ChatGPT Ads India: what it changes in your media plan

OpenAI turned on ChatGPT Ads in India on 27 August 2026. Self-serve access to ChatGPT Ads Manager opens on 4 September 2026, which gives most brands about a week to decide whether this belongs in the plan.

What is actually new here

Ads appear below a ChatGPT response when OpenAI determines there is a relevant sponsored product or service. They are clearly labelled as sponsored, visually separated, and they sit alongside the response rather than forming part of the AI answer. You are not buying a mention inside the answer, and no amount of spend changes what the model says.

Against search, the difference is where intent comes from. You are not bidding on a typed query, you are appearing beside a conversation the person is already having, and OpenAI decides when a sponsored product is relevant to it. Against social, the difference is that nobody was interrupted.

The reachable audience is also narrower than everyone on ChatGPT in India. Ads run for logged-in adults on the Free and Go tiers. Plus, Pro, Business, Enterprise and Education users see no ads, and neither do users under 18. Ads are not eligible alongside sensitive or regulated topics, including mental health and politics. India had crossed 100 million weekly ChatGPT users, a figure stated in February 2026, but the ad-eligible share of that has not been published.

Buying a slot below an answer is not the same as being cited inside one. Getting cited by AI assistants is a separate discipline, and money does not buy it.

A Rs 725 daily floor changes who gets to test

Daily campaign budgets start at Rs 725. That is a minimum daily self-serve budget and nothing else should be read into it: no CPM floor, no auction detail and no pricing model beyond that figure has been published.

That removes money as the reason not to test. Rs 725 a day sits inside the rounding error of most D2C media plans. The real cost is operator attention: someone has to build the campaign, watch it, and then be honest about what it did.

Campaigns can be optimised for CPM, CPC or conversion-oriented oCPC objectives. The reported formats are product-feed campaigns, described as always-on, seasonal campaigns, and conversion-optimised cost-per-click campaigns. Targeting controls, creative specifications, character limits, review timelines and auction behaviour have not been published, so plan a first week of finding out rather than a first week of optimising.

Every performance number quoted for this today is invented

Say that plainly to whoever asks for a forecast. The platform went live this morning, self-serve access does not open until 4 September, and more than 50 brands intend to go live this week without any of them being named. Nobody anywhere has a meaningful run of data on this surface.

Any CPM, CPC, ROAS, click-through rate or conversion rate attached to ChatGPT Ads in India right now was either made up or lifted off another platform and relabelled. When someone quotes you one, ask which account it came from and how many days it ran. There is no answer that survives the question.

Size the test by what you can afford to spend on learning, not by a return you cannot estimate. Write down the number you would need to see to keep going, before you have seen any numbers at all.

The measurement problem you inherit on day one

Advertisers get aggregate campaign-performance information such as impressions and clicks. They do not get access to individual conversations, and OpenAI says it does not sell user data to advertisers. That is a defensible privacy position and a hard reporting constraint.

The platform hands you the top of the funnel and nothing after it. Everything past the click comes out of your own stack: strict UTM parameters on every landing URL, a clean session view in analytics, and a post-purchase question asking how the customer heard about you.

The context that made a placement worth buying is the exact thing you cannot inspect. Users can also dismiss an ad, see why it was shown, manage advertising personalisation and clear the data used for advertising. None of that reaches your dashboard either.

At spend that matters, the honest answer is a lift test rather than a platform dashboard, the logic behind a geo holdout in quick commerce. At Rs 725 a day you will not have that volume, so do not claim you are measuring incrementality when you are measuring curiosity.

Where high-context moments plausibly matter

OpenAI frames targeting around high-context moments, the point where a consumer decision is forming. The categories where that plausibly helps an Indian consumer brand have a real research phase and a comparison problem: appliances, skincare where the question is about an ingredient, supplements, baby care, laptops, travel gear. Anything where a shopper asks which one should I get instead of typing a product name into a marketplace search box.

Impulse and replenishment plausibly do less, because that shopper opens a quick commerce app and never opens a chat window. Anything adjacent to mental health is excluded outright.

All of that is a hypothesis drawn from how people already talk to assistants, not a finding. Write it down before you spend, so the test has something it can disconfirm.

What to have ready before 4 September

WPP and Omnicom are the first agency partners on a non-exclusive basis, which matters mainly because you do not need an agency to get access.

Five things are worth doing in the week you have. Get the product feed accurate and in sync, because product-feed campaigns are one of the reported formats and feed readiness is a job of its own. Point the campaign at a page that answers the question a person would have asked, not at a category page. Agree the tracking plan before the first rupee moves. Name one owner. Set a decision date and a kill rule now, while you are still calm.

The four conditions that have to be true before you spend on any channel have not changed for this one, and they are laid out in our guide to how to advertise your website in India.

Nobody knows what this performs like, us included. That is an argument for a small, well-instrumented test with a written hypothesis and a hard stop. It is not an argument for skipping it, and certainly not for moving budget out of something that already works.