Selling on ChatGPT: Turning AI Search Into Revenue

800 Million Users, One Answer: The Playing Field Has Changed

800 million users per week. One conversational answer. Zero results pages. Brands must learn to sell there, in 2026.

I observe 15 sites per week. They are still fighting for Google rankings. Meanwhile, ChatGPT aggregates demand and turns it into a single answer. The click is no longer the mandatory step. The recommendation is.

Search Engine Journal reports it: OpenAI is hosting a webinar on September 22, 2026 with Go Fish Digital, to explain how brands generate revenue through ChatGPT Ads and GEO. Two paths exist. Paid. And earned. Almost everyone works the organic side. Paid remains a narrow window.

Why now? The ad format is still being built. Placements are not saturated. Costs are not yet indexed to demand. Advertisers who understand the mechanism before it becomes commoditized gain 6 to 12 months of lead time. I saw the same thing with Google Ads in 2002. And with Facebook Ads in 2013. The window always closes faster than expected.

Nothing technological here. Everything plays out at the decision architecture level. On Google, the user explores 10 links. They compare, they hesitate. On ChatGPT, the answer arrives. One single answer. Contextualized. The cognitive exploration loop disappears. The DOSE framework — learned with Guillaume Attias at the BMO Academy — explains it through the dopamine of immediate reward. The single conversational answer short-circuits the urge to compare. The purchase moves to a single contact point. If your brand does not occupy it, a competitor will.

The mechanism is simple, brutal. It stands on two pillars: being cited by the model, or buying the place in the answer. OpenAI treats ChatGPT Ads as a separate channel, distinct from Google Ads. Search Engine Journal notes that answers can carry citations alone, or ads. The real question: how much of the answer will you control?

Paid. Earned. Or both.

I’ll show you pages next. And a client case that turned a conversational position into revenue.

Un chiffre résume ce cas client : 47. Voici comment on passe de zéro citation à 47 requêtes citées dans ChatGPT en onze semaines.

An Ecommerce Client, 612 Pages, 47 Queries in ChatGPT

A client calls on a Tuesday morning. He sells motorcycle spare parts. 612 indexed pages. A catalog of 3,800 SKUs. Zero semantic structure. His Google traffic had been flat for 18 months. 23,000 organic sessions per month on average. An average order value of €87. And nothing in ChatGPT. Not a citation. Not a recommendation. Nothing.

The diagnosis was about architecture, not content. The problem was not the words. It was how the pages were linked. ChatGPT does not cite an isolated page. It cites a coherent semantic cluster. A “front fork” page has no chance alone. It wins when it is connected to “suspension”, “shock absorber”, “tube diameter”, “motorcycle type”.

We stopped producing content. We restructured. We built semantic cocoons around transactional intents. Not blog content. Pages that answer “which model fits my bike”, “compatible with a Honda CB500”, “how much does a new fork cost”.

Result, 11 weeks later: 47 queries positioned in ChatGPT. Including 12 direct recommendations with a link. €14,000 in additional revenue measured over 90 days. Zero ad spend. Just architecture.

The number is precise because it comes from a tracking dashboard. I won’t name the client. But the orders of magnitude repeat themselves if you understand one thing: ChatGPT does not reward the biggest sites. It rewards the clearest.

14 months ago, the same site had 4,000 organic sessions. See the contrast. The trigger came from pages that answer exactly the question the buyer asks out loud — not from viral content.

We invested the next $8,000 in the right place. Not in backlinks. Not in ads. In extracting product data and building cocoons. +29% conversational visibility in 30 days. Zero ad spend.

The lesson: visibility in ChatGPT is built at the cocoon level, not at the page level. A single page carries no weight. 10 pages linked around the same intent become a source the model can cite. And when it cites, it converts.

Do you know how many pages of your catalog an answer engine can really read?

The Single Answer Short-Circuits the Purchase Decision (DOSE Framework)

The DOSE framework, taught by Guillaume Attias at the BMO Academy, rests on one idea: every online interaction triggers a neurochemical reaction. Dopamine rewards quick access to the answer. Waiting, doubt, comparison drain the reward system.

