TL;DR
- Unlinked mentions pass no Google ranking credit but drive AI visibility.
- Mueller confirmed plain mentions stay out of the link graph.
- 25.1% of AI brand appearances are mentions with no citation.
- Only 13.2% of appearances both name the brand and cite it.
- Convert mentions with search value and correct the wrong ones.
An unlinked brand mention passes no ranking credit in Google Search, and Google has said so on the record more than once. It still earns you something links do not: AI engines name brands far more often than they link to them. In a June 2026 study of 3,981 domain appearances, 25.1% were brand mentions with no citation attached. A July 2026 study of 12,000 AI responses found only 23.1% of brand mentions came with a citation. The name in the answer is the visibility, with no link involved.
What counts as an unlinked brand mention
An unlinked brand mention is any reference to your company, product, domain, or people in someone else’s content with no hyperlink pointing back to you. A roundup that names your tool without linking it. A Reddit thread where a user types your domain as plain text. A news article that quotes your founder. All three are mentions. None of them is a link.
The term matters because two very different questions hide inside it. The first is whether Google treats the mention as a ranking signal. The second is whether the mention makes your brand visible to the systems people actually ask now. Those questions have different answers, which is why brand entity work and link building have stopped being the same job.
What Google has actually said about unlinked mentions
John Mueller answered the question directly in a Google SEO office-hours session on 17 December 2021. Asked whether brand mentions without a link help rankings, he said: “I don’t think we use those at all for things like PageRank or understanding the link graph of a website. And just a plain mention is sometimes kind of tricky to figure out anyway.” He then named the indirect route that does work: a mention makes people aware of your brand, they search for it, and they land on you.
He gave the same answer three years earlier. In a December 2018 webmaster hangout, asked whether a mention without a link counts as a positive signal, Mueller said Google would “pick that up as a mention on another web site, but it’s not a link. It doesn’t pass any page rank.”
Later in the 2021 session, someone asked how Google would handle spam pages calling a brand a scam, given that you can disavow a link but not a mention. Mueller agreed that is exactly the problem. Reading the intent behind a mention is hard: “Is it a sarcastic positive mention or a sarcastic negative mention? How can you even tell?” His conclusion was that mentions are “too confusing to use as a clear signal.” That is an engineering constraint on the signal, and it says nothing about whether being talked about is worth having.
Gary Illyes described the other side of it at BrightonSEO in 2017, in an interview Search Engine Watch published in full. Describing what a healthy site looks like, he said high quality content should be “highly cited on the internet,” and added that he was “not talking about just links, but also mentions on social networks and people talking about your branding.” Google’s public position holds both ideas at once: mentions do not enter the link graph, and widespread discussion is what strong brands have.
The “implied links” patent says something narrower than you were told
Most articles claiming unlinked mentions carry ranking weight trace back to one Google patent, US8682892B1, “Ranking search results,” granted on 25 March 2014. The patent does define the term. It separates an express link, “a hyperlink, is a link that is included in a source resource that a user can follow,” from an implied link, “a reference to a target resource, e.g., a citation to the target resource, which is included in a source resource but is not an express link.”
Two things get lost in the retelling. The patent’s headline mechanism is not the mention at all: it counts “reference queries,” which the document defines as search queries “classified as referring to a resource in the group.” That is branded search, not text on a page. And a granted patent records what a company filed, not what its ranking systems run. Google files thousands. Mueller’s statements describe the shipped system, and they are eleven years newer than the grant. Treat the patent as vocabulary. It carries the same weight as any other link building claim with no measured result behind it.
AI engines changed the value of a mention
Retrieval-augmented engines write answers in prose and attach sources separately. Those two layers do not line up, and the gap is large enough to change how you plan mention work.
Semrush and Kevin Indig logged 3,981 domain appearances across 115 prompts, 14 countries, and four AI engines in June 2026. Almost 62% (61.7%) were what they call ghost citations: the page was used as a source link, and the brand name never appeared in the answer text. Only 13.2% were cited and named together. The remaining 25.1% were brand mentions with no citation, where the engine named the brand and linked to nothing.
