OpenAI to test ads during ChatGPT image generation in the US; its free no-ads option has no image tools

262 days from the bottom of the answer to the wait for an image

On 16 January 2026 OpenAI said it planned to test ads "at the bottom of answers" in ChatGPT. On 5 October 2026, 262 days later (derived), it announced a visual ad format that it will first test "during image generation in ChatGPT", in the US, "later this month", with "an initial group of advertisers". It says: "Ads will be clearly labeled, and remain separate from the image being created."

Read beside OpenAI's help-centre page on ads, the placement has a second edge. That page offers Free users a no-ads option and lists its cost: "fewer messages per day and no access to tools like images". Image generation is the tool the new format sits inside. On that wording, a Free user who wants images without paying has to accept ads (inference). The page, read at 16:44 BST on 5 October 2026 and marked as updated seven days earlier, says ads appear "below the end of a response". It does not mention the image format.

What that does not establish.

• That the image prompt is used to choose the ad. The announcement does not say. The help page says selection considers "the context and intent of your current conversation"; whether this format reads the prompt is left open.
• How many people will see it. "An initial group of advertisers" is the only size given. The 1.2 billion weekly users OpenAI cites are ChatGPT as a whole, not people on Free or Go in a country with ads.
• A UK test. The test is announced for the US. That does not put the UK outside the programme: OpenAI's release notes list "Rolling out ads in the UK" for Free and Go in the entry group dated 4 June 2026, 123 days before this announcement (derived).

Stated and not stated

OpenAI's announcement of 5 October 2026, ads help page, ad policies and Europe privacy policy, read between 16:42 and 17:00 BST on 5 October 2026

1. Question

   Where the ad appears

   Stated by OpenAI

   During image generation, labelled, separate from the image. Elsewhere, below the end of a response.

   Not stated

   Screen position, how long it stays, whether it outlasts the wait.

2. Question

   Who sees it

   Stated by OpenAI

   Free and Go may see ads. Plus, Pro, Business, Enterprise and Edu will not. The test is in the US.

   Not stated

   Whether Free, Go or both join the test; signed-out use.

3. Question

   Image prompt used to pick the ad

   Stated by OpenAI

   Selection uses the current conversation and, with personalisation on, past chats and memory.

   Not stated

   Whether this format uses the image prompt or an uploaded photo.

4. Question

   Personalisation off

   Stated by OpenAI

   Ads continue, based on the current chat thread.

   Not stated

   Any switch that stops current-chat targeting.

5. Question

   What advertisers receive

   Stated by OpenAI

   Aggregated views and clicks. Never chats, history, memories, name, email, precise location or IP address.

   Not stated

   What an advertiser learns once a person clicks through.

6. Question

   Minors

   Stated by OpenAI

   No ads for accounts identified as under 18, by stated age and age prediction "where available".

   Not stated

   Prediction accuracy, UK availability, what a signed-out child sees.

7. Question

   Sensitive topics

   Stated by OpenAI

   No ads near personal health, mental health or politics. Ad policies v1.1 (April 2026): medical, legal and financial advice contexts no longer blocked by default.

   Not stated

   How exclusions apply to an image prompt; the policy is written about conversations.

8. Question

   UK

   Stated by OpenAI

   Release note dated 4 June: ads rolling out for Free and Go. Testing page, 11 August: "now launched".

   Not stated

   A UK date for the visual format.

9. Question

   Independent checks

   Stated by OpenAI

   Brand-suitability pilots with DoubleVerify and Integral Ad Science are being developed, without access to private conversations.

   Not stated

   Any outside check of answer independence or child protections.

