How to appear in ChatGPT: the two paths that exist and which one suits your company

When a CEO says “I want my company to show up in ChatGPT,” they’re usually asking for two different things without realizing they’re different. One is bought and the other is earned. One appears below the answer with a sponsored label, and the other appears inside the answer, spoken by the model itself. Confusing the two is what makes many companies spend budget in the wrong place.

Since August 31, 2026, the confusion has become more expensive, because OpenAI opened direct access to its ad manager in Europe and Spain is already listed as an available market. This article explains what exactly you’re buying when you buy an ad in ChatGPT, what you can’t buy even if you want to, how the other path works, and what criteria to use to decide which one to start with.

The two paths, in one table

	Ads in ChatGPT	Appearing in the answer
How you get it	Bought in the ad manager	Built with content and authority
Where it appears	Below the answer, labeled as sponsored	Inside the text the model generates
Who it reaches	Only Free and Go plans	Any user, including enterprise plans
Control	High over spend, low over the conversation	Low in the short term, high in the medium term
Measurement	Dashboard with impressions, clicks, CTR and conversions	No dashboard: citation and attributed traffic
Timeframe	Immediate	Months

The short read: the paid path buys presence today and the organic one builds recommendation tomorrow. They don’t compete, they solve different problems.

Path 1: ads in ChatGPT

Where the ad appears and what format it has

The ad is shown below the model’s answer, separated from the text and marked as sponsored. The format includes the advertiser’s name, its favicon, a headline, a short text, an image and the destination page. It isn’t inserted inside the answer and doesn’t change what the model replies.

This matters more than it seems: you’re not buying ChatGPT recommending you. You’re buying a space adjacent to the recommendation. If your ad contradicts what the model just answered, the user notices.

There are no keywords. There are context clues

This is where the mental model of anyone coming from search engines breaks down. In ChatGPT’s ad manager there’s no keyword field. Don’t look for it, it’s not there.

What there is are context clues, which are defined at the ad-group level and consist of a plain-language description of the conversations, topics and situations in which your offer is useful. OpenAI is explicit in its documentation: these clues guide allocation, they don’t work as exact matches and they don’t guarantee delivery in any specific conversation.

Translated into a sales conversation: if someone promises they’ll rank you for the word “forklift” inside ChatGPT, either they haven’t opened the platform or they’re selling you something that doesn’t exist.

What you can control:

• Geography, by market
• Your own data, to build audiences
• Product catalogs, when applicable
• The ad copy and its coherence with the destination page, which is the most underrated lever of the four

How it’s bought and how much it costs to start

It’s bought by impressions (CPM) if the goal is reach, or by click (CPC) if the goal is visits. The maximum bid is set at the ad-group level, and OpenAI recommends starting with a maximum CPC of between 3 and 5 dollars.

The auction is second-price weighted by relevance. That is, paying more doesn’t guarantee showing up: if your ad fits the conversation worse than that of a competitor bidding less, you lose. It’s the same principle you know from other platforms, applied to a conversational context instead of a written query.

The dashboard returns impressions, clicks, spend, CTR, average CPC, average CPM and conversions. Your link’s tracking parameters are preserved on click, so you can attribute in your own analytics without relying on the dashboard alone. The data you receive as an advertiser is aggregated and non-identifying.

The limit that changes the equation in B2B

This is the fact that decides whether the channel is useful to you or not, and it’s the one least talked about: ads are not shown to Plus, Pro or any enterprise-plan users. They only appear on the Free and Go plans, and they’re not shown to minors either.

Think about your real buying committee. The operations director who pays for their own paid subscription. The procurement manager at a company with corporate licenses. The engineer at a multinational on an Enterprise plan. None of them will see your ad. Ever.

This doesn’t invalidate the channel. It delimits it. It works well for reaching profiles who research on their own with the free version, who in technical categories are far more numerous than most marketing departments assume. But if your sale is decided in a committee at a large company, the paid path is not your main path.

Path 2: appearing inside the answer

When someone asks ChatGPT what providers exist for a problem and the model names three companies, those three haven’t paid anything. They’re there because the model has built an understanding of the sector in which those brands appear as a reference. This is called GEO, generative engine optimization, and it’s the natural continuation of SEO, not a separate discipline.

