In June 2026, Adthena scraped 169,560 UK search queries looking for paid ChatGPT placements and found exactly zero.
That is the number to sit with before anyone tells you this channel is inevitable. The same dataset counted 7,378 distinct advertisers on ChatGPT in a single week and 97 times growth in placements over a quarter, so the channel is unquestionably real and growing fast. It is also, in at least one large market, capable of serving nothing at all across a sample of 169,560 queries.
Both things are true. A channel can be technically live in your market and still not deliver meaningful inventory to you. That gap between availability and delivery is what this framework is designed to test cheaply, before it becomes a line item somebody has to defend.
This is a decision guide for mobile app teams, current as of September 2026.
The case for testing now
Three arguments hold up.
Entry cost is genuinely low. The minimum daily campaign spend in India is ₹725. For most teams that is a rounding error against monthly UA spend, which makes this a cheap option on future optionality rather than a budget decision.
Competition is thin and geographically lopsided. Of the 7,378 advertisers Adthena counted, 60.1% were in the US. India entered the channel on 29 August 2026 with almost no domestic competition in the auction, which is the main reason to expect low early CPMs. It is also the main reason they will not stay low.
The intent quality looks unusual. Similarweb found sponsored placements in 26% of ChatGPT responses at a 0.50% click-through rate. Adthena separately found retail took 39% of placements while appearing on 24% of US queries, which suggests ads concentrate on transactional and comparison-shaped questions rather than spreading evenly across conversations. A user asking which expense tracker to use is closer to a purchase decision than a user scrolling a feed.
If your app sits in a category people research before choosing, that third point is the strongest argument on this page.
The case against testing now
Four arguments, and they are not weak ones.
Delivery is not guaranteed by availability. The UK zero-placement finding is the cleanest evidence available that a live market can produce no impressions for a given advertiser set. Nothing published guarantees India behaves differently.
You cannot compute ROAS without manual work. There is no cost API for ChatGPT Ads, so spend does not flow into your measurement platform the way it does from Meta or Google. Attributed revenue is available; the spend side has to be imported by hand every week for the two numbers to meet.
Reporting is thinner than you are used to. Targeting splits iOS, Android and web. Insights returns two device buckets with mobile web folded into Mobile. View-through exists at a fixed one-day window and is excluded from bidding, billing and CPA. Attributed conversions lag 24 to 48 hours.
Volume at floor budget is small. At ₹725 a day, on any plausible CPM, you are looking at single-digit to low-double-digit installs per day. That is enough to answer "does this work at all" and nowhere near enough to answer "how well does this perform".
No performance data exists for India. Not from OpenAI, not from a third party, not from any MMP. Anyone quoting an India CPI figure in September 2026 has either not shown their sample or does not have one.
Six readiness gates
Run these before you spend anything. Each is a pass or a fail, not a score, because a scoring system lets you talk yourself past a gate you failed.
1. Can you attribute installs from a click-based channel today?
Pass if you have a measurement platform that generates attribution links and your SDK is live and validated. Fail if you rely on platform-reported conversions from Meta and Google and have never run an attribution link. This gate is absolute: on a click-based channel, attribution cannot be backfilled, so a fail here means the test produces no evidence at all.
2. Do you have one activation event you trust?
Pass if there is a single post-install event that reliably separates real users from curious taps, and it fires correctly today. Fail if your event structure is undocumented or your team argues about which event means "activated".
3. Is ₹725 a day genuinely immaterial to you?
Pass if a month of floor-budget spend would not require reallocating from a channel that is working. Fail if it would. A test you have to defend weekly is a test that gets killed before it produces a signal.
4. Can someone own fifteen minutes a week of manual spend import?
Pass if a named person will do it. Fail if the answer is "we'll automate it later". Without spend import you are running a cost-per-install test, not a ROAS test, and you should decide that deliberately rather than by accident.
5. Does your category match the channel's shape?
Pass if people research your category before choosing: fintech, insurance, marketplace, travel, comparison-driven eCommerce. Weaker if discovery is social and feed-driven, as it typically is for gaming, short-form content and dating.
6. Can you tolerate 30 days without a conclusion?
Pass if your reporting cadence can hold an inconclusive channel for a month. Fail if a new line item has to justify itself in the first fortnight, because at this volume it cannot.
Fail gate one and stop. Fail two or more of the rest and the honest answer is not yet.
If gate one is where you fall down, that is a broader problem than this channel, and the questions in what to ask in MMP demos that actually reveal product quality are a better use of the next fortnight than a ChatGPT Ads campaign. If you are weighing whether to build measurement yourself rather than buy it, the numbers in the true cost of custom attribution infrastructure are the ones that decide it.
The 30-day test design
Structure it to answer one question at a time.
Days 1 to 3: prove the plumbing, not the performance.
Launch one campaign with your attribution link in place. Confirm the click identifier arrives populated, installs attribute to ChatGPT Ads rather than to organic, and your activation event fires against the channel specifically. Wait the full 48 hours before concluding anything, because the reporting lag makes a correct setup look broken on day one. The mechanics are in the ChatGPT Ads install tracking setup guide.
