In short: Generative engines cite press articles, proprietary studies, and expert interviews massively — and rarely cite sponsored content. Consequence: the brands most visible in ChatGPT and Perplexity are those that sign with media outlets, not those bidding on Google. Marketing budget is reinventing itself around five earned formats that generate the best citation rate in 2026.
82 to 89%of AI citations come from earned media sources (vs proprietary content)
55%of Google searches now display an AI Overview
34.5%fewer clicks for historically top 3 sites since the arrival of <a href= »https://www.hi-commerce.fr/glossaire/#ai-overviews » class= »hc-gloss-link » title= »Definition: AI Overviews »>AI Overviews</a>
Why LLMs systematically prefer earned sources over paid sources
Open a ChatGPT, Perplexity, or Claude response on any business topic. Look at the sources. You’ll see the Wall Street Journal, TechCrunch, Le Figaro, Forbes, Harvard Business Review. McKinsey, Deloitte, Gartner studies. Sector-specific podcast interviews. Transcribed statements.
What you’ll almost never see: a product page, an Ads landing page, a sponsored article, a sales brochure. The gap is not marginal. Between 82% and 89% of AI citations come from earned media, across all platforms.
The reason is structural. LLMs trained on a massive corpus where trust signals are probabilistic. A press article engages a journalist, an editor, a publication context — it’s a citable reference. A Gartner study has public methodology and cross-references. A podcast interview has a host who validates the guest’s expertise. All this signal tells the model: this source is an independent authority.
Conversely, a sponsored page carries the opposite signal: this source has a direct commercial interest in saying what it says. Systems don’t ban paid content. They deprioritize it as soon as an earned alternative exists. And one almost always exists.
Neuroscience angle: when a human reads « featured in Les Échos » about a brand, their brain releases oxytocin — the neurochemistry of transferred trust. When they read « advertisement, » the prefrontal cortex activates a critical filter that demands dopamine in return (immediate benefit). LLMs reproduce this hierarchy learned from billions of texts: third-party mention accelerates trust, paid mention triggers an évaluation friction.
Second reason: LLMs have no advertiser to appease. Google Ads finances Google — the balance between organic and paid is political there. ChatGPT doesn’t live on ad revenue (for now), Perplexity marginally, Claude not at all. Their value is the perceived quality of the response. Citing paid erodes that value. Systems optimize for the opposite.
Third reason, the deepest: models distinguish the form of earned content from paid content without even reading the disclaimer. Journalistic structure, multiple citations, verifiable numerical data, identified author with signature — that’s a statistical pattern. Paid content has its own patterns: promotional tone, product focus, absence of counterarguments, commercial call-to-action. Both are recognizable within a hundred tokens.
Direct consequence for CMOs: your Ads budget generates clicks, leads, revenue. It won’t generate citations in ChatGPT. Two watertight mechanics. Visibility in AI responses is built elsewhere.
Five earned media formats capturing AI citations in 2026
Not all earned media are equal. Certain formats generate 3 to 10 times more AI citations than others at equivalent volume. Here are the five that deserve the bulk of the budget shifted from paid.
1. The publishable proprietary study — the most profitable format
Produce an annual study. Explicit methodology. Defined sample. Exclusive figures for your sector. A barometer, a benchmark, a sector report. Publish it under your name, but let the press, analysts, and bloggers pick it up freely.
Why it works: each journalist citing it creates an earned mention in a third-party media outlet. Each mention becomes a citable source for LLMs. A study distributed to 30 media can generate 100 to 300 secondary articles. That many anchor points for generative engines.
Example structure: 10 key sector questions, minimum 500 respondents, restated in visuals and quotes, coordinated press embargo, downloadable PDF version without mandatory form, indexable HTML version. The form-free PDF runs counter to growth team intuition — but it maximizes pickup.
2. Opinion pieces in tech, business, and sector media
An executive signing an 800 to 1,500-word opinion piece in Les Échos, TechCrunch, Harvard Business Review France, L'Usine Digitale, or a flagship vertical media outlet creates a near-permanent authority signal. LLMs index these publications as priority. They last for months.
