ChatGPT Ads in Japan: Live Since June 2026, and What the First Study Found

In June 2026, the system that shows ads inside the AI assistant "ChatGPT" started in Japan too. It targets the free plan and the low-cost "Go" plan; the paid Plus and Pro plans show no ads. Major Japanese advertising firms—Dentsu Digital, Hakuhodo DY ONE, and CyberAgent—are lined up to support advertisers in placing them.

This article organizes, from both the advertiser's and the user's point of view, "when it started in Japan," "how much advertisers pay," "how the ads have performed so far," "whether ads change the answers," and "whether you can turn the ads off." We also look at how competitor Anthropic (maker of the AI assistant "Claude") has flatly said it will not add ads, and why ads are coming to AI in the first place.

[Update, August 17, 2026] The ads are already running in Japan, and the price has more than halved

Four things have moved since this article was published. First, ad display began in Japan on June 19, 2026, and it is now clear which plans show ads. Second, what advertisers pay has fallen to less than half the original rate. Third, the first empirical study measuring "who is actually being shown ads" was released. Fourth, Anthropic (the U.S. AI company behind the assistant "Claude") has not changed its no-ads stance. We take them in order.

1. Japan started on June 19, 2026, and your plan decides whether you see ads

The "when" in the title now has an answer. According to Impress Watch, ad display in Japan began on June 19, 2026. Ads appear only for signed-in adult users on the free tier and ChatGPT Go. Plus, Pro, Business, Enterprise, and Edu show none. Accounts belonging to under-18s see no ads either, and neither do conversations on sensitive topics such as health, mental health, and politics. Advertisers cannot access chat history or personal information; they receive only aggregate figures such as impressions and clicks.

If you would rather not see ads, moving to a paid plan is the reliable route. Put the other way: as long as you stay on free or Go, there is no setting today that turns ads off. OpenAI sets out what it does and does not do with your data in its own note on testing ads in ChatGPT.

The three Japanese firms behind placement are also confirmed. On June 18, 2026, Dentsu Digital, Hakuhodo DY ONE (the Hakuhodo DY group company running its digital advertising business), and CyberAgent (a company operating internet advertising and media services) announced they were starting pilot operation of ChatGPT ads as domestic launch partners. AdverTimes reported the same. Dentsu Digital described itself as "one of a limited number of companies working directly with OpenAI," an extension of the dentsu Japan–OpenAI strategic partnership signed in February 2026. CyberAgent says it will generate ad assets with its in-house AI text prediction tool "Kiwami Yosoku TD" and run creative covering hundreds to thousands of conversation patterns.

The global picture is clearer too. On June 22, 2026, PPC Land reported that the Japan and South Korea launch was OpenAI's first Asian market and its first non-English-speaking market. At the same time, the UK gained a self-serve Ads Manager, which lets advertisers buy without going through an agency. Japanese advertisers got the same self-serve access at ads.openai.com (Ads Manager beta) around June 28–29, 2026. Then on August 11, 2026, OpenAI confirmed in an official update that ads were live in the UK, Mexico, Brazil, Japan, and South Korea. Together with the U.S. (February 9) and Canada, Australia, and New Zealand (late March), that makes nine countries (gHacks).

2. CPM fell from $60 to $25, and the minimum spend is gone

The "$60 CPM" in this article's original framing is now a historical number. "CPM" is the price per 1,000 ad impressions; "CPC" is the price per click. According to PPC Land, CPM had fallen to about $25 as of April 17, 2026. Actual clearing rates sat in a $25–45 range.

The minimum spend moved the same way. The pilot's $250,000 became $50,000, and on May 5, 2026 the floor was removed entirely when self-serve opened across the U.S. CPC bidding arrived the same day, with a suggested bid of $3–5. There was movement around the edges too: Criteo, the French-founded internet advertising company known for retargeting, cut its minimum spend for ChatGPT campaigns from $50,000 to $10,000 in June 2026. In half a year this went from inventory only a handful of large advertisers could buy to inventory you can try.

