ChatGPT Ads Arrive in India.

01What ChatGPT Ads actually are

The unit is a sponsored card that appears below a finished answer. Not inside it, not woven through it, not shaping it. ChatGPT answers the question, the answer ends, and beneath it sits a labelled commercial placement carrying an advertiser name, a logo, a headline, a short description, an optional image and a call to action.

That description sounds mundane. The position is the entire product. An ad that appears after a question has already been answered is reaching a person at a different moment than a search ad does - not at the point of query, but at the point where the query has been resolved and the next step is being considered.

What it is not

Three comparisons are being made in the market this week, and all three mislead.

It is not a search ad. A search ad is bought against a string a user typed. Here there is no string to bid on. It is not display, because it is not bought against an audience segment or a placement list. And it is not native or an affiliate placement, because it cannot be woven into the editorial content - the separation between the answer and the ad is the product's central design constraint, not a formatting preference.

The closest honest analogy is a shopping unit that appears after a considered recommendation, bought without the ability to specify what recommendation it appears after. That is genuinely new, and nobody has a playbook for it.

Who sees it

Logged-in users, aged 18 or over, on the Free and Go tiers only. This is the detail most widely reported wrong this week, so it is worth stating precisely.

Beyond the tier gate there is a second set of exclusions, and it does more commercial work than it first appears. These are surfaces and contexts where no ad will serve regardless of what an advertiser is willing to pay.

The sentence most coverage got wrong

Nearly every write-up this week carried a version of: advertisers receive aggregate impressions and clicks only, with no access to conversations or personal data.

That is true about advertisers and incomplete about the system. Advertisers genuinely never see a conversation. But conversations are exactly what rank the ads. The context of what a person is discussing is the primary targeting signal, and with personalisation enabled, past chats and stored memory feed into it too.

Both things are true at once, and an agency that cannot hold them together in a single sentence will get caught out in the first client meeting where someone asks a sharp question about privacy. The distinction is between data leaving the system, which does not happen, and data being used inside it, which is the whole mechanism.

02How the machine works

An ad is selected on two things: how relevant it is expected to be to the conversation in front of it, and what the advertiser is willing to pay. Relevance is assessed against the live context - what the person is asking about, what they have been discussing, what the ad's own copy and landing page are about.

What flows the other way is very thin. The advertiser receives counts. Not conversations, not queries, not the text that triggered the placement, not who the person is. The asymmetry between what goes in and what comes back is the single most important structural fact about this channel, and it is the thing to draw on a whiteboard when a client asks how it works.

The two absences that define the channel

Set aside everything else in the specification. Two capabilities that Indian performance marketers use every working day are simply not present.

There is no keyword matching. No exact match, no phrase match, no broad match, no negative keyword list, no search-term report to mine. Instead there are “context hints” - plain-language descriptions of the situations in which an advertiser wants to appear. You are describing a circumstance, not bidding on a string.

There is no demographic targeting. No age brackets, no gender, no income proxies, no interest categories. Geography, platform and a custom audience list uploaded from your own data is the entire available surface.

Everything that follows in this article - how the work changes, which clients it suits, what an agency can honestly charge for - descends from those two absences.

What it costs, so far

India has no published benchmarks. None. Anyone quoting you an expected CPM or CPC for this channel in India today is extrapolating, and you should ask them from what.

What does exist is the US trajectory since February, and it points in one direction. That direction is the more useful input for a budget conversation than a fabricated India number would be.

The policy gate

Before any of this matters, an advertiser has to get through review. This is where a meaningful number of Indian categories will discover they cannot participate, and it is worth checking before you build a pitch around a client who turns out to be ineligible.

Two absences define this channel: no keyword, no audience segment. Both are the primitives Indian performance marketing was built on, and both are the layers agencies staffed most heavily.

03The Indian digital marketing scene as it stands

India's advertising market reaches roughly ₹1,74,605 crore in 2026 on the Pitch Madison count, with digital at about 60% of it last year and heading toward 64% this year. Dentsu calls India the fastest-growing ad market in the world this year. Those are the headline numbers, and they are the last comfortable thing in this section.

Below the headline, the Indian ad-spend evidence base is a mess, and an article that pretended otherwise would be worth less to you. The two most-cited sources for what sits inside digital - WPP Media's This Year Next Year and the FICCI-EY Media & Entertainment report - use different bases, different definitions and different totals. They are not reconcilable. Both are shown below, separately, because averaging them would manufacture a precision nobody has.

Who captures it

Concentration is the defining feature. Two companies take roughly two-thirds of Indian digital advertising, and the fastest-growing challenger is not a media company at all.

