How to get ready for ads in ChatGPT by the team at Aiso

1. The Economics Behind ChatGPT Ads

1.1 The user base and revenue have exploded

OpenAI has already shared several milestones for ChatGPT usage. The product passed 200 million weekly active users in August 2024, then 400 million weekly users in February 2025, and third party estimates now put it at around 800 million weekly users by early October 2025. (TechCrunch)

On the revenue side, Reuters has reported that OpenAI's annualised revenue run rate rose from about 5.5 billion dollars in December 2024 to 10 billion dollars by June 2025, with later reporting suggesting around 12 billion dollars by July 2025 as adoption of ChatGPT and its API accelerated. (Reuters)

1.2 There is still a big profitability gap

Despite the scale, OpenAI is not yet a profitable business. Financial Times coverage of HSBC research suggests that OpenAI recorded a loss of roughly 5 billion dollars in 2024 and is expected to run large operating losses for years, with HSBC estimating a funding gap of about 207 billion dollars by 2030 because of massive data centre and cloud commitments. (Fortune)

That combination of huge audience, fast growing revenue and very large losses is the core reason an ads and commerce model is almost inevitable.

1.3 Could ads actually fill the gap? A back of the envelope estimation

Internal OpenAI projections reported by The Information and summarised by Search Engine Land show "free user monetisation" starting around 1 billion dollars in 2026 and rising to nearly 25 billion dollars by 2029, about 20 percent of a 125 billion dollar revenue target. (Search Engine Land) At today's scale of roughly 600 million monthly users and around 2.5 billion prompts per day, that would imply ad and commerce revenue per free user in the same ballpark as YouTube and not too far off Instagram. (Statista) By comparison, Google Search generates about 175 billion dollars a year from search ads on roughly 3 trillion queries, which works out to around 5 to 6 cents per search and 56 dollars per thousand queries. (Search Engine Land) If ChatGPT ever approached that level of monetisation per prompt, its current traffic could in theory support 50 billion dollars a year in ad revenue. The problem is that OpenAI's projected cash burn and infrastructure commitments are even larger, with HSBC estimating that OpenAI will need about 207 billion dollars of additional financing cumulatively between now and 2030 to cover data centre and compute costs, which averages out to tens of billions of dollars per year in extra capital over the rest of the decade. (Financial Times)

Looked at on a revenue per user basis, that 25 billion dollars from free users by 2029 would work out to about 10 dollars in annual ad revenue per global ChatGPT user if the product reached roughly 2.5 billion users, compared with our rough estimates of about 40 dollars per year for Google Search, 13 dollars for YouTube and 22 dollars for Instagram globally, and about 20 dollars per future US ChatGPT user versus around 250 dollars for Google Search, 34 dollars for YouTube and 180 dollars for Instagram in the United States. These numbers are approximate and based on public reporting about ad revenue and user bases, but they show how much headroom there still is for ChatGPT to catch up to mature ad platforms.

We derived the comparative ARPU figures using publicly available ad revenue and user base data. For Google Search, we used Alphabet's 2023 global search ad revenue (~$175B) divided by an estimated 4.4B users, and extrapolated a higher U.S. ARPU (~$250) based on U.S. ad revenue concentration. For YouTube, ARPU estimates came from its 2024 global ad revenue ($36.1B) and U.S. share ($8.3B), against global (2.7B) and U.S. (242M) users. Instagram ARPU was calculated from Meta's 2025 projected ad revenue ($67.3B global, $32B U.S.) and user counts (3B global, 172M U.S.). ChatGPT's current ARPU is $0 as it doesn't serve ads. The projected ARPU ($10 global, $20 U.S.) is based on leaked OpenAI targets of $1B–$25B in ad revenue by 2026–2029 and projected user growth to 2B, assuming moderate monetization efficiency relative to competitors.

By 2029, for example, if OpenAI reached around 2.5 billion users and generated about 10 dollars in annual ad and commerce revenue per free user, that would mean roughly 25 billion dollars of yearly ad revenue. If you compare that with a simple round number funding requirement of 30 billion dollars a year implied by HSBC's 207 billion dollar cumulative funding gap, that would cover a little more than 80 percent of the shortfall in that year. In 2026, a more conservative scenario with 1.5 billion users and around 5 dollars per user in annual ad revenue would still bring in about 7.5 billion dollars a year, enough to cover a meaningful share of the required external financing even if the actual funding need is higher in later years. These scenarios are only back of the envelope illustrations, not HSBC's own forecasts, but they show that even modest monetisation at scale could materially change OpenAI's funding picture, without being enough on their own to make the company self financing.

