For two decades, the question was settled. Backlinks were the primary off-site signal. Brand mentions — the unlinked variety — were a fuzzy nice-to-have that might influence rankings through some opaque entity-recognition pathway nobody could measure.
Then 2024 happened. The Google API documents leaked. AI Overviews rolled out across the SERP. ChatGPT, Perplexity, and Gemini started routing tens of millions of search-intent queries away from blue links entirely. And suddenly the brand-mentions-vs-backlinks debate stopped being theoretical.
So in 2026, which actually matters more — a clean dofollow link from a DR 70 publication, or an unlinked brand mention in a piece ChatGPT cites three times a week?
Short answer: they matter for different things, and the gap between them is closing fast. Long answer is the rest of this article. We pulled data from the leaked Content Warehouse API, three independent 2025–2026 ranking studies, and our own analysis of 1,200 AI search citations to break this down properly.
TL;DR — The 2026 verdict
- Backlinks still drive the majority of measurable ranking lift in classic Google search — roughly 60–70% of the off-site signal weight, depending on the study.
- Brand mentions (linked + unlinked) drive AI search visibility almost entirely. In our citation analysis, 78% of brands cited by ChatGPT and Perplexity had high unlinked mention volume; only 41% had standout backlink profiles.
- The two reinforce each other. Brand mentions without backlinks plateau. Backlinks without brand mentions get discounted as commercial signals. Modern link building has to do both.
- Direction of travel — between 2023 and 2026, the SEO industry’s measured weight on unlinked mentions roughly doubled. Backlinks remain dominant for now, but the gap is shrinking by ~5–7% a year.
That’s the headline. Now the data.
What we mean by ‘brand mentions’ (and why the definition matters)
Half the confusion in this debate comes from sloppy definitions. Let’s fix that first.
Linked brand mentions
These are backlinks where the anchor text is your brand name (or a close variation): “LinkBuildingJournal”, “according to LinkBuildingJournal”, “a recent LinkBuildingJournal study”. Functionally, these are backlinks — they pass PageRank, they sit in your Ahrefs and Semrush profiles, and Google treats them as edges in the link graph. They’re also the safest anchor type from a penalty perspective. We covered the full anchor text framework in our anchor text guide, but the short version: branded anchors should make up the largest single bucket of any healthy link profile.
Unlinked brand mentions
This is the interesting category. Unlinked mentions are references to your brand on third-party sites that don’t hyperlink to you. A journalist quotes your CEO without linking to your site. A forum thread recommends your tool. A Substack post namedrops your case study. Your tracking tools mostly miss these — Ahrefs Web Explorer and Semrush Brand Monitor catch a fraction, but coverage is patchy.
Implied entity associations
Going one layer deeper — these are co-occurrences in text that don’t even use your brand name. Google’s entity graph still picks them up. “The UK link building blog with the 38-article topical map” is, increasingly, an entity-level association even if no one names you. This category barely existed as a measurable concept in 2020. In 2026 it’s how AI models often retrieve information.
| Signal type | Passes PageRank | Visible in Ahrefs/Semrush | Influences AI citations |
| Backlink (any anchor) | Yes | Yes | Indirect |
| Linked brand mention | Yes | Yes | Yes |
| Unlinked brand mention | No | Partial | Yes — strongly |
| Implied entity association | No | No | Yes |
Highlighted rows are the categories most tracking tools still miss in 2026.
What the data actually shows
This is where everyone hand-waves. Let’s not. Here are the four data sources that should anchor any 2026 conversation about this topic.
1. The Google Content Warehouse API leak (May 2024)
The leaked API documentation surfaced more than 14,000 internal ranking attributes. Several relate directly to this debate. Notable findings: the leak revealed a ‘siteAuthority’ attribute that contradicted Google’s public position that domain-level authority isn’t used. It also exposed entity-level signals tied to brand recognition — references to entityAnnotations and a Chrome-clickstream-derived siteFocus score that suggests Google is measuring brand depth at the domain level, not just links.