How a model decides whom to cite

Models don’t rank by links the way the classic search engine did. They prioritize, among other factors:

1. Topical authority of the domain. A site that covers a topic with depth and consistency gets cited more than a generalist one that touches on it in passing.
2. Structural clarity. Precise definitions, direct answers in the first paragraphs, lists and tables. What is easy to extract gets extracted.
3. Concrete, verifiable data. Figures, percentages, comparisons. A text with citable numbers gets cited; one with adjectives doesn’t.
4. Frequency of mention across the rest of the web. The model doesn’t only read your website, it reads everything else. If no one mentions you outside your domain, your authority is just a claim you make.
5. Freshness. Content with a recent publication and review date is preferred when the model consults the web.

What you have to change in your content

Most industrial websites are written to convince a human who has already arrived. GEO requires you to write also so that a machine understands you before that human arrives. In practice:

• Answer the question in the first paragraph. Don’t build toward the conclusion: put it at the top and develop it afterward.
• Define the concepts of your category with closed, self-sufficient sentences, of the type “X is a system used for Y in Z contexts.”
• Structure with headings that are your buyer’s real questions, not catalog titles.
• Add structured data so the content is machine-readable without ambiguity.
• Sign the content with author, role and verifiable credentials. Authorship is a reliability signal for both Google and the models.
• Include frequently asked questions with the answer visible in the HTML. If your FAQs are in an accordion that only loads on click, for a model they don’t exist.

How you measure it if there’s no dashboard

There’s no Search Console for language models. What you can do:

• Regular, documented queries. Asking ChatGPT, Gemini, Perplexity and Claude the same thing every month, and recording whether you appear, in what position and with what description.
• Referral traffic. Visits coming from these platforms can be identified in analytics. It’s the most honest indicator that exists today.
• Brand mentions. Monitoring where you’re named outside your domain, because it’s the fuel of the organic path.

Which path suits you, depending on your situation

Your situation	Where to start
New category that has to be explained before selling	Organic, with paid support to accelerate the first visibility
Buyer in a large company’s committee	Organic. Paid doesn’t reach corporate plans
Product with a fast decision and an individual buyer	Paid, with a capped budget and your own measurement
You want to “show up when they search your category”	Organic. No one can guarantee you a specific query
You need to close the quarter	Neither of the two. It’s a young channel, not a closing lever

How to set up a test in 30 days

1. Define the question you want to win. Not a keyword, a complete question exactly as your buyer would phrase it. “What pedestrian detection system do I need for a forklift in a warehouse” is a question. “Pedestrian detection” is not.
2. Measure your starting point. Before touching anything, ask that question to four different models and save the answers. Without a baseline there’s no story to tell the committee.
3. Publish a piece that truly answers it. Structured, with data, with authorship and with the frequently asked questions visible. One good piece performs better than ten generic ones.
4. Launch an ad group in parallel with a test budget. Well-written context clues, a single destination page and tracking with your own parameters. The goal isn’t to turn a profit, it’s to learn what conversations exist in your category.
5. Review at 30 days and decide. If the ad generates clicks but no sales conversation, the problem is the destination page, not the channel. If the organic path doesn’t move, that’s normal: it’s the one that takes months, and that’s exactly why you have to start it now.

Common mistakes

• Treating the ad manager like a search engine and looking for keywords that don’t exist
• Promising internally a presence in specific conversations that the platform doesn’t guarantee
• Ignoring that the ad doesn’t reach the paid or enterprise plans
• Measuring the organic path with the metrics of the paid one
• Publishing content designed only to convince, with no structure a machine can extract
• Waiting for the channel to mature. The cost of entry rises as more advertisers come in

Conclusion

Appearing in ChatGPT isn’t a single thing. It’s a decision between buying a space adjacent to the answer, with its own rules and a clear limit in B2B, and building the authority needed to be inside the answer, which takes more time and pays off for much longer.

Most industrial companies should start with the second and use the first as controlled learning. But the decision depends on who buys, how they buy and how long you can wait.

At Sheridan we work both paths from the Digital Division and the AI Division. If you want to know whether your category has enough conversation for this to be worthwhile, let’s talk it over before you move any budget.