Question	Stated by OpenAI	Not stated
Where the ad appears	During image generation, labelled, separate from the image. Elsewhere, below the end of a response.	Screen position, how long it stays, whether it outlasts the wait.
Who sees it	Free and Go may see ads. Plus, Pro, Business, Enterprise and Edu will not. The test is in the US.	Whether Free, Go or both join the test; signed-out use.
Image prompt used to pick the ad	Selection uses the current conversation and, with personalisation on, past chats and memory.	Whether this format uses the image prompt or an uploaded photo.
Personalisation off	Ads continue, based on the current chat thread.	Any switch that stops current-chat targeting.
What advertisers receive	Aggregated views and clicks. Never chats, history, memories, name, email, precise location or IP address.	What an advertiser learns once a person clicks through.
Minors	No ads for accounts identified as under 18, by stated age and age prediction "where available".	Prediction accuracy, UK availability, what a signed-out child sees.
Sensitive topics	No ads near personal health, mental health or politics. Ad policies v1.1 (April 2026): medical, legal and financial advice contexts no longer blocked by default.	How exclusions apply to an image prompt; the policy is written about conversations.
UK	Release note dated 4 June: ads rolling out for Free and Go. Testing page, 11 August: "now launched".	A UK date for the visual format.
Independent checks	Brand-suitability pilots with DoubleVerify and Integral Ad Science are being developed, without access to private conversations.	Any outside check of answer independence or child protections.

Friendly names are claims about design

Three phrases carry the comfort in OpenAI's text. None is a control a user can inspect.

Conversation privacy means conversations stay private from advertisers. It concerns who receives the chat, not whether OpenAI reads it. The help page says the ads system uses the current conversation, and that with personalisation off a user will "still see ads based on the context of your current chat thread".

Advertising does not influence the answers is a statement about design. The help page says ads run on separate systems from the chat model; a user cannot see that. The independent evaluations announced on 5 October are brand-suitability pilots for advertisers, still in development. They are not described as an audit of answer independence.

Contextual is the shorthand, not OpenAI's word. The ICO says contextual advertising more readily complies because "personal data is not used to determine what advert a user sees". That reasoning was written for a page a publisher wrote. A conversation the user wrote can hold a name, a place or a photograph, and the guidance does not address that case (inference).

The controls that exist are real: under Settings, Ads controls, a user can switch off Personalize ads and Past chats and memory, see ads history, delete ads data (up to 30 days to remove) and open About this ad. Temporary Chats "will not show ads"; whether that holds for image generation is not stated. None of the controls switches off the current-chat signal.

A prompt you wrote, a screen you are watching

An ad during image generation lands while a person waits on something they typed. OpenAI's Images page says generation "may take a few minutes". Its 8 September release note says some users already see "animated dots and a Snake game" during the wait. OpenAI's illustration of the new format (not a live screenshot) shows an "Ad" pill, an advertiser name, a headline and a Learn more button under a placeholder marked 62 per cent. The help pages say ads are labelled "sponsored"; which word a UK user sees is not stated. The ASA's influencer guidance, not binding, advises against "sponsored" as "open to varied interpretation" and calls "Ad" very likely acceptable.

Personal data. OpenAI's Europe privacy policy (updated 24 August 2026, the version that covers UK users) counts prompts and uploaded images as User Content. For personalised ads it lists User Content among the data used, on the basis of consent "when we ask for it". For generic ads it lists "the context of your current ChatGPT chat, location and time of day", on legitimate interests. It names no exception for image prompts and does not say whether this format reads them.

Children. OpenAI shows no ads to accounts identified as under 18, using age information and age prediction "where available"; the prediction signals include "the general topics a person discusses". Signed-out ads are "appropriate for all audiences". Teens can create images: the Images page describes a teen experience with parental controls over image creation. The ICO's Children's code says profiling children for advertising should be off by default, and its May 2026 advice to government calls legitimate interests for advertising to children "likely to be difficult". None of the sources says what happens when ChatGPT cannot tell that a child is the user.

The UK position: ads are already here, the format is not

The 4 June release note says OpenAI was "beginning to roll out" ads in the UK. An update to its "Testing ads in ChatGPT" page says that on 11 August the programme had "now launched" there. The Ads Manager availability page lists the United Kingdom as available, but that concerns where advertisers are based, not where viewers are.