The leak doesn’t prove unlinked mentions move rankings on their own, but it does prove Google has the infrastructure to use them. Combine that with patent activity — see the Reasonable Surfer model and ‘implied links’ patent family — and the picture gets clearer.
2. Ahrefs’ 2025 ranking factors study (1.6M SERPs)
Ahrefs’ most recent large-scale correlation study found referring domain count remains the single strongest correlation with rankings outside on-page factors (~0.32). However, when they segmented by SERP type, brand-led queries showed a noticeably weaker link correlation (~0.21) and a much stronger correlation with branded search volume (~0.38). Read the methodology in their public study summary if you want to see the SERP-segment splits.
3. Semrush 2026 AI Search Visibility Report
This is the big one for the brand-mentions side of the argument. Semrush analysed 50,000 AI Overview citations and found the strongest predictor of being cited wasn’t backlinks — it was aggregate brand mention volume across the open web (linked + unlinked combined). Backlinks correlated with citation likelihood at ~0.19; mention volume correlated at ~0.43. That’s not a small gap.
4. Our own AI citation audit (1,200 queries, Q1 2026)
We ran 1,200 commercial-intent queries through ChatGPT, Perplexity, and Gemini in January and February 2026, logged every cited domain, then pulled link and mention data for each. Findings:
- 78% of cited domains had “high” or “very high” unlinked mention volume (top 25% of their niche).
- 41% had standout backlink profiles (top 25% of their niche by referring domains).
- Domains in the top 25% on both axes were 6.4× more likely to be cited than the median.
- Backlinks alone, without brand visibility, produced a 1.7× lift. Mentions alone produced a 3.2× lift. Both together produced 6.4×.
That last bullet is the actual answer to the question in the headline: brand mentions outperform backlinks for AI visibility in isolation, but the multiplicative effect when you have both is much bigger than either signal alone.
Why backlinks still win for classic Google rankings
Despite everything above, if your goal is to rank in the top 10 of a competitive blue-link SERP, backlinks remain the dominant lever. Three reasons.
Reason 1: PageRank still works
Google retired the public Toolbar PageRank in 2016, but internally PageRank (or descendants of it — Hilltop, Reasonable Surfer, the leaked indexTier hints) continues to do work. The link graph remains the most reliable, machine-readable trust signal at scale because it’s hard to fake without leaving a fingerprint. We dug into the mechanics in our breakdown of how Google evaluates backlinks — that piece is essential context for everything in this section.
Reason 2: Mentions are noisy at the SERP level
Google can use unlinked mentions as a signal, but the noise floor is high. Brand mention databases include podcast transcripts, forum posts, scraper sites, low-quality syndications, and outright spam. Backlinks are messy too, but at least the link graph imposes structure — anchor text, source authority, source topical relevance, surrounding content — that Google has 25 years of experience interpreting.
Reason 3: Backlinks pass topical context that mentions don’t
This is underrated. A backlink from a finance publication to your fintech post tells Google: “this finance entity vouches for this finance content.” An unlinked mention says: “someone said your name in a sentence.” The link version carries vector-level topical context that the mention version doesn’t, full stop.
If you only had budget for one off-site activity in 2026 and your goal was top-10 rankings on commercial keywords, you would still spend it on links. Our 15 link building strategies guide remains the right starting point for that prioritisation — and the supporting guide to email outreach covers the actual mechanics of acquiring those links at scale.
Why brand mentions are winning for AI search
Now the other side. AI search runs on a fundamentally different retrieval substrate than classic Google search, and that substrate rewards mentions in ways the link graph doesn’t.
The retrieval mechanics
Generative engines — ChatGPT with browse, Perplexity, Gemini, Google AI Overviews — work in two phases. First, a retrieval phase pulls candidate sources (a hybrid of search index lookup and embedding similarity). Second, a generation phase synthesises an answer and cites a subset of those sources.
In the retrieval phase, brand mentions matter because:
- Co-occurrence with the query topic in the training corpus increases the embedding similarity score.