1.4 What Sam Altman has actually said about ads

Sam Altman has been careful when talking about ads. In earlier interviews he described advertising as a "last resort" for OpenAI and said he "kind of hates ads as an aesthetic choice", calling "ads-plus-AI" "uniquely unsettling" in a May 2024 talk at Harvard. (Search Engine Land)

More recent conversations show his language softening. On Stratechery he said "I love Instagram ads" and that he has "actively liked" the products he discovered there, and in other interviews he has talked about the possibility of "some cool ad product" that could be a net positive for users. (Search Engine Land)

Taken together, these statements point to a likely direction: advertising that is useful, relevant and clearly labelled, not generic banners that degrade the chat.

1.5 Hiring and code signals that ads are coming

Beyond interviews, OpenAI is sending concrete signals that it is building ad and commerce capabilities.

On the people side, the company has been hiring leaders with deep ads and commerce backgrounds, including high profile executives from Meta and other consumer platforms. Bloomberg, for example, reported on the hiring of former Meta executive Irina Kofman to lead strategic initiatives, and industry commentary has highlighted leaders such as Fidji Simo, known for building Facebook's mobile ad business and scaling Instacart's ads. We estimate that OpenAI has already committed hundreds of millions, plausibly around a billion dollars, of multi year total compensation to ex Meta monetisation and ads talent, which underlines how serious this push is. (MediaPost)

On the product side, independent reverse engineers have found ad related strings and configuration hooks in recent ChatGPT Android beta builds, including references to search ads and targeting. These findings were shared in public threads and later picked up by media outlets as an early sign that OpenAI is preparing an ad system inside ChatGPT. (Dataconomy)

Wrap up pieces on when ChatGPT might get ads pull these code snippets together with job postings for a "marketing platform" to enable campaigns and real time attribution, reinforcing the view that OpenAI is building the underlying infrastructure for ads. (IndexLab)

2. What ads in ChatGPT will probably look like

2.1 Who will get access first, and at what price?

A natural question for marketers is whether OpenAI ads will be open to everyone from day one or whether the first wave will be a small private beta for very large partners.

If you look at how OpenAI usually rolls out major features, the pattern is clear. GPT-4 launched with deep integrations at companies like Duolingo, Stripe and Khan Academy before broader access. (Mint) ChatGPT plugins initially shipped with a limited set of large partners such as Expedia, Instacart, Shopify, Klarna and others before opening up a wider developer ecosystem. (OpenAI) Sora followed the same playbook, with early access for a few hundred artists, designers and filmmakers before being made available more widely. (OpenAI)

Commerce is already following this pattern. Instant Checkout in ChatGPT first launched with Etsy and Shopify, and more recently Walmart joined through a dedicated partnership, with documentation and commentary describing a broader Agentic Commerce Protocol that other platforms can adopt later. (OpenAI)

Based on this history, it is reasonable to expect that the very first ad and sponsorship experiments will run with a short list of large retailers, marketplaces and brands that are already integrated into ChatGPT commerce flows. Self serve access for the long tail of advertisers is likely to come later, once OpenAI is confident that the formats work and the safety and measurement stack is in place.

In fact, on December 3rd, 2025, the first appearance of a suggestion to continue on the large American retailer Target appeared. This early example shows how OpenAI is testing commerce integrations through app suggestions that appear below ChatGPT responses, allowing brands to be promoted without disrupting the core chat experience.

View tweet on X/Twitter →

Pricing is likely to follow a similar pattern to other new ad platforms. In auction based systems, prices start low when there are few advertisers and increase as more bidders enter the auction and competition intensifies. (Google Ads Help) Mature platforms like Google Search already show high average costs per click and steady CPC inflation. Recent benchmark work puts the overall average Google Search CPC at around 8 dollars, with double digit percentage increases year on year for many industries. (WordStream)

Newer networks, on the other hand, are often much cheaper in their early years because supply of impressions grows faster than demand from advertisers. TikTok is a recent example, with multiple studies and agency reports showing that TikTok CPMs and CPCs started significantly below Facebook and Instagram, which created arbitrage opportunities for early adopters. (Tlinky)

Given OpenAI's need to learn quickly and attract budgets from incumbent channels, the most realistic expectation is that early ChatGPT ad inventory will be priced aggressively, likely well below fully mature Google Search and Meta auctions on a cost per click or cost per acquisition basis. Of course this presents a huge opportunity for the companies who will get ready for ads early. We expect that even after broadening of access, ads will be significantly cheaper for a comparable number of impressions, clicks or conversations than similar ads at Google.