Engine behaviour splits hard. Gemini names brands in 83.7% of appearances and generates a citation link only 21.4% of the time. ChatGPT does the reverse, citing brands 87% of the time and naming them in 20.7% of answers. A brand that measures only citations is invisible to itself on Gemini. This is the same per-engine divergence we found in our own citation work, where AI Mode and Gemini shared just 3.5% of cited domains across 42,971 citations.
BuzzStream ran 12,000 AI responses across Google AI Mode, AI Overviews, Gemini, and GPT-5.4-mini in July 2026, covering roughly 200 brands. Only 23.1% of brand mentions were backed by a citation in the same response, while 69.9% of citations also named the brand. On open list and category prompts, the kind buyers use before they know any vendor, the number collapses to 7.2%. If your category page is where the buyer starts, the mention is nearly the whole prize.
Our own data points the same direction from the retrieval side. Across four identical ChatGPT runs in our citation variance study, the share of pulled-up sources that got credited landed at 23.1%, 24.9%, 27.9% and 22.0%. Roughly three quarters of what the model reads never reaches the visible source list. A mention of your brand inside one of those uncredited pages still feeds the answer. In our May 2026 study of 153,425 citations, 76.95% of cited URLs were not in the organic top 10 for their query, so the pages carrying your name are often not the pages you rank with.
Ahrefs studied 75,000 brands in December 2025 and found branded web mentions correlating with AI visibility at 0.66 to 0.71 Spearman, well above domain rating and backlink counts, while number of site pages sat near zero at about 0.194. Ahrefs states the disclaimer themselves: correlation is not causation, and these may be markers of brand strength rather than causes of it. Read it as a description of which brands get named. It is not a lever with a guaranteed return. Our entity resolution work suggests the causal path runs through the engine recognising the brand as a distinct entity in the first place.
Google’s own documentation closes off the on-page shortcuts. Its AI features page states there “are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” No schema type, no text file, no markup. That leaves the off-site layer, where mentions live, carrying most of the load.
Linked and unlinked mentions do different jobs
| Outcome | Link | Unlinked mention |
|---|---|---|
| PageRank and the Google link graph | Yes | No, per Mueller in 2018 and 2021 |
| Referral clicks | Yes | Only if someone types your name into search |
| Named in an AI answer | Not required | This is the mechanism, 25.1% of appearances are mention only |
| Feeds entity co-occurrence with your category | Yes | Yes, the surrounding text does the work |
| Correlates with AI visibility across 75,000 brands | Backlinks correlate weakly | Branded web mentions correlate 0.66 to 0.71 |
| Can be disavowed if hostile | Yes | No |
| Visible in Search Console | As a linking page | Only indirectly, through branded query impressions |
How to find your unlinked brand mentions
Four methods cover almost everything, and they stack rather than compete.
1. A content index with an unlinked filter
Ahrefs Content Explorer searches an index of 21.4 billion pages and ships a stated use case for exactly this: find “websites that mention your brand but aren’t linking to you.” Search your brand name in the page text, exclude your own domain and the social platforms, then use the unlinked domain highlight to separate sites that have never linked to you from sites that already have. Export the survivors.
2. Ongoing monitoring rather than one-off pulls
A quarterly export misses the window when a mention is fresh and the writer still remembers writing it. Semrush’s Brand Monitoring app sends an email every time a new mention appears, flags negative sentiment, and alerts on unusual spikes in mention volume. Set queries for the brand, the product names, and the two or three people who get quoted. The same alerts double as an early warning system for reputation problems that reach AI answers before they reach you.
3. Verify the mentions are genuinely unlinked
Tool indexes lag. Confirm before you email anyone. Run your exported URL list through Screaming Frog in list mode with custom search, which supports up to 100 filters set to “contains” or “does not contain,” using text or regex against the HTML. A does-not-contain regex on your domain inside an anchor tag isolates the pages that mention you in text and link nowhere. It also catches the awkward case where the link exists but is nofollowed or points at a redirect chain.