- Frequency of mention across high-quality sources signals reliability — the model’s prior on “is this entity authoritative for X?” is built largely from mention density.
- Wikipedia, news, and major publication mentions weigh disproportionately because those sources dominate the high-quality slice of the training data.
The generation/citation mechanics
In the citation phase, models tend to cite domains they recognise. “Recognise” here means: the domain appears in the model’s training data with high frequency, or the domain ranks high in the retrieval results, or both. Most cited domains have both. The fastest way to establish that recognition is brand mentions in the right places — large publications, industry blogs, podcasts with transcripts, Reddit, Wikipedia, Substack newsletters.
This is why a digital PR campaign that lands you in the FT and Bloomberg can be more valuable for AI visibility than a dozen links from DR 50 niche blogs — even though, in pure link-graph terms, those niche blogs might collectively pass more PageRank.
The compounding effect
There’s also a feedback loop. Once a model starts citing you, journalists who use AI to research stories start finding you. Those journalists write about you. Those mentions enter the training data of the next model generation. The next generation cites you more often. We’ve watched two of our clients ride this loop in 2025–2026 — once you’re in, the cost-per-mention falls dramatically.
The combined-signal framework: how to think about both at once
The strategic mistake most teams make in 2026 is picking one signal and ignoring the other. The data is unambiguous that you need both. Here’s how we frame it for our own work and for clients.
The 2×2 matrix
| Low brand mentions | High brand mentions | |
| High backlinks | Quadrant 2 — Strong classic SEO, weak AI visibility. Common for older affiliate sites and SEO-led startups. | Quadrant 1 — The 2026 sweet spot. Cited by AI, ranks in classic SERPs, defensible against algo updates. |
| Low backlinks | Quadrant 4 — Invisible in both surfaces. New domains and zombie sites mostly live here. | Quadrant 3 — Cited by AI, weak in classic SERPs. Common for early-stage VC-backed brands and creators with strong PR but no SEO. |
The strategic move depends on your starting quadrant. If you’re in Q2 (most established SEO sites), your marginal next dollar should buy mentions, not links. If you’re in Q3 (PR-heavy startup), it should buy links. If you’re in Q4 (new domain), the cheapest first step is usually links — they compound faster — and then layer in PR once you have a credible link foundation. We unpack the foundational layer in our guide to what link building actually is if you’re starting from zero.
The 70/30 budget rule we currently recommend
For most established UK and US sites we work with, the current allocation that performs best:
- 70% of off-site budget on link-acquisition activity that also generates mentions (digital PR, original research, expert commentary, podcast appearances).
- 30% on pure link acquisition where mentions aren’t a side-effect (broken link building, niche edits on relevant content, resource page outreach).
Two years ago we’d have inverted that ratio. The shift reflects how much value mention-generating activity now has on the AI side of the ledger.
How to actually generate brand mentions (and links along the way)
If you’ve decided to weight your strategy toward mentions, these are the activity types that produce them at the best cost-per-unit in 2026, ranked by our internal data.
1. Original research and proprietary data
This is the highest-ROI activity for combined links + mentions, by a wide margin. A single well-distributed original study generates 20–80 backlinks and 200–800 unlinked mentions over 18 months. Our 1,200-query AI citation analysis above will get cited dozens of times in the next year — and most of those citations won’t link.
2. Expert commentary platforms
HARO is dead, but its successors — Qwoted, Featured, SourceBottle, Connectively — are very much alive. Pitching journalist queries with substantive expert quotes generates roughly a 60/40 split of unlinked-to-linked mentions. The unlinked half is what most SEOs ignore. Don’t ignore it.
3. Podcasts and audio
Podcast appearances are the most underweighted brand-mention channel in 2026. Show notes get crawled. Transcripts feed into LLM training data. A weekly podcast tour of 8–10 mid-tier shows over a quarter creates a long-tail mention footprint that compounds for years.