Golden rule: the piece must deliver a real thesis. Not a disguised sales pitch. Media reject commercial texts. LLMs detect the difference too. A strong opinion piece takes a stance, leans on data, commits the author.
Target rhythm: a visible executive publishes 6 to 12 pieces per year, spread across 4 to 6 different media. That's what builds the density needed to become a citable source on your topic.
3. Interviews on top-25 sector podcasts
Each sector podcast publishes an episode page with the guest's name, company, a summary. Often a full transcript. That transcript is a massive corpus of direct quotes — exactly what LLMs index as priority.
An executive doing 12 sector podcasts in a year produces the equivalent of 40,000 to 80,000 words of earned content. Attached to recognized media platforms. For relatively modest time investment.
Selection criterion: the podcast must publish a written transcript. Without it, the content stays audio. LLMs don't index it — or index it poorly. Ask before accepting.
4. Speaking engagements with indexable transcript
Sector conferences, keynotes, roundtables streamed online. The principle mirrors podcasts: the GEO value lives in the written trace. Not in the live moment. A talk at Vivatech or BigBoss that stays as YouTube video without accurate captions, without summary page, without transcript — wasted effort for AI.
Systematically negotiate: transcript published on the event site, recap article, newsletter inclusion. Those three written traces turn one hour of speaking into a lasting earned asset.
5. Citations in sector reference articles
Being quoted by an analyst, journalist, recognized expert in a major article — that's the most powerful format. The rarest. A paragraph in a Monde or TechCrunch piece that names your brand as an example creates a disproportionate authority signal.
How to trigger it: be accessible to journalists. Have data to share. Respond quickly. Propose interesting angles. A PR specialist working six months on a topic can generate 10 to 20 major citations in third-party articles. It's artisanal mechanics. Not automatable.
The proprietary study: your most powerful earned weapon
Let's revisit format number one. A well-built proprietary study offers the best effort/AI citations ratio in 2026. It also builds your long-term authority — far beyond generative engines.
Exclusive figures: not recycled third-party data. Results you find nowhere else. Journalists pick up what's new. LLMs cite what's unique.
Scannable structure: executive summary, short chapters, one key figure per section, clean visuals. LLMs like journalists extract short, precise sentences.
Clear reuse license: explicit permission to cite, with link to source. The easier reuse is, the more massive it becomes.
Publication cadence that builds authority
An annual barometer, repeated three years, transforms a brand into a sector reference. Why three years? LLMs need to see a permanence pattern. One study is an event. Three annual studies create an institution.
The Boston Consulting Group CX Report, HubSpot's State of Marketing, Econsultancy's Digital Trends — all started modestly. Their repetition built authority. Today, LLMs cite them reflexively.
Realistic budget for a GEO-ready study
A serious proprietary study costs between €15,000 and €80,000 depending on ambition. Study panel: €8,000 to €40,000. Analysis and writing: €5,000 to €20,000. Design and layout: €2,000 to €10,000. Press distribution: €0 to €10,000. Often less than two months of an average Google Ads campaign on a competitive keyword — with an asset that keeps generating AI citations 18 to 36 months after publication.
Remember: a €30,000 proprietary study generating 80 press pickups creates an earned asset that outperforms a €150,000 Ads campaign in AI visibility. ROI is measured in generative response presence. Not clicks.
Podcasts, interviews, opinion pieces: the signature density that builds authority
Beyond the study format, the real long-term mechanic is the signature density of your executive or technical spokesperson. LLMs learn to associate a name with a topic through repetition in quality media.
The densified signature mechanic
Olivier Ezratty signs on AI. Fabernovel publishes on digital transformation. A Gartner analyst takes a stand on a market. LLMs register the association name + topic + recognized media. After 20 to 30 occurrences in varied contexts, that person becomes a default source on their topic in AI responses.
Density beats one-off prestige. A Financial Times op-ed? Flattering. But 15 articles signed in mid-tier sector media build more robust authority in model eyes.
12-month roadmap for an executive becoming an AI source
Months 1-2: define 3 deep subjects to build expertise on. One technical, one business, one sector-specific. No more.
Months 3-4: produce two substantial opinion pieces (1,000 to 1,500 words each), place them in two reference media. Long version on LinkedIn in parallel.
