For two decades, the currency of search was the link. Get more links from better sites and your pages rose. A whole industry grew around earning, buying, trading and measuring them. So when research started suggesting that, for AI-generated answers, something else might matter more, it was bound to cause an argument.
By the iSuggest.ai Team ยท Updated for 2026
That "something else" is the brand mention: your name appearing in other people's content, with or without a link. This post explains what the research reports, why it makes sense, where it should not be over-read, and how to earn mentions without gimmicks. We have kept the framing of every figure intact, and we do not claim that any tactic here guarantees a citation. For the wider background, see what GEO is.
What the research reports
One analysis of generative engine optimisation reports that brand mentions correlate with AI search visibility about three times more strongly than backlinks do, with correlation coefficients of 0.664 for mentions and 0.218 for backlinks (Media Officers). Put plainly, in that dataset, brands that were mentioned widely tended to be the brands that surfaced in AI answers, more reliably than brands with many links.
A second data point comes from a Meltwater study of more than 8 million citations across eight language models, which reports that earned and news media accounted for 37.6% of citations while press releases accounted for 0.2% (Meltwater). Again, that is one study with its own method and, as a media-monitoring vendor, its own interest in the importance of media. It is nevertheless consistent with the first finding: what other people publish about you appears to count for more than what you publish about yourself.
Why mentions might matter more to a language model
It helps to think about how these systems work, in broad terms. A language model learns patterns from large amounts of text. It learns which names appear alongside which topics, which descriptions repeat, and which brands are discussed in which contexts. Systems that retrieve live pages weigh sources that look credible and relevant to the question.
Three differences from classic search help explain the shift.
- Text, not only links. A link is a structural signal. A mention is language. A model reads sentences such as "for small clinics, many people use X" and absorbs the association between X, clinics and the use case, with or without a hyperlink.
- Corroboration. A claim about your company repeated by several independent sources is more believable than the same claim on your own site alone. Models and the systems around them are built to prefer information that agrees across sources.
- Entity recognition. Systems need to recognise "your brand" as a single thing. Consistent mentions across the web help build that picture, which is the subject of our post on entity clarity.
This is explanation, not proof. Nobody outside the model developers can state the exact weighting, and it differs between systems and changes over time. Treat it as a useful working theory consistent with the data.
What the research does not say
Correlation figures are easy to misuse, so here are the limits.
- It does not prove causation. Large, well-known brands tend to have many mentions and many links and be surfaced often. Mentions may be a symptom of being established, not the cause of visibility.
- It does not say backlinks are worthless. A coefficient of 0.218 is still positive, and links remain central to conventional search ranking. Many mentions arrive with links attached anyway.
- It does not describe your niche. A study across many brands may not hold for a small local business or a specialised B2B tool.
- It does not say all mentions are equal. A passing mention in a list is not worth what a detailed, positive, specific description is worth.
The sensible conclusion is modest: put more effort into being discussed credibly than you probably have, and do not drop the SEO fundamentals to do it. We explore how the two disciplines fit together in SEO versus GEO versus AEO.
What counts as a useful mention
Not every mention helps. A good one has most of these qualities.
- Independent. Written by someone who is not you and not paid by you in a way that is hidden. Disclosed sponsorships exist, but they are weaker than organic coverage.
- Specific. It says what you do, for whom, and ideally why it is good or different. "A tool for scheduling clinic appointments" is more useful than "a company".
- Consistent. It uses your name as you use it. Variants split the signal.
- In a relevant context. A mention in an article about your field, in a comparison, a how-to or a review, rather than in an unrelated list.
- On a crawlable page. If the page is behind a login or blocks AI crawlers, a model may never see it.
Where mentions come from
Mentions come from many places, and a healthy profile has several. Think in terms of types, and pick the two or three that suit your business.
Editorial and news coverage
Journalists, bloggers and newsletter writers describing you in their own words. This is the hardest to earn and, per the Meltwater data, apparently the most heavily cited. It comes from having something worth covering: original data, a clear opinion, a notable launch or a useful expert.
Reviews and comparison sites
Independent review platforms and "best of" lists describe what you do and how you compare. They are especially relevant for buying questions, which is where assistants are often asked to recommend options.
Community discussion
Forums, Q&A sites and professional communities where people recommend tools and services. A genuine answer from a user who mentions you carries weight; a promotional post by you does not, and can backfire.
Partners, customers and suppliers
Case pages, integration listings, supplier directories and customer announcements that describe the relationship. These are often easy to get and well worth asking for.
