysis on mention quality versus quantity across our dataset. The finding was clear: 20 mentions in publications with domain authority above 70 produced more AI citation lift than 200 mentions in publications with domain authority below 30.

The distribution of high-value mentioning sources varies significantly by vertical. For enterprise software, the highest-signal publications are the Wall Street Journal tech section, TechCrunch, The Information, and trade publications like CRN, SDxCentral, and DarkReading. For B2B fintech, it's American Banker, Barron's, and Bloomberg Finance. For healthcare technology, it's STAT News, Health Affairs, and Modern Healthcare. For professional services, it's Harvard Business Review bylines and appearances in industry conference proceedings that get written up by trade journalists.

Building a target publication list organized by DA threshold and topical alignment is one of the foundational steps of an AEO-informed PR strategy. The traditional PR approach of pitching the largest possible audience regardless of relevance produces volume without quality. The AEO-informed approach targets the 40 to 60 publications that have both the authority score and the topical alignment to produce high-signal mentions.

The Mention-to-Training Pathway

Understanding how a mention in a publication in May 2026 affects AI citation behavior in September 2026 requires understanding the training pipeline. The simplified version:

1. Publication and indexing. An article is published and indexed by major search crawlers within hours to days. AI lab crawlers — GPTBot, ClaudeBot, PerplexityBot — typically visit high-authority publications within days of publication, assuming those publications are not blocking AI crawlers in robots.txt.

2. Training data curation. Crawled content goes through quality filtering before entering training datasets. High-DA publications pass quality filters more reliably than low-DA sources. Content from established news organizations is weighted upward; content from thin, low-signal sources is weighted down or excluded.

3. Model training or fine-tuning. Major model updates incorporate new training data on cycles that have ranged from 3 to 9 months historically, though rapid fine-tuning and retrieval-augmented generation (RAG) systems can accelerate the lag. Real-time retrieval systems like Perplexity can surface very recent mentions almost immediately through live web fetching rather than training.

4. Citation behavior shift. Once the model has incorporated the new data, the brand's representation in the model changes — it appears more frequently in response to relevant queries.

The practical lag from publication to measurable AI citation lift was approximately 3–6 months for pure training data pathways in 2025. For Perplexity and other retrieval-augmented systems, the lag is much shorter — sometimes days. This distinction matters for strategy: if the goal is to influence Perplexity and Google AI Overviews citations in the near term, getting into high-authority sources that those systems fetch in real time is the fastest pathway. If the goal is to shift the underlying model's entity associations over the longer term, the sustained cadence of authoritative mentions builds the slower-compounding, more durable signal.

For a deeper view on how citation tracking works across AI systems, see the AEO citation tracking playbook.

How to Build Unlinked Mention Density

The playbook for building the kind of unlinked mention density that drives AI citation lift is different from traditional link-building, though it borrows some of the same media relationships. The key distinction: the goal is not to get a link back to your domain, but to get your brand name and a brief accurate description of what you do into high-authority text that AI crawlers will ingest.

1. Build a target publication list by domain authority and topical alignment. Start with the top 50 publications that cover your category and score their DA. Identify the 15–20 with both high DA and consistent coverage of your specific category. These are your primary targets for earned media placement.

2. Create a "citation-ready brand descriptor" and use it consistently. Work with your PR team on a 10–15 word description of your company that is accurate, specific, and categorically clear. Something like "Wiz, the cloud security posture management platform with $300M ARR" or "Replit, the collaborative browser-based IDE used by 30 million developers." Train all spokespeople and PR partners to use this descriptor. When journalists write about you, they will often use the language you give them. That language — not a hyperlink — is what gets indexed and learned by AI models.

3. Shift PR toward "landscape" and "comparison" story formats. The story types most likely to generate co-citation alongside category leaders are market roundups ("the five cloud security platforms CISOs are evaluating"), comparison pieces ("how does X compare to established players"), and category explainers that enumerate the major vendors. Pitch these angles proactively. Offer journalists structured comparison data that makes it easy for them to include your brand in a multi-vendor piece rather than a solo profile.

4. Invest in analyst relations as an AEO channel. Analyst firms — Gartner, Forrester, IDC, Redpoint — publish reports and blog content that is heavily indexed by AI training pipelines. Being named in a Gartner Magic Quadrant or Forrester Wave has always been a sales credibility signal; it is now also a high-signal AI citation input. Companies that have not historically invested in analyst relations because of the cost should reconsider the ROI when citation authority is included in the calculation.