2.2 Who is the ideal type of business for ChatGPT ads?

To understand who will benefit most from early ChatGPT ads, you need to look at two things at the same time: who actually uses ChatGPT today, and how competitive the auction is likely to be in the first few years.

On the audience side, ChatGPT has become both very broad and very skewed toward certain groups. Exploding Topics estimates that more than 45% of ChatGPT users are under the age of 25 and that only about 15% are in the United States, with the rest spread across Europe, Asia and Latin America. (Exploding Topics) OpenAI's own research and third party surveys show that usage is strongest among students, knowledge workers, freelancers and tech forward professionals, with very high penetration in categories like software, consulting, finance and marketing. (OpenAI)

In terms of what people actually do with ChatGPT, OpenAI's economic research paper and other analyses find that around 75–80% of usage falls into three buckets: practical guidance, information seeking and writing or editing. (NBER) Aiso's analysis of more than 2.1 million real ChatGPT conversations adds two new intent categories on top of traditional search – Productive and Creative – and shows strong volumes of transactional queries in areas like travel, software tools, online education, e commerce and personal finance. (Aiso)

Put simply, ChatGPT's early audience is young, global, digital first and heavily concentrated in high intent research and decision making tasks. That means the businesses that stand to gain the most from early ads are those that:

• sell products or services that can be researched and purchased fully online (software, education, e commerce, travel, financial services, healthcare adjacent information products)
• rely on considered decisions with lots of questions, comparisons and trade offs
• already see meaningful organic traffic or brand searches coming from AI tools today

Brands that are purely offline, mass market and price led will still benefit eventually, but the near term upside is much higher for categories where the buyer journey already looks like a long ChatGPT conversation.

Competitiveness is the other half of the equation. Even once OpenAI opens beyond a small beta group, we expect large enterprises to move comparatively slowly: they have more stakeholders and compliance reviews, and many rely on traditional SEO or media agencies whose incentives are tied to existing channels and reporting stacks. Historically, new ad platforms such as early Facebook and TikTok were under priced in their first few years because direct to consumer brands and smaller challengers moved faster than big incumbents, and it took time for agencies and procurement teams to catch up. (Tlinky)

Taken together, the ideal early adopter for ChatGPT ads is a digitally native, mid sized or challenger brand with:

• a clear online funnel and the ability to ship landing pages quickly
• a product that benefits from explanation, comparison and personalised advice
• enough budget and risk appetite to test a new channel before their largest competitors do

Non profits and mission driven organisations may also find ChatGPT ads attractive, especially for education, fundraising and cause awareness campaigns, because the audience skews young, curious and research oriented. The key is whether your ideal user is already the kind of person who asks ChatGPT for help when they are deciding what to buy, where to learn or which cause to support. (Aiso)

2.3 Protecting the core chat experience

OpenAI leaders are also aware of the risk. In a recent episode of The Verge's Decoder podcast, the head of ChatGPT, Nick Turley, said that ads are "not off the table" as a path to monetisation, but that any ad experience would need to be implemented in a way that maintains user trust and transparency. (The Verge)

That makes it unlikely that the first wave of ads will be interruptive units inside every answer. A more plausible path is to start with formats that feel like native recommendations and clear commerce flows, in parts of the product that already look and feel like discovery surfaces.

2.4 The experimental surfaces: Shopping Research, Pulse, Instant Checkout and ChatGPT apps

OpenAI has quietly been building those surfaces.