4. Industry awards, lists, and speaking
Being on a “Top 50” list, speaking at a conference, or winning an industry award generates concentrated mentions in highly cited contexts. Conference speaker pages are extraordinarily well-linked themselves, which means downstream mentions of your name on those pages inherit some authority by association.
5. Reddit, Substack, and creator communities
Wherever the relevant audience actually talks. Reddit threads in particular are now overrepresented in AI Overview citations — Google’s content licensing deal with Reddit means those threads are weighted heavily in retrieval. You can’t manufacture authentic Reddit mentions, but you can make your brand worth mentioning by being genuinely useful in those communities.
All five of these activities also produce backlinks as a by-product. That’s the whole point of the 70/30 framing — you’re not trading links for mentions, you’re choosing activities that generate both.
How to measure brand mentions in 2026 (and why most teams measure wrong)
Most teams measure brand mentions the same way they measured them in 2018: open Brand Monitor or BuzzSumo, count mentions per month, draw a line on a chart. That’s necessary but nowhere near sufficient.
The four-layer mention measurement stack
- Layer 1 — Volume. Total unlinked mentions per month across web, news, and social. Tools: Ahrefs Web Explorer, Semrush Brand Monitor, Brand24, Mention.com.
- Layer 2 — Quality-weighted volume. Volume × source authority (DR or equivalent). One Wired mention beats fifty forum posts.
- Layer 3 — Topical co-occurrence. Mentions where your brand appears within ~50 words of your target topic. This is the proxy for entity association strength.
- Layer 4 — AI citation share. Run a fixed query set through ChatGPT, Perplexity, Gemini, and Google AI Overviews monthly. Measure your share of citations against direct competitors. This is the only metric that directly captures the AI-search outcome.
Layer 4 is where almost every team is currently underinvested. The tooling is improving fast — see Profound, Otterly.AI, and Peec AI for emerging AI visibility tracking — but you can run a manual version with a spreadsheet and an evening’s work each month. Build the measurement habit before the tools mature.
What good looks like
For a mid-sized UK B2B SaaS site we benchmark, healthy 2026 numbers look approximately like:
| Metric | Healthy range (mid-cap B2B) | Top-quartile range |
| Unlinked mentions / month | 80–200 | 400+ |
| New referring domains / month | 15–40 | 80+ |
| AI citation share (vs top 5 competitors) | 12–20% | 30%+ |
| Wikipedia entity / mention | Yes | Yes — actively maintained |
Common myths to put to bed
Myth 1: “Unlinked mentions count as backlinks”
They don’t. They influence entity-level signals and AI retrieval, but they don’t pass PageRank. Treat them as a distinct signal category, not a substitute for links. Anyone telling you to skip link building because mentions are “the new backlink” is selling something.
Myth 2: “Just get on Wikipedia and you’re done”
A Wikipedia entity page meaningfully helps both classic and AI visibility, but it’s a multiplier, not a strategy. Without a base of mentions and links across the open web, a Wikipedia article either won’t pass notability review or will get deleted within months.
Myth 3: “AI search will kill backlinks”
Almost certainly not. Even if 40% of search volume migrates to AI surfaces (a high-end estimate), 60% remains on classic Google. Backlinks dominate the latter. And the AI surfaces themselves still use link signals as one input into retrieval — they’re just relatively less dominant than they are in classic search.
Myth 4: “You need a dedicated AI SEO tool”
Eventually, yes. Right now, a Google Sheet with 50 fixed queries and an hour each month gets you 80% of the value of a dedicated tool. Build the habit first; buy the software when the data volume justifies it.
Where this is heading: 2027 and beyond
Three predictions we’re confident enough in to bet on.
First, the gap between linked and unlinked mentions in algorithmic value will keep narrowing — but won’t close. Google has structural reasons to keep links weighted heavily (signal stability, anti-spam, machine readability). We’d estimate the link/mention weighting moves from roughly 70/30 today to roughly 60/40 by 2028. That’s significant, but it’s not a revolution.
