In short: Paying an agency to do link building and optimise individual keywords delivers less and less. ChatGPT, Perplexity and Gemini cite the brands that the press names, not the ones with the most backlinks. Technical SEO is still a prerequisite, but the spend that moves the needle today is PR / earned media combined with a single intelligence tool that unifies analytics, Search Console, GEO and competitor visibility.
- 0.664 vs 0.218: unlinked brand mentions correlate with AI citations roughly 3x more than backlinks — Ahrefs analysis of 75,000 brands (2026)
- 84% of AI citations come from earned media, versus 0.3% from paid content (Muck Rack, 25M+ links, 2026)
- -25% organic traffic by the end of 2026, -50% by 2028 (Gartner)
What is the difference between SEO and GEO?
SEO (Search Engine Optimization) optimises a page to rank among Google's blue links and earn clicks. GEO (Generative Engine Optimization) optimises so that a brand gets cited inside the answer generated by an AI engine — Google AI Overviews, ChatGPT, Perplexity, Gemini, Claude. Two different games: in the first you compete for a position in a list, in the second to be the source the model chooses to name and recommend.
The part that catches pure-SEO shops off guard: ranking on Google no longer guarantees being cited by AI. Fewer than 10% of the sources cited by ChatGPT, Gemini and Copilot sit in Google's organic top 10 for the same query. The overlap between the top 10 and AI Overview citations collapsed from 75% in mid-2025 to 17-38% in early 2026 (WordStream, 2026). Translation: you can pay to rank first on a keyword and stay invisible where reputation is now decided.
Why does paying an agency just for link building deliver less and less?
For twenty years off-page SEO was, in practice, link building. That engine has stalled for two converging reasons.
First: language models don't read the link graph, they read text. ChatGPT and Claude are trained on raw text: they learn who a brand is and what it is known for from how often and in what context it is named, not from how many hyperlinks point to its site. That's why an unlinked mention in a review, an industry article or a discussion is worth more than an optimised backlink. The Ahrefs study of 75,000 brands quantifies the gap: unlinked mentions 0.664, backlinks 0.218 correlation with AI citations (Ahrefs / Soar, 2026).
Second: backlinks aren't dead, but they're worth a fraction of what they were. They remain a factor for traditional ranking, yet their relative weight has collapsed as traffic shifted inside AI answers. Gartner forecasts a 25% drop in organic search traffic by the end of 2026 and 50% by 2028 (Gartner, 2026). 65% of Google searches already end without a click, and with AI Overviews active that rises to 83% (Omnibound, 2026). Pouring most of your budget into a lever whose market is halving is a bet against the numbers.
This does not mean abolishing technical SEO: a fast, crawlable, well-structured site stays the prerequisite for being extracted by AI. It means the marginal spend on bought links and keyword-by-keyword optimisation has a decreasing return, while spend on media-mediated reputation has an increasing one.
What makes a brand appear inside ChatGPT and Perplexity answers?
The short answer: earned media. AI engines recommend the brands they find cited by sources they trust — editorial publications, analyst reports, review platforms. Muck Rack, analysing over 25 million links, found that 84% of AI citations come from earned media, while paid content accounts for 0.3% (Muck Rack, 2026). Other measurements converge: earned media reportedly drives around 325% more citations than owned content alone (AuthorityTech, 2026).
Anyone who knows marketing science recognises the mechanism instantly: it is mental availability applied to machines. Just as a consumer recalls the brand most "available" in memory at the moment of need, a language model recalls the brand most frequently and consistently associated with a topic in the text it was trained on. Repeated mention on authoritative sources builds that association. The same logic governs the GEO guide we link below.
There is also a second, technical and structural level: the content must be extractable. Synthetic answers up front, data with inline sources, tables instead of numbers buried in images, FAQs in plain HTML, headings phrased as real questions. These are the operational rules detailed in How to appear on ChatGPT, Perplexity and Gemini: the GEO guide. The Princeton / Georgia Tech / Allen AI study presented at KDD measured that applying GEO techniques lifts visibility in generative answers by up to 40%.