• Shopping Research: a feature that turns a shopping related question into a personalised buyer guide that compares products, explains trade offs and links to merchants. (OpenAI)
• Pulse: a feed for Pro users that delivers personalised digests and buyer guides based on their conversations, with Shopping Research guides now integrated so they can appear as part of an ongoing update stream. (OpenAI)
• Instant Checkout: an integration for merchants that lets users buy some products directly inside ChatGPT, with a secure checkout and no need to click out to a separate website. (OpenAI Help)
• ChatGPT apps and app suggestions: third party integrations that appear as suggestions below ChatGPT responses when relevant to the conversation. These apps can help users complete tasks directly within ChatGPT, such as design tools like Figma appearing when users ask about design work, or productivity apps for specific workflows. App suggestions appear non intrusively below the main response, allowing users to continue their conversation or connect to external tools as needed.

The Target ChatGPT app suggestion that appeared on December 3rd, 2025 represents an early experiment in using app suggestions for advertising purposes. While app suggestions are typically functional tools (like Figma for design tasks), the Target example shows how retailers could use this surface to nudge users toward commerce opportunities. In this case, the suggestion appeared below a response about Windows BitLocker, demonstrating how app suggestions can surface even when the connection to the query is indirect. (X/Twitter) What is an ad if not a nudge? This format allows brands to promote themselves without disrupting the core chat experience, creating a non intrusive way to surface commerce opportunities.

OpenAI's own documentation for Shopping with ChatGPT Search explains that product cards in search results link out to multiple merchants, ranked on factors such as price, availability and whether Instant Checkout is enabled. (OpenAI Help)

When you combine Shopping Research, Pulse, Instant Checkout and ChatGPT apps, you get a clear blueprint: conversational shopping journeys that can include both organic recommendations and paid placements, plus a way for OpenAI to earn commissions on purchases. (OpenAI)

2.5 From commerce flows to ad products

Reporting on "Commerce in ChatGPT" suggests that OpenAI is already working with partners like Shopify on a payment checkout system within ChatGPT so that it can earn commissions on purchases. (VentureBeat)

Once you have:

• a shopping query surface
• a feed that can remix those guides
• an in chat checkout
• and a history aware memory system
• suggestion of ChatGPT apps for design tools like Figma

it is not a big step to add:

• clearly labelled sponsored slots inside Shopping Research guides
• commission based recommendations in Pulse
• or performance based placements in Instant Checkout journeys
• suggestion of ChatGPT retailer apps for purchasing intent

If OpenAI does this, it can keep the core chat experience mostly unchanged while building a serious ads and commerce business around it.

2.6 Will history and memory be part of ChatGPT ads?

Sam Altman has also been open about what he thinks will make ChatGPT different from search engines when it comes to relevance: history.

He has described a "platonic ideal" of a small reasoning model with a trillion token context window that you "put your whole life into", and said that his goal is for ChatGPT to remember your whole life, as reported from a Sequoia hosted event and covered by TechCrunch, Windows Central and others. (TechCrunch)

OpenAI has already shipped this idea in product form. ChatGPT now includes a persistent memory feature that stores important details across sessions when the user opts in, and similar ideas are being extended into business products. Official documentation explains that memory allows ChatGPT to remember information and personalise responses over time. (OpenAI)

If you put those pieces together, you get a clear hypothesis: ChatGPT will lean on long term conversational history to deliver more relevant, personalised experiences than what is possible with traditional search alone, which would also make it a powerful signal for future ad targeting and measurement.

2.7 Video and group chat as testbeds

Video and group chat are likely to be testing grounds, but for a slightly different reason than Shopping Research or Pulse.

It is true that, today, Sora style video experiences and group chats are weaker on personalisation than the core one to one ChatGPT experience with memory. A shopping conversation that spans weeks and dozens of prompts will always carry more granular intent signals than a single Sora clip or a casual group thread. That is exactly why these surfaces are attractive as sandboxes: if an early ad experiment feels clumsy or off brand, it damages a secondary experience rather than the core product people rely on every day.

On video, OpenAI has introduced advanced video models such as Sora and has announced creator tools and programmes around AI generated clips. These products are still in a clearly experimental phase and are used by a much smaller subset of ChatGPT users than the main chat interface. Ad industry writers have already speculated that short video formats will be an obvious testbed once OpenAI starts experimenting with ads, even though there is no dedicated video ad product today. (OpenAI)

On group chat, OpenAI's own description makes it clear that group chats in ChatGPT are separate from individual conversations and that personal memory is never shared with other people in the chat. (OpenAI) This separation reduces the personalisation signal but also reduces the reputational risk: OpenAI can try new interaction patterns, sponsorships or lightweight promotions in a context where users already expect a more experimental, collaborative feel, without changing the classic one to one assistant that carries most of the long term history.