5. Pursue speaker slots and industry proceedings. Conference proceedings, published speaker abstracts, and post-conference write-ups in trade publications generate a class of unlinked mentions that is particularly high-signal because the context is always topically aligned. A slot at RSA Conference, KubeCon, or Dreamforce that generates three trade press write-ups naming your company in the context of what you presented is worth far more AI citation signal than 50 directory listings.

6. Activate customer voices in external media. Customer case studies published by analysts, customer quotes in industry press, and customer testimonials cited by trade journalists are among the most powerful unlinked mention sources — because they are third-party validation, not self-promotion. AI models that encounter a customer quote naming your product in a neutral third-party publication register that as a high-quality signal. Building a systematic program for getting customers to speak publicly about your product — at conferences, in trade interviews, in analyst surveys — is one of the highest-leverage AEO investments available.

PR Strategy: Mentions vs. Links

The coexistence of link-based SEO goals and mention-based AEO goals creates a tension in how PR programs are measured and optimized. Many PR teams are still evaluated primarily on media hits and, in more sophisticated organizations, on backlink acquisition from those hits. The AEO mandate requires adding a third metric: mention quality score, defined as a weighted sum of DA-adjusted mentions with appropriate topical alignment scores.

The good news is that the two goals are largely complementary. High-authority publications that generate strong AI citation signal are exactly the publications that also generate high-DA backlinks. A Wall Street Journal mention drives both. The divergence happens at the margin — in decisions about where to spend discretionary PR budget.

The link-only framing says: prioritize placements that include a dofollow link to your domain. If the Wall Street Journal mentions you without a link, that is a nice-to-have but doesn't move the needle on domain authority.

The AEO-informed framing says: an unlinked Wall Street Journal mention is extraordinarily valuable. The AI citation signal from that mention is as strong as if the article had linked to you — possibly stronger, because the model weights authoritative unlinked mentions in editorial content more highly than corporate site backlinks. Chasing a link in that article at the expense of the mention itself is the wrong optimization.

In practice, this means PR teams should stop declining coverage opportunities that do not include links, stop requesting link insertions in a way that makes journalists less likely to write the story, and start measuring unlinked mentions in DA-qualified publications as a primary KPI alongside other standard metrics.

For teams still building their understanding of how brand mentions translate into AI search visibility signals, the ChatGPT citation engineering playbook provides the complementary view from the content side of the equation.

Measuring Mention-to-Citation Conversion

Closing the loop from PR activity to AI citation outcome requires a measurement stack that most teams are not yet running. The components:

Mention tracking. Set up comprehensive mention monitoring using Meltwater, Brandwatch, or a similar platform. Filter to sources with DA 40+. Log each mention with date, source, DA score, topical category, and whether a brand descriptor was included. Export monthly aggregates.

Citation probe queries. Run a battery of 50–100 category-relevant queries across ChatGPT, Perplexity, Claude, and Gemini on a weekly or biweekly basis. Document whether your brand is cited, in what context, and with what accuracy. Tools like Profound or Otterly automate this at scale.

Lag-adjusted correlation analysis. Plot monthly mention volume (DA-weighted) against AI citation rate with a 3-month lag. In our dataset, this lag-adjusted correlation was r=0.71 in the second half of 2025 — one of the strongest predictive relationships we found. If you are generating consistent authoritative mentions, your citation rate should be rising 3 months later. If it is not, the likely causes are either a robots.txt configuration blocking AI crawlers from your key sources, a mismatch between your brand descriptors in press and how you describe yourself in your own content, or model training lag that requires waiting for the next major update.

Share of model tracking. The broadest metric is share of model: in all AI responses about your category, what percentage name your brand? This is the output measure that all the input investments ultimately drive. Comparing your share of model at the start and end of each quarter, against the PR activity you ran in the prior quarter, creates the feedback loop that lets you optimize the program over time.

For the full measurement framework across all AEO inputs, see share of model: AI search measurement without vanity metrics.

The Brands Getting This Right in 2026

The clearest evidence that the mention-based AEO playbook works comes from looking at brands that are over-indexed for AI citations relative to their domain authority.

Wiz has a Domain Rating in the mid-70s — respectable but not exceptional for a cybersecurity vendor. Its AI citation rate across security-related queries is disproportionately high. The explanation is a sustained, methodical earned media program that has placed Wiz executives and customer quotes in the Wall Street Journal, Bloomberg, TechCrunch, Dark Reading, and SC Magazine on a near-monthly cadence since 2023. Each of those placements includes the phrase "cloud security" and a brief descriptor. The model has a clear, reinforced representation of what Wiz is.

