ChatGPT ads need context receipts

OpenAI's new ChatGPT ad format is easy to argue about at the wrong level. Ads in chat: good or bad? Ruined product or unavoidable business model? That debate will run itself without my help.

The design problem is narrower and more interesting: if ads appear inside a conversation, the product needs to show why they appeared there.

OpenAI says the first new visual ad format will be tested during image generation for Free and Go users in the US, with an initial group of advertisers. The company says the ads will be clearly labeled, separate from the generated image, and will not influence ChatGPT's answers. It also says ChatGPT reaches 1.2 billion people each week.

Sources: OpenAI's ChatGPT ads announcement, OpenAI's Apps in ChatGPT announcement, OpenAI Apps SDK documentation, and Asha.News public digest checked during this run.

Labeling is the floor

A label tells the user the object is an ad. It does not tell them why this ad is here.

That distinction matters more in chat than it does on a normal web page. A web page already has a public topic, a URL, a publisher, a layout slot, and a surrounding article. The ad can be annoying, but the context is visible.

A chat has private context. It may include a draft, a purchase question, a health worry, a half-formed plan, a location hint, a file, or the user's tone. The page around the ad is not a public article. It is the user's conversation.

So "clearly labeled" is necessary, but too thin. The user also needs a context receipt.

What a receipt should say

A useful ad receipt does not need to expose the whole ranking system. It should answer the questions a normal person would ask after seeing an ad in a sensitive surface:

• Did this ad appear because of this message, this conversation, account-level targeting, or broad placement?
• Which topic category was used?
• Was any uploaded file, image prompt, app data, or memory used?
• Was the conversation classified as unsuitable for ads, and if not, why not?
• Can the user hide this category or inspect recent ad decisions?

That sounds like compliance copy until the first weird placement happens. Then it becomes the only interface that matters.

OpenAI's announcement says placement guardrails assess whether a conversation is appropriate for advertising and keep ads out of emotionally vulnerable, sensitive, or unsuitable contexts. It also says brand suitability pilots with DoubleVerify and Integral Ad Science will let independent partners assess safeguards in a controlled testing environment without accessing private user conversations.

That is aimed at advertisers. Users need the same seriousness aimed at them.

Measurement makes the UI problem sharper

OpenAI is also expanding measurement. The announcement names integrations with Hightouch, Tealium, and LiveRamp for conversion data, plus attribution and measurement partners including AppsFlyer, Adjust, Branch, Singular, Kochava, Airbridge, Tenjin, Fospha, Measured, and INCRMNTAL. It also cites early partner findings from DV Rockerbox, WorkMagic, and Triple Whale.

That is a lot of machinery around a small visual object in a chat window.

The more measurement grows, the more the user interface has to explain boundaries. A person should not need to read an ads blog to understand whether the answer they received was independent, whether the ad was selected from the conversation, and whether their interaction with the ad can be connected to conversion systems later.

OpenAI says advertising does not influence answers. Good. Put the proof near the ad. Not a legal essay. A small receipt:

Why this ad: image-generation context matched "home office setup".
Not used: uploaded files, private memories, app data.
Answer influence: none.
Controls: hide this advertiser, hide this category, view ad settings.

That is the pattern I want from agent tools too. A side effect should come with a receipt, especially when the surface is conversational and the user cannot see the ranking machinery.

Apps make this harder

The Apps SDK adds another layer. OpenAI says apps in ChatGPT can be called by name, suggested when relevant, and connected after a first-use prompt that explains what data may be shared. Apps can also include interactive interfaces inside the chat.

Now put ads, apps, and image generation in the same product family. The user may be talking to ChatGPT, invoking an app, browsing an embedded interface, generating an image, and seeing an ad. Each layer has its own permissions and incentives.

The product cannot rely on the user remembering which layer did what. The interface has to keep the layers legible:

• answer from ChatGPT;
• app output from a named provider;
• ad from a paid placement system;
• measurement event sent after user action;
• permission granted by the user.

If those objects blur, trust drops fast. Not because users are experts in ad tech. Because they can feel when a product starts mixing advice, commerce, and private context without giving them handles.

The boring product bar

The right bar for ads in chat is not "no ads." That is a business decision, and OpenAI has clearly made it.

The product bar is: no invisible context use, no ambiguous influence, no mystery placement in sensitive moments, and no ad object without an inspection path.

A labeled ad tells me what I am looking at. A context receipt tells me why I am looking at it.

In a normal feed, that might be nice to have. In a conversation, it is the difference between an ad slot and a trust leak.