ChatGPT Ads and GEO: Allocating Paid Budget and Earned Visibility

37 000 organic sessions. 14 months ago, 4 000.

37 000 organic sessions per month. 14 months ago, the same site had 4 000. The catalog stayed the same. The ad budget stayed the same. What changed? The semantic architecture. And a new visibility surface: AI answers.

Search Engine Journal published a cross-analysis between OpenAI and Go Fish Digital on September 23, 2026. The specific topic: where to place budget between ChatGPT Ads and earned AI visibility. A question that keeps e-commerce leadership up at night. Rightly so.

Visibility no longer plays out only on Google. It plays out in the answers from ChatGPT, Gemini, Perplexity. Some brands buy this visibility through ChatGPT Ads. Others earn it through GEO, Generative Engine Optimization. Both exist. Not necessarily at the same time on the same queries.

Take an e-commerce site with 945 product pages. It hits a plateau. 47 thematic clusters poorly structured. 310 product pages rewritten. The rest is internal link architecture. 14 months later, 37 000 sessions. +820 %. Growth came entirely from organic improvements.

According to OpenAI public data, ChatGPT exceeds 800 million weekly active users. A growing share of these exchanges contains purchase intent. The 2026 marketing budget must account for this. Paid and earned are complementary. They work together.

Investing in AI? Wrong question. The right one: on which queries to pay, and on which to let GEO work. Full clarity. Paid. Earned. Both.

ChatGPT Ads is a Response Surface, Not a Classic Auction.

ChatGPT Ads is not Google Ads with a different logo. In the Search Engine Journal article, OpenAI describes sponsored responses as contextual. They appear in the conversation thread, tied to the intent expressed in natural language. No frozen keyword. No results page. An integrated response.

Direct consequence: the format rewards conversational relevance. An ad poorly aligned with the question creates friction. An aligned ad strengthens credibility. Everything plays out in a few turns of dialogue.

Go Fish Digital, cited in the article, draws a clear distinction. On one side, paid visibility: a sponsored placement, bought on a conversational surface. On the other, earned visibility: an organic citation in the response, stemming from authority, content clarity, and site structure.

• Paid: sponsored response, bought on conversational queries.
• Earned: organic citation in the response, triggered by content quality and semantic structure.
• Hybrid: pay to accelerate, then capitalize on the earned citations.

A client sells electric bikes. On the query « best urban electric bike 2026 », ChatGPT Ads places their brand at the top. GEO gets their comparison cited 47 times per week in non-sponsored responses. Both channels coexist. They serve different intents and different decision moments.

What I observe with my clients: paid brings immediate visibility. Earned brings lasting visibility. Pitting the two against each other means missing the mechanics.

GEO: Earned Visibility Is Measured Beyond Clicks

Earned AI visibility is measured beyond clicks. Some sites are cited in an answer without receiving direct traffic. The citation anchors the brand in memory. It prepares the conversion later, on a direct channel or a branded search.

Go Fish Digital, in an analysis picked up by Search Engine Journal, advises auditing GEO proof. Concretely: identify the queries where the site appears, isolate the pages used as sources. Then verify whether the citation matches the intended angle.

A GEO audit on 120 transactional queries holds its share of surprises. A secondary page may cannibalize the main page. Elsewhere, a product page is cited for a category query. The fix is often structural, rarely editorial.

I observe a frequent gap: teams track sessions and Google positions. They ignore AI citations. Result: invisible acquisitions go unnoticed. Or worse, an unsuitable page becomes the reference source. GEO is measured like any other asset.

The good news: 20 strategic queries are enough for a first diagnosis. Full clarity.

Avant de dépenser un euro, on passe par quatre étapes qui transforment une intuition en décision chiffrée. Voici la mécanique, du catalogage à la mesure.

The Allocation Protocol: 4 Steps Before Spending

Opening a ChatGPT Ads budget blindly is the worst reflex. The protocol I apply fits in 4 steps. Each step gives a quantified decision.

1. Segment the catalog: isolate the 20 to 30 high transactional intent queries, the ones that generate 80 % of margin. The rest waits.
2. Audit current GEO presence: for each strategic query, note whether the site is cited, on which page, in which position. A manual extraction on ChatGPT and Gemini is enough for a first overview.
3. Test ChatGPT Ads on a limited scope: 15 to 20 queries, geolocated, with a capped budget. Measure cost per click, cost per citation, and above all cost per order after click.
4. Compare marginal cost: is the paid acquisition cost lower, equal, or higher than the margin generated by an earned citation? If yes, accelerate. If no, invest in semantic architecture.

At an e-commerce client, this protocol shifted 4 500 $ from paid budget to GEO over a quarter. Organic traffic continued to grow. Total acquisition cost dropped. Visibility on hot queries remained intact.

The allocation is far from binary; it moves. A query can be assigned to paid at launch, then switch to earned visibility once authority is established. It all comes down to continuous measurement.

Is your site cited nowhere? Invest in GEO first. Paid alone costs too much in the long run.

Toutes les requêtes ne méritent pas le même traitement. Selon le cycle de vie du catalogue, le paid et le earned n’ont pas le même rendement.

When Paying Accelerates, When Architecture Suffices

Not all queries deserve the same treatment. On an e-commerce catalog, I see three situations.

First situation: launch. You release 120 new seasonal references. Nobody cites them yet. ChatGPT Ads gives immediate visibility on product queries. Meanwhile, GEO prepares the ground: structured pages, product data readable by crawlers, coherent internal linking. Paid covers the first days. Earned takes over.

Second situation: stable catalog. 2 500 established references, existing comparison content, decent domain authority. Here, GEO suffices on most queries. Paying to be cited first on « best washing machine 2026 » when your comparison is already cited in second position is wasting budget. Better to reinforce proof and internal structure.

Third situation: competitive attack. A competitor buys ChatGPT Ads placements heavily on your core queries. Your earned citation slips down the conversation. You can respond with paid to hold presence. But the lasting counterattack goes through editorial and semantic work.

A simple rule I apply: paid protects short-term visibility, earned builds long-term visibility. Budgets complément each other. They are allocated according to time horizon.

And if you hesitate, look at margin. A query generating 38 € average margin does not support a paid acquisition cost of 12 € per click. GEO, on the other hand, compounds. Your marginal cost tends toward zero. Not right away. But it tends.

Pour un e-commerce de 500 à 1 000 références, voici la répartition que je testerais sur 10 000 $. Le earned reste majoritaire, le paid sert d’accélérateur ciblé.

What I Would Test with 10 000 $ in 2026

The allocation I would test, as an order of magnitude, for an e-commerce with 500 to 1 000 references. Total budget: 10 000 $. The idea: install a lasting AI presence, with a balanced paid share.

• 6 500 $ for semantic architecture and GEO: restructuring 30 clusters, rewriting 150 product pages, updating structured data, internal linking.
• 2 500 $ for ChatGPT Ads: 20 transactional queries, precise geotargeting, 3 months of testing, weekly tracking of cost per order.
• 1 000 $ for measurement: initial GEO audit, monthly tracking of 20 strategic queries, iterations.

Why 65 % earned? A paid click in an AI interface remains expensive on high-intent queries. An earned citation lasts. And a well-structured site gets cited more often, including on questions you never bought.

Nothing dogmatic here. If your margin is very high, reverse the split. If your catalog is seasonal, temporarily shift more budget to paid. The key point: measure.

Remains the question that matters. Is your site already cited by ChatGPT on your 20 queries that bring revenue? If not, start there. Not with Ads.

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.

Follow on LinkedIn