A note on what we are not writing. Specific figures circulating in agency blogs as Japanese market averages (a CPM of 773 yen, a CPC of 86–104 yen, CTRs of 0.74–1.34%, and the like), as well as claims of a 2,500-yen minimum daily budget in Japan, could not be traced to a primary source. This article does not use them. Nor has actual data on how advertisers are spending their budgets been published as of August 17, 2026. For market scale, the research firm EMARKETER forecasts that ads inside AI chatbots will bring in less than $1 billion in 2026, roughly 3% of total AI ad spending.

3. The first empirical study: accounts signaling lower income see more ads

OpenAI states that "ads are always clearly labeled and separated from ChatGPT's responses, and do not influence the answers ChatGPT provides" (OpenAI - Our approach to advertising and expanding access). On August 5, 2026, a study appeared that tried to test that claim from the outside: arXiv:2608.05008, "The Beginning of ChatGPT Ads" by Emma Lurie, Ro Encarnación, Sorelle A. Friedler, and Danaé Metaxa.

The method: a 3×3 factorial design crossing three race conditions with three income terciles produced 91 sock-puppet accounts (research accounts created for the study, not real people), with ZIP codes (U.S. postal codes) spoofed via residential proxies, a setup that routes traffic through ordinary home connections to disguise where a request comes from. The team fed 335 prompts daily and collected 3,602 impressions from 191 advertisers.

The findings. First, the initial ad took about 14 days to appear after account creation. Once exposure started, the median exposed account saw ads on 24% of prompts. The most debated result: accounts signaling lower income were significantly more likely to be shown ads (odds ratio 0.98 per additional $1,000 of income, p=0.044). An odds ratio of 0.98 means the likelihood of seeing an ad drops about 2% for every $1,000 of additional income. No significant difference by race was detected, though the paper itself notes this may reflect insufficient statistical power.

By sector, retail and information accounted for 57% of all impressions. The top advertisers were Target (the large U.S. general merchandise retailer), Top10.com (a product and service comparison site), Preply (an online language-tutoring marketplace), and Advance Auto Parts (a U.S. auto parts retailer). By prompt type, product recommendations and health/fitness questions drew the highest ad rates, roughly 14–19%. Meanwhile, the areas OpenAI said it excludes—health conditions, mental health, and politics—showed close to zero ad delivery, so the exclusion policy appears to have been broadly honored.

The limits of this study need stating precisely. The paper does not test whether ads influence the organic text of ChatGPT's answers. What it tests is who saw which ads and how many; it goes no further than describing ads as clearly separated from the answer text. So on the question in this article's own headings—do ads change the answers—no peer-reviewed study or reporting has addressed it directly as of August 17, 2026. What we now know is about skew in delivery, not about neutrality of answers. On how AI answers can bend toward the user, see also Stanford's demonstration that ChatGPT tends to agree with its users.

4. Anthropic's no-ads policy still stands

In "Claude is a space to think" on February 4, 2026, Anthropic stated that "Claude will remain ad-free. Users will not see 'sponsored' links next to their conversations with Claude," and that "Claude's responses will not be influenced by advertisers, nor will they contain third-party product placement that users did not ask for." It says its revenue comes from enterprise contracts and paid subscriptions. As of August 17, 2026, no change or withdrawal of this policy has been confirmed.

One thing worth adding: the same text also says that "if we ever need to revisit this approach, we will explain transparently why." That phrasing leaves room for a future change of direction, and it is not the same as a promise never to run ads. Best not to read it as one.

When did ChatGPT ads start in Japan?

The Japanese rollout began in June 2026. According to ITmedia and other outlets, three firms—Dentsu Digital, Hakuhodo DY ONE, and CyberAgent—announced on June 18, 2026 that they were joining the ad pilot as domestic launch/research partners working directly with OpenAI, and trial ad display began in Japan on June 19. The Nikkei also reported that Dentsu- and Hakuhodo-affiliated firms broker the in-ChatGPT ad display.

Ads appear for users on the unpaid free plan and the low-cost "Go" plan. The paid Plus and Pro plans show no ads. The familiar "use it for free in exchange for seeing ads" model, well known from TV and web services, has now arrived in AI assistants.