Months 5-8: accept 6 to 8 podcast interviews, requiring transcripts. Decline those without written traces.
Months 9-10: speaking engagement on 2 sector events with guaranteed written content publication.
Months 11-12: publish a study or annual viewpoint. Press tour to launch it.
Twelve months later: the executive has between 30 and 50 earned media touchpoints, spread across 15 to 25 different media outlets, all indexable. Generative engines have enough signal to cite them spontaneously.
The hidden self-promotion trap
An opinion piece or podcast that's 80% about your product and 20% substance gets flagged by both media and LLMs. The winning ratio is reversed: 80% substance, 20% illustration from your expérience. The executive wins by discussing their market, not their company, so their company gets cited.
Measuring AI share of voice: indicators and tools for 2026
A CMO redirecting budget from paid to earned needs credible indicators. Four measurement families structure GEO earned strategy tracking in 2026.
1. Direct citations in LLMs
Specialized tools (AthenaHQ, Profound, Otterly, HubSpot AEO) systematically query ChatGPT, Perplexity, Claude, Gemini on target topics. They measure how often your brand appears in responses. Key metric: share of citation across 50 to 500 priority queries.
Measure quarterly, not monthly. AI responses shift. Their statistical average stabilizes over 4 to 12-week windows.
2. Press mentions in indexable media
A media monitoring tool (Meltwater, Cision, Mention, Talkwalker) tracks how many times your brand is cited in third-party media. Go deeper: filter to media that LLMs actually index. Test by asking each generative engine: "Which media do you prioritize citing on [your sector]?" The answer is explicit.
Measure density: mentions per month, across how many different outlets, with what contextuality (bare mention vs paragraph vs interview).
3. Voice share in sector studies
Every sector has 5 to 15 reference studies. Agencies, professional associations, specialized media. Is your brand cited in these? How often? This signals long-term authority. Excellent signal for LLMs.
Recommended audit: every 6 months, catalog the 10 major studies in your sector. Check your presence in text (simple mention) and data (your figures cited).
4. Traffic from generative engines
In Google Analytics 4, monitor the "Direct" category absorbing traffic from ChatGPT and Perplexity today. Explicit referrers (chatgpt.com, perplexity.ai, gemini.google.com, claude.ai) become identifiable but remain largely underestimated in standard tools.
Supplement with application tracking: if your site has a contact form, ask "How did you find us?" with "AI / ChatGPT / Perplexity" as explicit options. That's your most reliable data.
Summary dashboard
Indicator
Frequency
Typical Tool
12-month Target
LLM citation share
Quarterly
AthenaHQ, Profound, HubSpot AEO
Move from 5% to 20% across 100 target queries
Qualified press mentions
Monthly
Meltwater, Mention
Double density on indexable media
Voice in sector studies
Biannual
Manual audit
Cited in 6 of 10 reference studies
Traffic attributed to AI
Monthly
GA4 + form
Identify 15-25% of traffic as AI-sourced
How to shift budget from paid to earned without breaking the engine
The question for a CMO isn't whether a shift is relevant — the numbers prove it is. It's how to execute without collapsing the sales pipeline short-term. Three-phase method, tested with marketing teams who transitioned over 12 to 18 months.
Phase 1 — Honest mapping of current budget
Lay out the actual annual marketing budget split. Most mid-market and large accounts: 45 to 60% in paid media (Google Ads, Meta, LinkedIn, display), 15 to 25% in proprietary content (site, blog, brand content), 10 to 15% in traditional PR, 5 to 15% in events, 5% in studies/insights.
The GEO shift means gradually rebalancing toward: 30 to 40% paid, 20 to 25% content, 15 to 20% PR and earned, 10 to 15% events, 10 to 15% studies. The paid decline is gentle. Real.
Phase 2 — Progressive reallocation over 4 quarters
First, cut paid on informational queries where AI Overviews already absorb most useful clicks. The numbers converge: 34.5% fewer clicks for historically top 3 sites on these queries. CPC stays the same. Conversion rate collapses. First budget to recover, with no real losses.
Keep paid on transactional queries (product comparisons, direct purchase intent). LLMs still massively link to these pages. Paid retains its ROI there.