Directories and reference sources
Industry directories, professional associations and reference listings. Consistent details here reinforce your identity.
Your own content, used as a source
Your site is still the origin for facts only you can state: pricing, specifications, policies, original research. Clear, quotable pages give others something accurate to cite. See how to write content AI models want to quote.
How to earn mentions honestly
There is no shortcut, but there are practical habits that produce results over time.
- Create something citable. Original data from your own operations, a clear explanation of something complicated, a checklist people actually use. Others cite what is useful.
- Be easy to quote. State your key facts in plain sentences with a name and date, so that a writer can lift them accurately.
- Offer expertise. Answer journalists' and newsletter writers' questions about your field. Keep it genuinely helpful, and do not turn it into a sales pitch.
- Help customers talk about you. Make it easy to leave a review or describe how they use you, and ask at the moment they are most satisfied.
- Join the conversation. Contribute real answers in communities where your customers ask questions, and disclose your connection.
- Build relationships, not lists. A handful of people who understand what you do will mention you again and again. Mass outreach rarely does that.
- Keep your own description stable. One name, one short description and one category across every profile.
The press release deserves a note. The Meltwater figure for press releases, 0.2% of citations, suggests that simply distributing an announcement does little for citations by itself. A release can still be useful as a source document for writers who do cover you, but do not expect it to do the job of earning coverage. We unpack this in earned media versus press releases.
What to do about backlinks
Do not abandon links. Keep doing the things that earn them naturally, and keep an eye on quality over quantity. A link from a relevant, trusted page is a mention with a pointer attached, and it helps both classic search and AI. What you should deprioritise is link-building that produces no real coverage: bulk directory submissions, swapped links, and paid placements that exist only for the link. They add little and carry risk.
A useful reframing is to stop asking "how many links do we have?" and ask "who is talking about us, and what do they say?". Links will follow from that where they should, and the rest still builds the association models are learning.
A worked example
Imagine a small company that sells scheduling software to independent clinics. This is an illustration, not a case study. It has a modest link profile from a few directory listings and almost no independent descriptions of what it does. It rarely appears in assistant answers to "what scheduling tools do small clinics use?".
Over a quarter it takes five steps. It publishes a short original piece of research, using anonymised and aggregated data from its own product, on why clinics lose appointments. It offers the findings to two newsletters aimed at clinic owners. It asks pilot customers whether they would describe their experience in a short written review. It contributes useful, non-promotional answers in a clinic-owner community. And it makes sure its name and description are identical across every profile.
Some of this lands and some does not, which is normal. What it ends the quarter with is a handful of independent, specific descriptions of the company in relevant places, and a clearer entity. It tracks whether it is named in assistant answers over the following months, and treats the result as evidence to inform the next quarter, not as a guarantee.
How to track whether it is working
- Mention log. Keep a simple record of new independent mentions: date, source, what it says and whether it links.
- Assistant checks. Monthly, ask several assistants the questions your customers ask and record whether you are named. Answers vary between runs, so look at patterns over weeks.
- Referral and branded search. Watch referral sessions from assistants and branded queries in Search Console. Our guide to measuring AI search traffic in Google Analytics shows how.
Mistakes to avoid
Buying coverage or reviews
Paid reviews and fake testimonials violate platform rules, can be detected, and destroy the trust you are trying to build.
Spamming communities
Promotional posts get removed and associate your name with spam. Contribute answers first.
Inconsistent naming
If you appear as three slightly different names, the signal is split three ways. Pick one.
Neglecting the basics
A glowing mention does no good if your site blocks crawlers or hides its text behind scripts. Access and clarity come first; see how to check if GPTBot can crawl your website.
Expecting speed
Mentions accumulate slowly. Plan in quarters, not weeks.
Make your own site worth mentioning
Mentions point at something, and the thing they point at should be in good shape. The free iSuggest.ai audit checks the machine-readable side of your site: crawler access, rendering, structured data, headings and identity signals. You can read what an audit measures for the full list. Audits are free and not metered, so you can run one on every page you hope to be cited for.
Once your pages are ready, you can publish a factual snapshot of your business to the iSuggest.ai AI directory, which has pages for ChatGPT, Claude and Gemini. It adds one more consistent, crawlable description of you. Publishing spends credits, which are non-refundable, and it improves findability without guaranteeing a citation; see how credits work. See how it works, run an audit on your homepage, then sign in at your account page when you are ready.