Worldwide, OpenAI had begun testing ads in the United States in February 2026. As stated in OpenAI's announcement and reported by CNBC, it first tried the U.S. and signaled an expansion to several countries, including Japan, in May. This Japan launch follows that plan.

Where and how do the ads appear?

Ads appear within the screen where you interact with ChatGPT. OpenAI says it shows ads clearly distinguished from the AI's answers, with a label indicating the sponsor (advertiser) so they are visually separable from a normal reply. Ads relevant to the conversation's context are expected, but details like the exact banner position and size have not been publicly spelled out so far.

As for the ads that actually began appearing in the U.S., Search Engine Land reported they showed up "more prominently (aggressively) than expected," fueling debate over their appearance. How exactly they will look in Japan is likely to take shape as the trial progresses.

Who can advertise, and how much does it cost?

From here, the advertiser's point of view. ChatGPT ads started as something quite expensive that only a limited set of large players could buy. Six months later, the terms are much looser. The period-by-period history is in the update at the top of this article; here we set out the terms as of August 2026.

"CPM" is the rate per 1,000 impressions; "CPC" is the rate per click. According to Adweek, the February 2026 launch terms were roughly $60 CPM with a minimum spend in the hundreds of thousands of dollars—conditions for a small group of large players. Ad holding giants Omnicom, WPP, and Dentsu reportedly lined up brands early. CPM then fell to about $25 by April, and on May 5 click-based pricing (a suggested bid of $3–5 per click) arrived alongside the removal of the minimum spend floor. The self-serve Ads Manager opened across the U.S. at the same time, so the field has clearly widened from the early "large players only." All figures are press-based approximations that vary with exchange rates and conditions.

In Japan, Dentsu Digital, Hakuhodo DY ONE, and CyberAgent act as the windows and support for placing ads.

How have the ads performed since launch?

What advertisers care about most is "does it work." A few months after the February 2026 launch, the honest assessment is "not bad, but hard to judge." Click-through rates (the share of impressions that get clicked) were reported at roughly 0.8–2.3% depending on the vertical—higher than ordinary banner ads—while one early case came in at 0.91%, reportedly about one-seventh of Google Search ads. Views are split.

The reason it is hard to judge is that the measurement tooling is still immature. Campaign reported it is "still finding its feet, with underwhelming early returns," and there was a period when a glitch in the ad-management tool kept campaign data out of reach, making return-on-investment hard to calculate. With no published industry benchmarks, it is hard to tell whether your own results are good or bad.

On the revenue side, however, OpenAI is reportedly doing well. According to CNBC, the ad business reached $100 million in annualized revenue (ARR) in under two months. Even with mixed advertiser sentiment, the sheer pulling power of "the place" that is heavily used ChatGPT is large.

User questions: do ads change the answers, and is my conversation used?

Now the user's point of view. What worries most people is whether the AI's answers themselves get skewed to favor the advertiser. On this, OpenAI says ads do not influence the content of answers. Ads are framed as a separate slot, shown with a sponsor label and kept distinct from the reply.

The handling of conversation content needs a careful read. Per reporting, OpenAI says it does not hand conversation content, chat history, memory, or personal information to advertisers. At the same time, it is said to use signals like the user's conversation content, history, and number of questions "on the ad-serving side" to decide which ads to show. In other words, it is framed as "not selling your content to advertisers" but "using it within OpenAI to target ads." If this concerns you, it is worth reviewing memory and data-use settings.

Can you turn ChatGPT ads off or reduce them?

The surest way to avoid ads is to use the paid Plus or Pro plan. In this system, ads are limited to the free and "Go" plans, and Plus/Pro reportedly show none. It comes down to "pay monthly to avoid ads" or "accept ads in exchange for free use."

Even on free or Go, the precision of the ads (how they are tailored to you) likely depends on your memory and conversation-history settings. If privacy concerns you, review memory use and data-for-training settings. Because this is a trial stage, how ads appear and what adjustment options exist may change going forward.

How rivals move: Anthropic says it will not put ads in Claude

Approaches to advertising in AI assistants differ by company. Notably, Anthropic—maker of "Claude," ChatGPT's biggest rival—has made clear it will not put ads in Claude. The difference in direction also bears on which AI users choose.