Where the new money is

The growth is not concentrated where agencies are. Three pools have expanded fast, and Indian agencies are structurally absent from all three - a point made at length in Growth Without Margin and worth restating here, because ChatGPT Ads at ₹725 a day lands squarely in the segment agencies do not currently serve.

V12. MSME and quick-commerce figures: Pitch Madison Advertising Report 2026. Creator-market figure: Zefmo Media, May 2026 - a single vendor report with no disclosed methodology; treat as directional.

What the work is actually made of

Here is the part that decides whether the rest of this article matters to you.

Strip an Indian performance-marketing engagement down to its load-bearing structure and you find two primitives underneath almost everything: the keyword and the audience segment. Campaign architecture is built on them. Job descriptions are written around them. Optimisation cycles consist largely of operating them - adding negatives, splitting ad groups, refining segments, building lookalikes. Reporting is organised by them. A significant share of what an Indian agency bills for is the competent, repeated operation of those two objects.

ChatGPT Ads has neither.

Remove the keyword and the audience segment, and what is left of a campaign is the two ends: deciding what to do, and finding out whether it worked. Those were always the parts an agency was really for. They were just bundled inside the middle.

04ChatGPT Ads enter India

On 27 August, sponsored placements went live for Indian users on the Free and Go tiers. More than fifty brands are running this week, bought through WPP and Omnicom as OpenAI's first Indian agency partners on an explicitly non-exclusive basis. From 4 September, self-serve access opens to businesses of all sizes through the ChatGPT Ads Manager, with daily budgets starting at ₹725.

Dave Dugan, OpenAI's VP and Head of Global Ads Solutions, framed it as: “India is one of the most dynamic markets in the world for how people use ChatGPT to learn, build, and discover.”

The audience behind that sentence is the reason this matters. India is OpenAI's second-largest market, and the base was assembled deliberately.

Why now

Commercial pressure, and it is not subtle. OpenAI filed confidentially for an IPO in June. Second-quarter revenue reached $6.7 billion against $5.7 billion in the first quarter - and operating losses widened over the same period from $9.3 billion to $12.3 billion. The US ad pilot reportedly reached over $100 million annualised within six weeks of launch.

An agency planning around this channel should assume OpenAI is motivated to grow ad revenue quickly, and should read the pace of the rollout in that light rather than as a product-quality signal.

05The details that matter

The framing that has travelled furthest this week is that India has been given a small pilot - fifty-odd brands, two agencies, a controlled trial with a long tail before general access. That framing is wrong, and getting it wrong will cost an agency the only planning advantage available.

The agency-gated window is eight days.

What “open” will probably mean

The US precedent is the best available guide, and it is unambiguous. The pilot began with a $250,000 minimum commitment. That fell to $200,000, then $50,000, then was removed entirely within fourteen weeks, at which point self-serve opened to advertisers of all sizes. There was no long queue and no rationing.

India opens at ₹725 a day. Applying the same pattern, the reasonable planning assumption is that from 4 September this channel is genuinely open, and that the constraint on who runs campaigns will not be access or budget.

The regulatory position

This is more settled than most people assume, and in one specific respect it lands directly on this format.

ASCI issued draft guidelines on AI in advertising on 22 May 2026, with consultation closing on 13 June. They set out a three-tier risk model. The middle tier - mandatory labelling - explicitly names “paid or sponsored product suggestions, which must specifically be labelled as ‘sponsored by’.” That is this format, described almost exactly.

Two honest caveats. The guidelines remain in draft as of today; we found no evidence they have been finalised. And no enforcement mechanism is specified in the draft or in any coverage of it - ASCI's regime is self-regulatory.

On data protection, the position is emptier. We found no MeitY statement, no regulator comment and no published DPDP analysis addressing conversational AI advertising. The obvious pressure point is the Personalised Ads setting, where past chats and stored memory feed ad ranking - a consent question the DPDP framework plainly reaches but has not yet been applied to in public.

What Indian marketers are actually saying about it is, so far, cautious rather than enthusiastic - and notably, the caution is about trust rather than about mechanics.

What nobody knows yet

06What changes, and how to be ready

6.1  What this does to the work

Take the standard Indian performance-marketing cycle - brief, plan, build, optimise, report - and run this channel through it. Three of the five stages change materially, and one of them stops existing in its current form.

The build stage collapses. There is no keyword research, no ad-group architecture, no negative list, no audience construction, no lookalike seeding. What remains is writing a context hint in plain language, uploading a customer list if you have one, setting geography and platform, and producing static creative. Work that took a competent executive two days takes an hour.