Dates from OpenAI's own pages; the gaps are derived day counts, checked on 5 October 2026

1. Date (2026)

   16 January

   What OpenAI's page says

   Plan to test ads at the bottom of answers for Free and Go in the US

   Days after row above

   start

2. Date (2026)

   9 February

   What OpenAI's page says

   US test begins (release notes)

   Days after row above

   24

3. Date (2026)

   4 June

   What OpenAI's page says

   Release notes: beginning to roll out ads in the UK for Free and Go

   Days after row above

   115

4. Date (2026)

   11 August

   What OpenAI's page says

   Ads "now launched" in the UK, Mexico, Brazil, Japan and South Korea

   Days after row above

   68

5. Date (2026)

   16 September

   What OpenAI's page says

   Sponsored Agents, a chat with a business's agent after an ad, tested with select US advertisers

   Days after row above

   36

6. Date (2026)

   5 October

   What OpenAI's page says

   Visual format to be tested during image generation, in the US, later in October

   Days after row above

   19

Date (2026)	What OpenAI's page says	Days after row above
16 January	Plan to test ads at the bottom of answers for Free and Go in the US	start
9 February	US test begins (release notes)	24
4 June	Release notes: beginning to roll out ads in the UK for Free and Go	115
11 August	Ads "now launched" in the UK, Mexico, Brazil, Japan and South Korea	68
16 September	Sponsored Agents, a chat with a business's agent after an ad, tested with select US advertisers	36
5 October	Visual format to be tested during image generation, in the US, later in October	19

What the UK sources say. The ICO's online-advertising guidance, finalised in April 2026, says using storage and access technologies for online advertising "requires consent", including for ad selection, delivery and measurement. That PECR rule concerns reading or storing information on a user's device. OpenAI does not say whether the in-chat ad unit does either, so whether it applies is not established. OpenAI names OpenAI OpCo, LLC, in San Francisco, as controller for residents outside the EEA and Switzerland, which includes the UK. The ICO's advice says special category data in advertising needs explicit consent. Its consent-or-pay guidance (January 2025, under review after the Data (Use and Access) Act) asks whether the core service is "broadly" the same across options. Whether image generation is core or an extra is what the Free no-ads option raises; no ICO source read applies that test to OpenAI.

The CAP Code covers "online advertisements in paid-for space" and paid-for communications aimed at people in the UK. Rule 2.1 requires marketing to be "obviously identifiable as such"; rule 2.3 requires it to "make clear their commercial intent, if that is not apparent from the context", and CAP says it reflects a prohibited practice in Schedule 20 of the Digital Markets, Competition and Consumers Act 2024. The CMA's guidance (CMA207, 18 November 2025) calls failing to identify commercial intent a misleading omission. The ASA says the Codes have no AI-specific rules and apply "regardless of how content is generated, edited, or targeted". None of these sources addresses an ad slot inside a chatbot or an image screen; none rules on OpenAI.

Which plans carry ads depends on which page you read

Six OpenAI pages give different plan lists. The ads help page names five plans as ad-free: Plus, Pro, Business, Enterprise and Edu. The Europe post of 18 August names three. The advertiser-facing page says "Plus, Pro, or any Business plan". The Go plan page says OpenAI "may start testing ads in ChatGPT Go in the future", while the help page lists Go as ad-eligible.

No plan changes sides; no single page is the contract. One more line matters to anyone paying: the help page says Ads-Free is a Free-plan option, so a paying Go subscriber who wants no ads must switch to Free, and lose image tools, or upgrade.

For UK organisations whose staff use free or low-tier accounts

Staff using personal Free or Go accounts for work are on the ad-supported side of OpenAI's own plan boundary. On the other side, OpenAI's data-controls page says Business, Enterprise and Edu content is not used to train its models by default, and its enterprise privacy page (updated 8 January 2026) offers a data processing addendum for Business, Enterprise and the API. The consumer privacy policy says that when an account uses a work email address, OpenAI "may share the fact that you have an account" with the employer: a possible inventory source, not a guarantee.

Earlier briefings show the same data from other angles: a Featured ad blocker whose server sent programs aimed at AI chats, and 13,000 public screenshots that AI agents put in repositories, a reminder that images carry data nobody typed.

Method, and commercial interest

This briefing follows OpenAI's pages, not the press report. BleepingComputer's account (5 October, 06:28 US Eastern) matches OpenAI's announcement and is used only as a pointer. OpenAI sells these ads and reports its own results; its 31 August post claims a 1 billion dollar annualised run rate, not checked here. Nothing here says OpenAI has broken a rule or a law. Its comfort phrases describe design, and the controls it ships are narrower than the phrases.

The question the announcement cannot answer

When a member of staff types an image prompt into a free account and an ad appears while the image renders, which of your records shows that your organisation decided an ad-supported tool could read that prompt?

Key facts