Notion has achieved extraordinary AI citation rates not primarily through backlinks — its link profile is strong but not category-leading — but through a combination of user community content on Reddit, YouTube, and Twitter that generates millions of unlinked mentions, plus a sustained presence in productivity and knowledge management coverage in major tech publications. The breadth of authoritative unlinked mentions is wider than any link-building campaign could replicate.

Perplexity itself provides an instructive case study. The company's AI citation rate in queries about "AI search tools" or "alternatives to Google Search" grew from near-zero to among the highest in the category in under 18 months — driven almost entirely by mention density in technology journalism as reporters covered the AI search story. Perplexity benefited from the meta-irony of AI systems citing a company that was itself disrupting the sources those systems learned to trust.

The common thread is not a specific tactic. It is a systematic approach to generating consistent, authoritative, contextually specific mentions in publications that AI training pipelines treat as high-signal.

The Action Playbook

For operators who want to make the shift from link-focused to mention-aware authority building, here is the prioritized sequence:

1. Audit your current mention footprint. Pull the last 12 months of DA-qualified unlinked mentions from your media monitoring tool. Categorize by source DA, topical alignment, and whether the brand descriptor was included. This baseline tells you whether your current PR program is generating AEO-relevant signals.

2. Write and socialize a citation-ready brand descriptor. One sentence, 12–18 words, accurate, specific, categorically clear. Include the category name, a signal of scale, and your differentiated position. Train every spokesperson, PR partner, and agency on it. Use it in all media briefings as the preferred description of the company.

3. Rebuild your target publication list. Score your current media targets on both DA and topical alignment. Drop targets below DA 40 unless they are exceptionally high topical alignment. Add analyst firm publications and specialist trade press that you are not currently covering. The goal is a list of 40–60 high-quality targets, not 200 generic ones.

4. Shift your PR KPIs. Add DA-weighted unlinked mention count to your primary monthly PR metrics. Report it alongside traditional media hits. Build a monthly dashboard that plots mention density against the citation probe results from two to three months prior.

5. Build out co-citation positioning. Identify the two or three category leaders your brand is most frequently evaluated against. Actively pitch story angles that put you in the same conversation as those leaders — market comparisons, category analysis pieces, multi-vendor roundups. Co-citation with established names is one of the fastest ways to move a model's representation of where you sit in the competitive landscape.

6. Activate analyst and conference channels. If you are not in any analyst firm research, prioritize the ones that cover your category and have publication DA above 60. If you are not speaking at category-defining industry events, build a conference PR program focused on generating post-event trade press write-ups.

7. Launch a customer voice program. Design a systematic program for generating third-party quotes, case study citations, and customer testimonials in external media. Customer voices in neutral third-party publications are the highest-signal unlinked mention type in the dataset.

8. Run citation probes quarterly. Twice a year, run a full battery of category queries across all major AI assistants. Measure your citation rate. Correlate the movement against the PR activity from 3–4 months prior. Use the correlation to allocate next period's PR budget toward the highest-signal publication types.

What This Does Not Replace

The mention-over-links framing is a correction to an imbalance, not an argument for abandoning link-building. Links still matter for three things that are important in 2026.

First, links remain load-bearing for Google organic search, which still drives a significant share of discovery traffic alongside AI search. Brands that defund link acquisition entirely will see Google organic traffic decline, and that traffic still converts. The right answer is portfolio rebalancing, not replacement.

Second, links in high-authority editorial content still generate the same valuable unlinked mentions — they just also carry link equity. A Wall Street Journal article with a link to your domain is not worse than one without a link. It is better on both the link and the mention dimension. The tactical change is to stop requiring the link as a condition of accepting or pursuing coverage.

Third, link-rich content earns more secondary citations. High-DA pages that link to your site often also mention your brand in text that AI crawlers ingest. The link is the mechanism by which your brand earns secondary text mentions from other authoritative pages — the "being linked to by people who write about your category" pathway that has always been one of the most durable forms of authority building.

The rebalanced view is: mentions are the primary AEO signal, and links are valuable primarily as mechanisms that generate more mentions and that maintain Google organic performance. Programs that acquire links without generating meaningful text coverage of your brand are doing the lower-value half of the job.

Takeaway: The 12-month data is unambiguous — unlinked brand mention density in authoritative, topically aligned publications has become the strongest predictor of AI search citation rate, surpassing domain authority metrics that have dominated marketing authority thinking for a generation. The brands capturing this shift are running systematic, high-cadence earned media programs focused on mention quality over link acquisition, and they are seeing their AI citation rates compound at a pace that link-building alone cannot match. For operators, the mandate is clear: audit your current mention footprint, rebuild your PR KPIs around DA-weighted mention density, and shift discretionary media budget toward the high-authority, topically aligned placements where unlinked mentions produce outsized AI citation return.