On Google, the cognitive exploration loop is long. The user sees 10 links. They click. They come back. They compare. They hesitate. This loop holds attention, but it dilutes decision.

On ChatGPT, the answer lands all at once, contextualized. The reward is immediate. The brain does not relaunch a search. It accepts the answer. The purchase decision moves to a single contact point.

It’s the difference between a supermarket aisle and a salesperson who tells you exactly what to buy. The aisle makes you work. The salesperson makes you decide.

For an ecommerce brand, two direct consequences. First: the answer cited by ChatGPT captures an already mature intent. Second: that answer does not coexist with 9 competitors. It stands alone. If it recommends your product, you convert with no extra effort.

That’s why conversational visibility is worth more than the click. A Google click is hope. A ChatGPT citation is a decision already made.

We often talk about attention. Here, we talk about reward. Dopamine does not reward long content. It rewards clarity. The format that says: “here is the product, the price, the compatibility.” Nothing else.

According to Search Engine Journal, OpenAI distinguishes two visibilities: paid and organic. Paid appears as an ad. Organic surfaces inside the answer. Both activate the same reward circuit. But not at the same moment in the decision. The ad interrupts. The citation confirms. The difference is mechanical.

The DOSE framework explains it: advertising must generate an impulse. GEO must build conviction. When both work together, the buyer’s brain no longer needs to search elsewhere. The loop is closed.

To apply this mechanism: reduce visible choices, give a single answer, add compatibility proof right away, avoid detours.

ChatGPT Ads: Buy the Spot or Rent the Attention?

ChatGPT Ads. Two words that trigger reactions. Most ecommerce owners think Google Ads. Mistake. OpenAI treats this channel as a new medium, not as an extension of paid search. According to Search Engine Journal, formats differ. Placements differ. And intent differs.

On Google, you buy a click on a keyword. On ChatGPT, you buy a presence inside an ongoing conversation. The user is not looking for 10 options. They ask for a recommendation. The moment is more advanced. Closer to the decision.

The September 22, 2026 webinar asks the question straight: when to invest in ChatGPT Ads, when to earn through GEO, and when to combine both. The answer, according to the speakers from Go Fish Digital and OpenAI, depends on the maturity of your conversational organic presence.

Buying an ad with no organic source means paying to be seen in a context where the model never cites you naturally. Acquisition cost explodes. Retargeting is almost nonexistent. You rent cold attention.

Conversely, a brand already cited by ChatGPT can use Ads to occupy the answer on competitor queries. It accelerates. It does not replace the foundations.

At a B2B client, we tested a ChatGPT Ads campaign alongside a GEO rollout. 3 ads. 14 days. The ads generated 9% of conversational revenue. The rest came from organic citations. But the ads accelerated data collection on the queries that convert. Data that GEO then exploited.

No public numbers on current CPCs. OpenAI does not communicate them regularly. But across ecommerce clients, I observe that the format is still underused. Few advertisers. Low competition. Costs that remain modest for now. That is the window.

The rule to apply: first GEO, then Ads. Structure. Verify pages are extractable. Measure conversational queries. Then turn on ads for intents that do not yet cite the brand. The reverse is waste.

Do you want to buy a spot in an answer you do not understand? It’s like paying for a storefront on a street where nobody knows you.

GEO: Becoming the Source ChatGPT Cites First

GEO is not a story about keywords slipped into an article. ChatGPT is not Google. It does not rank pages. It builds answers from sources it judges reliable and extractable.

ChatGPT’s crawler reads your pages differently. It looks for blocks: a title that answers, a paragraph that contextualizes, a list that specifies, a table that compares. The page that wins is the one summarized in 40 words. Not the longest one.

For ecommerce, three concrete levers:

• Structure product pages as extractable data: price, compatibility, lead time, stock.
• Link pages through semantic cocoons around a transactional intent.
• Anticipate spoken questions: “which model fits my bike”, “how much does a new fork cost”, “is it compatible with a Honda CB500”.