The optimise stage changes character entirely. There is no search-term report to mine, no placement report to exclude from, no demographic breakdown to shift budget across. The levers are creative, context hint, bid and budget. Optimisation stops being inspection and becomes experimentation - you cannot see why something worked, so you have to design a test that tells you.

The report stage has to be rebuilt. A monthly report organised around keyword and audience performance has nothing to render. Worse, with no impression-level attribution, the honest answer to “did this work?” requires a holdout or a geo test rather than a platform dashboard screenshot.

Where the money would come from

Nobody has incremental budget sitting unallocated. Any rupee that goes into this channel comes out of something, and being honest with a client about which something is the difference between a test that survives its first review and one that gets quietly cancelled.

The honest counter-case

Before the preparation advice, the argument against all of it. This channel may stay small for two years. Inventory is constrained by design - ads serve only on two tiers, only to logged-in adults, never in the highest-volume sensitive contexts. A large Indian advertiser may find there is simply not enough of it to matter. If that is how it goes, an agency that built a practice around this in September will have spent a quarter on something that never paid.

The counter to the counter is that the preparation described below costs very little - a few days of senior attention and a ₹725-a-day test - and the capability it builds transfers to whatever the next intent-based surface turns out to be. That asymmetry is the whole case.

6.2  How to prepare

The instinct will be to buy or build tooling. Resist it. Growth Without Margin made the point that no Indian independent wins a tooling arms race, and it applies here with extra force: there is nothing to tool yet. The preparation is entirely about people and process.

On the capability question: the person who should own this is not your best campaign operator. Operating skill is precisely what does not transfer. What transfers is the ability to form a hypothesis, design a test that could disprove it, and write down the result honestly - including when it is unflattering.

Most Indian agencies have that person. Most have them buried in a delivery role because delivery was what got billed.

6.3  How to advise the client

The temptation is to arrive with enthusiasm. Do not. Every Indian marketing head has read the same coverage you have, and several of them have already been pitched this week by somebody promising early-mover advantage on a channel that launched three days ago.

What is honestly claimable today is narrow and worth saying precisely: this is a real placement, on a real audience, at a low entry cost, with genuinely novel targeting, no benchmarks, immature attribution, and an unproven relationship between clicks and outcomes in India. Every one of those clauses is defensible. Anything beyond them is not.

The strongest position an agency can take this month is the one nobody else will take: tell the client what you do not know, then tell them what it would cost to find out. That is a different proposition from selling a channel, and it happens to be the proposition Growth Without Margin argued is the only one left with margin in it.

V24. Language framework, following the treatment established in Growth Without Margin. The right-hand column is harder to say and considerably more persuasive to a marketing head who has been pitched three times this week.

6.4  Which clients this actually suits

Most of your clients should not test this in September. Saying so is the fastest way to be believed about the ones who should.

Two gates and then a score. The first gate is policy: if the category is banned, the conversation ends. The second is measurement: if a client cannot run a holdout or does not have clean conversion tracking, they cannot learn anything from this channel and will end up arguing about a dashboard. Only after both gates does the scoring matrix apply.

6.5  How to pitch it

The structure that works is the one Growth Without Margin described for any post-automation conversation, applied to a specific channel: concede what you do not know, lead with their numbers rather than your credentials, and put the commercial architecture on the table before anybody asks for a discount.

Concretely, in order. Open with the two absences - no keywords, no audiences - because it demonstrates you have read the specification rather than the press release. State what you personally ran with your own money and what happened. Score their category against gate one in front of them. Then propose a test, not a campaign, with a defined end date and a stated condition under which you will recommend stopping.

On budget: ₹725 a day is a floor, not a plan. A test that cannot reach statistical usefulness is worse than no test, because it produces a number everyone then argues about. Size it to the question you are asking.

Not every enquiry should become a test, and the decision tree below is deliberately weighted toward saying no. An agency that runs four disciplined tests in September and declines eleven will be in a far stronger position in December than one that ran fifteen.

What to charge for it

Not a percentage of spend. At ₹725 a day a percentage produces a fee that would not cover the meeting to discuss it, and more importantly it prices the wrong thing - there is almost no media management to perform here.

The work is test design, context-hint writing, and independent evaluation. Price those.

Four of the five decision paths end without selling media. On a channel with no benchmarks, credibility is the only asset in the room - and it is spent every time you recommend a test that should not have been run.

07Conclusion

If you have read nothing else, this is the short version.