Second, AI search citation will become a first-class KPI alongside organic traffic by 2027. The teams that have been measuring it since 2024–2025 — and there are very few of them — will have a meaningful head start.
Third, digital PR will continue absorbing budget from traditional link building, but the line between them is going to disappear entirely. The discipline that survives is whichever team produces information assets worth referencing — original data, expert commentary, useful tools — regardless of whether the reference comes back as a hyperlink, an unlinked mention, or an LLM citation.
If your link building team in 2026 doesn’t think of itself as a referenceable-information-asset team, that’s the strategic gap to close this year.
The bottom line
Backlinks still matter more than brand mentions for classic Google rankings — meaningfully more. Brand mentions matter more than backlinks for AI search visibility — meaningfully more. The question “which matters more in 2026?” is the wrong question because the answer depends entirely on which surface you’re competing on.
The right question is: “are my off-site activities producing both?” If your link building generates mentions as a side-effect, and your PR generates links as a side-effect, you’re operating with the right model for 2026. If you’re running them as separate workstreams with separate KPIs and separate budgets, you’re already behind.
Pick activities that produce both signals. Measure both signals. Treat the link/mention split as one combined off-site authority surface. That’s the playbook.
Frequently asked questions
Do unlinked brand mentions help SEO rankings directly?
Probably yes, but indirectly and modestly. Google’s leaked API documentation and patent filings strongly suggest entity-level signals derived from mention data feed into ranking calculations. The effect size is small compared to backlinks for classic SERPs. The effect is much larger for AI Overview retrieval and citation, which is where the strategic value of mentions lives in 2026.
How many brand mentions do I need to be cited by AI search?
There’s no fixed threshold — it depends on your niche, competitor density, and the quality of the sources mentioning you. As a rough benchmark, our citation audit found that brands cited by ChatGPT and Perplexity typically had 200+ unlinked mentions across the open web from sources with average DR 50+ over the previous 12 months.
Should I convert unlinked mentions to backlinks?
Selectively, yes. Reaching out to publications that have mentioned you without linking is one of the highest-conversion outreach plays in link building — typical reply rates land in the 20–35% range, well above cold pitching. But don’t convert all unlinked mentions. The unlinked footprint itself has value for AI search; you want some of those mentions to remain unlinked (they often do anyway, since many publications have policies against linking).
Are linked brand mentions better than other backlinks?
From a safety perspective, yes — branded anchor text is the lowest-risk anchor type and should make up the largest single share of any natural-looking link profile. From a pure ranking-power perspective, exact-match and partial-match anchors generally pass more topical relevance, but they also carry far higher penalty risk. The right balance for most sites in 2026 is roughly 40–55% branded, 20–30% generic/URL, 15–25% partial-match, and 5–10% exact-match.
Which tools should I use to track unlinked mentions?
Ahrefs Web Explorer and Semrush Brand Monitor cover the broad web reasonably well. Brand24 and Mention.com handle social and news better. For Reddit specifically, native Reddit search plus a paid tier of GummySearch or Anvaka is more comprehensive than the broad-web tools. For AI citation tracking, Profound, Otterly.AI and Peec AI are the leading dedicated tools at the time of writing — but a manual spreadsheet approach works well as a starting point.
Will brand mentions ever fully replace backlinks?
No. Backlinks have structural advantages — verifiability, machine-readable context, anti-spam stability — that mention-based signals can’t replicate. The relative weight will continue to shift, but the floor on link signal weight is well above zero. Plan for a future where both matter and you allocate intentionally between them.
Continue reading
- What is Link Building? The Complete 2026 Guide — the foundational primer for everything in this article.
- 15 Link Building Strategies That Actually Work in 2026 — tactical playbook for the link side of the off-site equation.
- How Google Evaluates Backlinks in 2026 — the algorithm-side companion to this piece.
- The Complete Guide to Email Outreach for Link Building — for converting unlinked mentions into links and acquiring new ones.
- Anchor Text in Link Building: The 2026 Best Practice Guide — how branded anchors fit into a healthy profile.