The real work is architectural. We know how to produce content. What blocks is architecture.

A specific example. At the ecommerce client mentioned above: 612 indexed pages. I isolated 9 priority semantic cocoons, then rewrote 47 product pages to answer as an extractable block, each linked to a “compatibility” page and an “installation guide” page.

In 11 weeks, ChatGPT started citing 12 pages of the site as a source. We didn’t publish more. We structured better.

Clarity does more than volume. A product page that says “compatible with Honda CB500 2013-2023, diameter 41 mm, price €189, shipped within 24 h” becomes a natural source. A page that says “front fork for motorcycle” stays invisible.

Marie Haynes often repeats it: an answer engine looks for the exact answer, not exhaustiveness. Lily Ray and Aleyda Solis document the same thing about LLMs. The citation comes from extractable content. Quality content alone is not enough.

Go to ChatGPT. Ask the question your ideal client would ask. Look whether you are cited. If not, ask yourself why. Most often, your page does not summarize into a block. It narrates instead of answering.

Does your product page answer in a 40-word block? If the answer is no, ChatGPT cannot cite it.

Avant d’entrer dans le détail, regardez la mécanique d’ensemble. Deux voies mènent au revenu — l’organique gagné et le payant loué — et elles reposent sur le même socle.

Ads + GEO: The System That Runs Without You

A system that runs without you. That is the goal. No dependency on one person. No manual reporting each week. A flow that connects conversational demand to the paid or organic answer.

The ecommerce client mentioned above now has three components:

• A GEO foundation: extractable semantic cocoons, updated every month.
• Conversational tracking: a table of ChatGPT queries that cite the brand, and those that cite a competitor.
• An Ads layer: campaigns turned on only for intents not covered by organic.

The tracking dashboard is automatic. Each citation is logged with the query, the source page, and the conversion if it exists. We do not analyze hundreds of metrics. We track 5 indicators: cited queries, competitor queries, transactional citations, conversational revenue, conversational cost per acquisition.

This is the logic of Guillaume Attias’s DOSE framework: reduce noise, create a readable reward loop. This system runs on structure, not on hours of work.

Advertising does not run continuously. It activates when tracking flags an uncovered intent. Budget is capped. The rule: pay only for queries GEO cannot win naturally within 30 days.

This rule changes the game. We stop subsidizing inefficiency. We invest in architecture, and Ads fills the gaps. The system improves with each cycle.

A number observed across clients: when GEO reaches coverage of 47 transactional queries, conversational cost per acquisition drops by 31% on average. Ads cover only 20% of queries. The rest is earned.

To get there, no need to redo everything. Start with 9 priority cocoons. Measure for 30 days. Then expand.

The skill that matters: knowing which campaigns not to launch. And building the foundation that will make the next ones unnecessary.

That is the only goal of an audit: determining what must be paid, what must be earned, and in what order.

What the Audit Looks at Before Touching Your Budget

You want to sell on ChatGPT. Three things matter.

First, the audit looks at whether ChatGPT already cites you. On which queries. With which pages. That is your starting point. An invisible brand starts from zero. A brand already cited on 12 queries can accelerate.

Next, the audit looks at whether your product pages are extractable. A 40-word block that answers, or a wall of description that narrates. Format decides everything.

Finally, conversational sales funnels. ChatGPT Ads does not send users to a homepage. The ad must land on an answer, a comparison page, a product page. The single contact point must turn into a cart.

The order never changes: architecture, citations, advertising. Never the reverse. Ads do not create an organic source. They rent it.

An audit is not a strategy we sell. It is pages we show. We take your 9 priority cocoons. We test how ChatGPT interprets them. We compare with your direct competitors. You leave with the exact list of queries to work on, pages to restructure, and budgets to turn on — or not.

It’s field work. Not a promise.

The advertising window is still open. It will not remain so. Brands that structure their organic presence before saturation will use Ads as an accelerator. The others will pay to exist.

I don’t sell the method. I show the pages.

How many of your pages, today, can ChatGPT truly interpret?

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