1. The eight-day agency window is not the opportunity. Self-serve opens on 4 September at ₹725 a day with no announced queue. Access will not be scarce.
2. Read the ad policy against your client list this week. Some of your largest accounts are banned outright and some sit in an approval process whose India status nobody has confirmed.
3. Spend your own ₹725 before you spend a client's. There is no substitute and no shortcut, and it takes two weeks.
4. Name one owner, chosen for reasoning rather than platform seniority. The craft that transfers is experimental design, not campaign operation.
5. Score your book, and expect most of it to fail. Recommending against is the service in September.
6. Price judgment and verification, not media management. There is barely any media management to sell.

Where this argument could be wrong

Five ways, in order of how much damage each would do.

• OpenAI adds keywords or demographic targeting. The entire argument rests on two absences. If either is filled in the next two quarters, this becomes an ordinary channel with an ordinary playbook, and the window described here closes before anyone walks through it. Nothing in the public roadmap says this will happen; nothing says it will not.
• Inventory stays too thin to matter. Ads serve on two tiers, to logged-in adults, never in sensitive contexts, never in Temporary Chats or after image generation. That is a lot of exclusion. If Indian delivery at realistic budgets is negligible, the correct strategy was to ignore this entirely until 2027.
• The primitives argument is asserted, not measured. We could find no published figure for what share of Indian billable agency work is keyword and audience operation. Section 3.6 argues it from campaign architecture and hiring patterns. It is a reasoned claim, and a reader is entitled to weigh it as one.
• The India market data is unreliable in ways that matter. The two dominant sources disagree, both are agency- or industry-published, and neither splits digital finely enough to say confidently how much of Indian spend is genuinely intent-driven. Part II is the shakiest evidence in this article and is presented as such.
• Indian buyers may simply not care. One of the few named Indian marketers on record this week set a bar of 3× ROAS against Google and Meta before allocating anything. If that is the market's actual posture, a channel with no benchmarks will not get a test budget regardless of how well an agency prepares.

The preparation recommended in Part IV holds under all five, which is the test it was built to pass. Every action on the readiness list is cheap, and the capability it builds - reasoning about intent, designing tests, evaluating honestly - transfers to whatever the next intent-based surface turns out to be.

The closing argument

Indian agencies were paid, for a long time, to operate two objects competently and repeatedly: the keyword and the audience segment. Around those two primitives an entire industry structure was built - job titles, team shapes, retainer sizes, reporting formats, the junior tier that fed the senior one.

A channel has now launched in India that has neither. Not as a limitation to be worked around, but as a design choice. What is left when you remove them is the two ends of the job: deciding what to do, and finding out whether it worked. Those were always what an agency was actually for. They were bundled inside the execution, invisible on the invoice, and subsidised by a labour pyramid that has stopped making economic sense.

So this channel is not really an opportunity to sell media. It is a place where the thing an agency ought to be selling is briefly the only thing there is to sell, because no playbook exists to sell instead.

Somebody is going to write the Indian playbook for this channel in the next few months. Until they do, judgment is the only thing anyone can sell here. That is the whole opportunity, and it is the only kind an agency should want.

Primary sources

• OpenAI - advertising announcements, help documentation on ads in ChatGPT and campaign creation, and published advertising policies. Accessed August 2026. Used for all product mechanics, tier eligibility, targeting capability, exclusions and policy categories.
• ASCI - Draft Guidelines on AI in Advertising, 22 May 2026. Consultation closed 13 June 2026. The “paid or sponsored product suggestions” quotation is verbatim from the draft.
• Indian business press - Business Standard, Business Today, Free Press Journal, Storyboard18, BestMediaInfo, Exchange4Media, for the India launch, user figures, executive quotes and platform revenue filings.
• Advertising trade press - PPC Land, Adweek, Search Engine Land, Search Engine Journal, eMarketer, for CPM trajectory, commitment-floor changes, and Ads Manager mechanics.
• Market data - Pitch Madison Advertising Report 2026, WPP Media This Year Next Year (December 2025), FICCI-EY Media & Entertainment Report 2026, Zefmo Media (May 2026).

Method and its limits

Research was conducted on 27 August 2026, the day of the India launch. Figures are attributed and dated throughout. Agency-published, vendor and single-source figures are flagged where they appear. SEO listicles and marketer content-marketing were excluded at triage.

A note on the visual set. Thirty-one figures were specified for this article; twenty-eight were built. V16, a standalone diagram of the agency-gated versus open-access states, was folded into the magnified inset of V15, where the compression it illustrates reads more sharply against the full timeline. V17, a chart of the collapsing US commitment floor, was merged into V6 so it sits beside the CPM trajectory it explains. V9, a channel-split-over-time chart, was cut outright because the underlying sources could not be reconciled across years without manufacturing a precision that does not exist - the two-panel treatment in V8 is what honesty allowed instead. Recording the cuts is part of the method: a visual that survives only because it was planned is decoration.