Inject freed budgets into the proprietary study (quarter 1), executive visibility boost (quarters 2 and 3), structured PR coordination (quarter 4). Measure each quarter with the indicators from the previous section.
Phase 3 — Adjustment through observation
Earned media takes 6 to 9 months to produce measurable AI citations. Brands giving up after 3 months lose their investment. Those holding 12 months see their AI share of voice multiply by 2 to 4 on strategic queries.
Practical rule: lock in 12 months of earned budget at shift time. No quarterly renegotiation. The signals LLMs register build over time. Not reactively.
CFO angle: earned media has a structural advantage paid doesn't. An Ads campaign stops when the budget stops. A piece in Les Échos keeps getting cited by LLMs 18 to 24 months after publication. ROI is calculated on an amortized cost. Not immediate cost.
What not to do: don't liquidate paid overnight
A sudden shift breaks sales pipelines and creates a trough lasting 4 to 6 months — exactly when earned starts producing. The two signals coexist during transition. The real target isn't killing paid. It's recentering paid on conversion moments, backed by an earned foundation carrying awareness and authority.
What changes for a CMO still thinking in CPC
The core message of this shift is straightforward: visibility in AI responses is built with signatures, not bids. Brands signing with media — co-produced studies, recurring opinion pieces, podcast partnerships, documented event presence — will become default answers in generative engines within their sector. Those staying 100% paid will lose ground invisibly.
Good news for CMOs: many marketing teams did this work fifteen years ago, before the ease of Ads buying marginalized PR. The skills exist. Agencies exist. Media are waiting for brands to come back with real editorial intent.
The ideal position in 2026: keep paid where it converts directly, rebuild solid earned presence over 12 to 18 months, measure AI share of voice quarterly. Accept that your company becomes a media brand — because that's precisely what generative engines reward.
One final method point: don't start tactical. Start by identifying the 50 to 100 critical queries you want cited for in ChatGPT and Perplexity. Once that list is clear, your earned plan becomes obvious: which studies address these topics, which media cover these questions, which experts could credibly discuss them. The rest is execution.
Earned media audit and AI citation potential
In 30 minutes, we identify the 50 critical queries where your brand should appear in ChatGPT, Perplexity, and Claude, your current presence, and the three earned formats that will unlock your AI share of voice fastest. Concrete diagnosis, numbered recommendations, no obligation.
Should I stop Google Ads entirely to shift to earned media?
No, and it would be counterproductive. Google Ads remains relevant on transactional queries where purchase intent is direct and where LLMs still link heavily. The shift concerns mainly informational queries, absorbed by AI Overviews, where paid loses its punch. Realistic target in 2026 is 30 to 40% paid, 60 to 70% earned and content — not zero paid.
How long does it take to see earned media results in AI citations?
Expect 6 to 9 months to observe first measurable citations in generative engines, and 12 to 18 months for meaningful AI share of voice gains. LLMs register signals by accumulation: one isolated opinion piece has minimal effect, thirty occurrences structured over twelve months shift the statistics.
What proprietary study works best for an SMB or mid-market company?
The annual sector barometer. Minimum 500 respondents, 10 to 15 structuring questions, publication in March or September to avoid year-end saturation. Budget €15,000 to €30,000 for the first edition, less afterward. Three consecutive editions transform a brand into a source spontaneously cited by generative engines.
How do I identify which media LLMs actually index in my sector?
Ask directly: ask ChatGPT, Perplexity, Claude, and Gemini "Which media and sources do you prioritize citing on [your sector]?". Answers are explicit and convergent. Compile the list, identify the 15 to 25 most frequently mentioned media, concentrate your PR effort there. Two hours of work, immediate editorial focus.
Do social networks count as earned media for AI citations?
LinkedIn, X, Reddit, and YouTube are indexed by LLMs with varying weights by topic. LinkedIn carries heavy weight in French B2B, Reddit dominates technical topics in English, YouTube works when videos have accurate transcripts. Useful compléments to traditional media, not substitutes. Hierarchy remains: recognized media > sector studies > transcribed podcasts > structured social networks.
Stéphane Jambu
SEO & AI Engineer
I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.