This fits with OpenAI's broader "fail fast" pattern. Plugins, early browsing modes and other features were launched, iterated on and sometimes shut down or replaced entirely when they did not work, with OpenAI explicitly framing them as experiments. (OpenAI Help) We should expect ad formats to follow a similar path: start in lower risk surfaces like video and group chat, then gradually move closer to the highly personalised core chat experience once the team is confident the formats are useful and trusted.

3. How to prepare for AI ads in ChatGPT

3.1 Explain who your product is for

If we assume that future ad formats will be highly personalised, then the main job for marketers is to make their differentiation machine readable.

Altman's comments about wanting ChatGPT to remember your whole life, combined with OpenAI's investment in memory features, are widely interpreted as building a rich first party profile for each user that can power more relevant recommendations. (TechCrunch)

That means you should be crystal clear about who your product is for, which jobs it solves and which trade offs it makes, both in plain language and in structured data.

3.2 Make personalisation and classification machine friendly

OpenAI's description of Shopping Research and Pulse shows that recommendations can take into account your needs, preferences, budget and previous interactions with ChatGPT. (OpenAI)

Reviews describe conversational flows where ChatGPT asks clarifying questions about budget, use cases and must have features, then compares options across those dimensions instead of just listing links. In other words, the system is building a classification space for products in real time.

AI search optimisation playbooks now recommend including structured data for attributes such as audience, use case and context so that AI systems can better match content and products to personalised queries, and many AISO guides stress describing who something is for and in what context as part of schema and on page content. (Aiso)

If you start encoding these attributes in your content and schema today, you are essentially following the same pattern that current AI search features and legacy ad platforms already use, where structured fields like audience, budget, use case and key attributes power both organic recommendations and paid targeting, so it is a low regret way to make it easier for future ad systems to discover and match you accurately even if the exact implementation changes. (Google Merchant Center)

3.3 Learn from the history of conversations

Finally, there is the question of history.

Altman's discussion of a tiny reasoning model with a trillion token context window and AI agents that can act across your whole life appears in multiple summaries of his Sequoia remarks, and analyses of GPT 5 and beyond emphasise memory and personalisation as the next frontier. (TechCrunch)

Aiso's public materials describe a proprietary dataset of more than 2.1 million real ChatGPT conversations that it structures to extract intent patterns and conversational journeys at scale. (Startup Seeker)

We expect a whole new category of research to emerge around understanding conversational history: which sequences of prompts lead to purchase, which objections matter, how people change their minds over time.

Brands that learn from this kind of data will be better prepared to brief future AI ad systems and to interpret whatever reporting OpenAI eventually provides.

Conclusion: use 2025 to build your advantage

ChatGPT is on track to remain one of the largest consumer surfaces on the internet, with around 800 million weekly active users and an annualised revenue run rate already above 10 billion dollars, while still running multi billion dollar annual losses. (Reuters) That economic reality makes an ads and commerce model almost inevitable, but the combination of Sam Altman's public statements and OpenAI's product choices so far suggests that the company will prioritise relevance, safety and user trust instead of interruptive display inventory. (Search Engine Land)

For marketers, this creates a two stage opportunity. First, organic AI search: being mentioned in ChatGPT answers, linked as a source and measured with proper server side observability. Early data from Aiso's analysis of more than 2.1 million real ChatGPT conversations shows that brands can materially increase the frequency and quality of mentions by aligning their content with real prompts, improving AI crawlability and fixing schema markup issues. (Aiso) Second, future AI ads: when OpenAI introduces paid placements on top of this, the winners will be those who have already done the hard work of clarifying positioning, audiences and use cases in a way that is machine readable and compatible with high levels of personalisation.

Our recommendation is therefore simple. Use 2025 as the year to: 1) audit how often and in what context your brand appears in ChatGPT today; 2) fix the basics such as page speed, crawlability, structured data and third party distribution; and 3) start instrumenting your stack to measure ChatGPT traffic and conversions based on real logs, not just estimates. (Aiso) Teams that treat this as a compound learning process now will be best placed when ads arrive in ChatGPT at scale.