4. Watch the answer layer separately
Web mentions and AI answer mentions are different inventories, and no crawler-based tool sees the second one. You need prompt-level tracking that records whether the engine named you, cited you, or both, per engine. Aggregated visibility scores hide the Gemini and ChatGPT split described above. Our manual AI visibility audit covers the low-cost version if you are not buying a platform yet, and the metrics set worth tracking is smaller than most dashboards suggest.
One caution on scope. A two-word generic brand name will over-collect badly, and no filter fixes that cheaply. Track the product names and the founder’s name alongside the company name, since those are usually unambiguous, and expect a manual review pass on anything that looks like a common noun.
When to convert a mention into a link, and when to leave it
Outreach costs goodwill, and you spend it once per contact. Triage before you write.
Convert when the page has search value you want. It ranks for a query you care about, or it shows up as a source in AI answers for your category. A link there earns referral clicks and ranking credit that the mention alone does not deliver. The pages that get cited in AI answers are worth extra attention, because a small set of domains carries a large share of citations.
Convert when the mention is incomplete rather than wrong. A listicle that names you but describes only one feature is an easy ask, because the writer benefits from the correction too. Send the missing detail first and the link request second.
Leave it when the page is weak or hostile. A link from a scraped or spun page adds nothing, and asking a critic for a link hands them a reason to revisit the criticism. Mueller’s point about sarcastic mentions applies to your outreach list as much as to Google’s systems: read the sentence before you send the email.
Leave it when the outlet has a no-link policy. Many news desks and wire services do not add outbound links after publication. The mention still does its job in the AI answer layer, where the brand name in the text is what gets read.
Fix the facts instead of chasing the link when the mention is wrong. A page that puts you in the wrong category, quotes a discontinued price, or misnames your product is feeding that error into every model that reads it. Correcting the sentence beats adding an href, because AI answers repeat text far more readily than they follow links. That is the same reasoning behind defending against brand hallucinations at the source page.
Never buy mentions. Google’s spam policies now define spam to include “attempting to manipulate generative AI responses in Google Search,” alongside ranking manipulation. Paid mention placements sit in the same category as paid links, with the added problem that you cannot disavow what you cannot see.
Measuring mentions without fooling yourself
Four numbers are enough, and each has a trap.
- Unique domains mentioning you, per month. Count domains rather than mentions, because one prolific site distorts the total. Ahrefs found brands were not at a large disadvantage when the mentions came from low-view YouTube videos, as long as they were mentioned widely.
- Share of mentions against your named competitors. An absolute count has no meaning on its own. The comparison set should be the two or three brands your buyers evaluate you against, which is the same framing as AI share of voice.
- Mention rate and citation rate, tracked separately per engine. The Semrush data shows the two moving independently, and their engines disagreed about whether to name a brand in 22% of the prompt and domain pairs they tested.
- Branded query impressions in Search Console. This is the closest measurable version of the effect Mueller described. Mentions create awareness, awareness creates branded search. Track impressions on brand queries, not just clicks.
The trap that catches most teams is single-run measurement. Our variance study found the same ChatGPT question returning a full source list in one run and no sources at all in another on the same day. Any single reading of your AI mention rate is close to a coin flip. Sample repeatedly, average, and treat week-over-week movement smaller than your own measured swing as noise.
What to do with this
Stop treating unlinked mentions as failed links. In Google Search they pass nothing, and no amount of outreach changes the mechanism. In AI answers they are the unit of visibility, and the engines that name brands most generously are the ones least likely to link. Build the mention inventory, convert the small slice where a link earns its own return, correct the mentions that are factually wrong, and measure the answer layer per engine rather than in aggregate. The entity groundwork and a clean reference presence make every one of those mentions resolve to the right brand, which is the part that has to be right before any of the rest counts.