Symbolizing this rivalry was an ad video Anthropic released for Claude. OpenAI CEO Sam Altman's reaction to the video—which asks "Can I get a six pack quickly?"—was reported by TechCrunch, drawing attention to the gap between the two companies over advertising.

This is the author's view. "Run ads or not" is more than a revenue question; it is a statement of how each AI intends to build trust with users. OpenAI wants ads to widen its base of free users; Anthropic wants to sell an ad-free experience. Which is right will be decided by users, but the era in which "free AI assistants are supported by ads" has arrived—and it likely will not be reversed.

Why are ads coming to AI, and what's next?

Why did OpenAI turn to ads at all? Behind it lies the enormous cost of developing and running AI. The latest models cost vast sums both to train and to compute answers to user queries, and the more free users there are, the bigger the losses grow. From here is the author's view. When paid members alone cannot cover the costs, ads become a realistic move to "monetize while keeping free users."

OpenAI's finances are something we have covered repeatedly. Looking at the shutdown of the Sora video generator amid heavy losses and the internal conflict over an IPO, you can see why it wants to grow an ad revenue stream quickly. Going forward, expect the barrier to advertising to fall further, opening it to small and midsize advertisers, and the measurement tooling to mature. Meanwhile, debate over answer neutrality and the handling of conversation data is only getting started.

Frequently asked questions (FAQ)

Summary

On June 19, 2026, ChatGPT ads started in Japan. They target signed-in adult users on the free tier and "Go," and do not appear on Plus, Pro, Business, Enterprise, or Edu. Dentsu Digital, Hakuhodo DY ONE, and CyberAgent support placement, and since late June advertisers can also buy self-serve without an agency. Advertiser costs have moved from the early "~$60 per 1,000 impressions, $250,000 minimum" to roughly $25 CPM with no minimum at all.

The performance verdict is not yet settled and measurement is still developing, but OpenAI's ad revenue itself is growing fast. For users, answers are said to be left unchanged by ads, while the handling of conversation data is worth watching. August 2026 brought the first empirical study, which found a skew in delivery: accounts signaling lower income saw more ads. Even that study, though, did not test whether the answer text changes because of ads. If you would rather not see ads, a paid plan is an option. With Anthropic flatly saying it will not put ads in Claude, the question of how to support free AI assistants is where each company's character will show.

Sources

• ▸ITmedia NEWS - Dentsu and Hakuhodo join ChatGPT ads
• ▸Nikkei - OpenAI rolls out in-ChatGPT ads in Japan
• ▸OpenAI - Testing ads in ChatGPT
• ▸OpenAI - Our approach to advertising and expanding access
• ▸Adweek - Omnicom, WPP and Dentsu line up brands for OpenAI pilot
• ▸CNBC - OpenAI ads pilot tops $100M in ARR in under 2 months
• ▸Campaign - Underwhelming early returns for ChatGPT ads
• ▸TechCrunch - Sam Altman got testy over Claude Super Bowl ads
• ▸Dentsu Digital - Starting pilot operation of ChatGPT ads as a domestic launch partner (June 18, 2026)
• ▸AdverTimes - Dentsu Digital, Hakuhodo DY ONE and CyberAgent join ChatGPT ads (June 19, 2026)
• ▸Impress Watch - ChatGPT ads begin in Japan on the free tier and Go (June 19, 2026)
• ▸PPC Land - ChatGPT ads go live in Japan and South Korea, and UK gets self-serve (June 22, 2026)
• ▸PPC Land - ChatGPT ad CPMs drop to $25 as OpenAI races toward global auction
• ▸gHacks - OpenAI expands ChatGPT ads test to UK, Mexico, Brazil, Japan and South Korea (August 13, 2026)
• ▸arXiv:2608.05008 - The Beginning of ChatGPT Ads (August 5, 2026)
• ▸EMARKETER - FAQ on ChatGPT advertising: formats, costs, early strategies
• ▸Anthropic - Claude is a space to think (February 4, 2026)