Why write for many parallel searches, not one keyword?
A generative engine doesn't take a query and return ten links. It decomposes the user's question into sub-questions (query decomposition), queries multiple sources for each and recomposes a single answer. The user, in turn, no longer types "SEO agency Milan": they ask "what's the difference between an SEO agency and one that also does GEO, and how much does it cost for an SMB?". A page built on a single keyword doesn't cover that semantic spread and gets skipped.
So content must be conceived as a cluster: one topic owned in depth, every H2 answering a concrete sub-question, data the model can extract sentence by sentence. You no longer optimise "the keyword", you own the meaning. And this is where you need to know — with data, not guesswork — which questions to cover, which are already saturated, and where competitors show up.
PR plus a single intelligence tool: the Deep Marketing approach
This reading drives the way we work, and why it beats both the standalone tool and the SEO-only agency.
PR to build the authority AI reads. If 84% of AI citations stem from earned media, getting a brand onto major outlets is not an image exercise: it is the most direct way to tell AI who you are and lend you authority. Deep Marketing's press office service places clients on recognised publications, building the authoritative, repeated mentions models use to decide who to name. It is the job link building promised and can no longer deliver. The same logic applies to us: Deep Marketing was listed by Techreviewer among the top digital agencies of 2026 — verifiable earned media, not self-celebration.
A single digital intelligence tool, not five disconnected subscriptions. Measuring GEO with one tool, Search Console with another, analytics with a third and competitor visibility "by feel" is the surest way to never understand what works. The DM Intelligence platform unifies analytics, Google Search Console, GEO monitoring and content insight in one place: it tells you which topics are worth covering and which to ignore, where and how often competitors get cited by AI, and how your share of voice in generative answers moves. That's the difference between acting on data and guessing — exactly what neither the isolated tool nor the SEO-only agency offers.
The combination — authority built through PR, extractable content on a technically blazing-fast site, and a single dashboard closing the measure-decide-act loop — is what makes the difference today. It isn't "SEO vs GEO": it's about stopping paying for the wrong lever and investing where the numbers are moving.
FAQ
Is SEO dead?
No. Technical SEO — a fast, crawlable, well-structured site — remains the prerequisite for being extracted by AI. What's losing value is paid link building and single-keyword optimisation, not the technical quality of the site.
Are backlinks useless now?
They still help traditional ranking, but they're worth a fraction of before and correlate with AI citations far less than unlinked mentions (0.218 vs 0.664 per Ahrefs). It pays to shift budget from link quantity to the quality of media mentions.
Why does PR matter for AI visibility?
Because 84% of ChatGPT, Perplexity and Gemini citations come from earned media (Muck Rack, 2026). Appearing on authoritative outlets builds the repeated mentions language models use to associate a brand with a topic and recommend it.
Do I really need a single tool or are separate tools fine?
Technically you can use five different tools, but disconnected they don't tell you what works. A single platform combining analytics, Search Console, GEO monitoring and competitor visibility lets you decide what to cover and what to ignore on real data, not intuition.
How long does it take to see a brand cited by AI?
New content typically starts appearing in AI answers about 14-21 days after publishing and indexing. Authority from earned media accumulates more slowly but is more stable over time.
Sources
- Ahrefs / Soar - Backlinks vs Brand Mentions: 75,000 brands, unlinked mentions 0.664 vs backlinks 0.218 (2026)
- Muck Rack - Earned media drives 84% of AI citations (May 2026, 25M+ links)
- Gartner - Organic search traffic -25% by 2026, -50% by 2028
- Omnibound - Zero-Click Search Statistics: 65% no-click, 83% with AI Overviews (2026)
- WordStream - GEO vs SEO 2026: top-10 / AI-citation overlap collapsed to 17-38%
- AuthorityTech - Evidence That Earned Media Drives AI Citations (2026)
- Deep Marketing - Press Office & Events service (earned media on authoritative outlets)
- DM Intelligence - Analytics, Search Console, SEO & GEO platform
- Deep CMS - Blazing-fast, GEO-ready website