Days 4 to 14: measure delivery, not efficiency.
Hold budget flat and change nothing. The question in this window is whether impressions are being served at all, which is the UK finding made concrete. A campaign that under-delivers and a campaign that delivers expensive installs are different problems with different responses, and cost per install cannot tell them apart.
Days 15 to 30: one variable.
Split by platform, or test a second creative angle. One, not three. At single-digit daily installs you cannot read a multi-variable test, and the discipline in the five-day creative testing sprint applies with the timelines stretched to match the smaller sample.
Throughout: track activation rate, not install count.
The number that decides this is how ChatGPT Ads installs activate relative to your Meta and Google installs. Install volume at floor budget is noise. Activation rate is signal even at small samples, because a channel delivering users who behave like your good users is worth scaling and a channel delivering users who never activate is not.
Kill criteria, written before you launch
Decide these now, in writing, with a date attached. This is the part of the framework that does the most work, because a channel with no kill criteria does not get killed, it gets forgotten while it spends.
Reasonable defaults, to be adjusted to your own numbers:
- Delivery floor. If daily impressions stay below a threshold you set in week one, stop. You are not testing the channel, you are testing an empty auction.
- Cost. If cost per install after 30 days is more than double your blended average, stop, unless activation rate is materially better and justifies the premium.
- Quality. If activation rate on ChatGPT Ads installs is below half your Meta and Google baseline at day 30, stop regardless of how cheap the installs were.
- Attribution integrity. If you cannot confirm by day 5 that installs are attributing rather than landing in organic, stop and fix rather than continuing to spend.
Also write down what a pass looks like, which teams skip. A pass is usually: delivery adequate, cost per install within a defined multiple of blended, and activation rate within a defined distance of your best channel. Without that written down, a marginal result becomes an argument rather than a decision.
The general version of this discipline, applied across channels, is in the budget allocation framework for multi-channel app growth.
What a reasonable "no" looks like
Deciding not to test is a legitimate outcome and is under-represented in content about new channels, for obvious reasons.
The strongest case for waiting is that India has no published performance data, and by waiting a quarter you get to read someone else's evidence instead of paying for your own. The cost of waiting is that CPMs rise as the auction fills, so you buy the same evidence later at a higher price.
That trade is genuinely close, and it turns mostly on gate three. If ₹725 a day is immaterial, the option value of testing early is worth more than the information you would gain by waiting. If it is not immaterial, waiting is the better call and nobody should feel bad about it.
What is not reasonable is a third position that is common in practice: launching a campaign without attribution, watching organic installs rise, and concluding that the channel works. That produces a decision based on nothing, at full price.
What would change this decision
The framework above is calibrated to September 2026. Four developments would move it, and it is worth knowing which ones you are waiting for so that "not yet" has an exit condition rather than being a permanent position.
A cost API arrives. This is the big one. The moment spend flows automatically into measurement platforms, ROAS becomes computable without weekly manual work, gate four disappears, and the channel becomes directly comparable to Meta and Google in your existing reporting. Until then, every ROAS figure for this channel is assembled by hand and carries whatever error that introduces.
India performance data gets published. Right now nobody can tell you what a good ChatGPT Ads CPI looks like in this market, which means your own test is the only evidence available and you are paying for it. Once benchmarks exist, from any credible source, the cost of waiting drops sharply because you can calibrate against someone else's spend.
Reporting granularity improves. Two device buckets is the constraint that hurts most in Android-first markets. Platform-level reporting in Insights would remove a workaround that currently requires running separate campaigns per platform.
View-through becomes actionable. The one-day window is currently excluded from bidding, billing and CPA. If that changes, the channel's measurable surface widens considerably, particularly for categories with longer consideration cycles where a click-only model understates contribution.
Note what is not on this list: CPMs falling further. Cheaper inventory is pleasant and it does not change whether you can measure what you bought. A channel you cannot measure does not become a better decision by getting cheaper, it becomes a cheaper way to generate unattributable installs. Price is the variable most likely to be quoted at you and the least relevant to this particular decision.
Set a calendar reminder for the quarter and check these four rather than re-litigating the whole question from scratch.
Frequently asked questions
Are ChatGPT Ads worth it for app installs?
Worth testing at floor budget if you can attribute click-based installs and have an activation event you trust. Not yet worth forecasting on, because no India performance data has been published and the channel has no cost API, so ROAS has to be assembled by hand.
How much do I need to spend to get a signal from ChatGPT Ads?
Enough days, not enough rupees. At the ₹725 minimum you will accumulate a readable activation-rate comparison in about 30 days. Raising budget before you have confirmed attribution and delivery buys you noise faster.
What is the biggest risk in testing ChatGPT Ads?
Spending without attribution. Because the channel is click-based, installs from a campaign that ran without a tracking link cannot be recovered retrospectively, so the money produces no evidence at all.
Should agencies test ChatGPT Ads on client budgets?
Only with the kill criteria agreed in advance and in writing. The reporting gaps on this channel, particularly the absent cost integration and the two device buckets, will generate client questions that are much easier to answer before the spend than after it.
How does ChatGPT Ads compare to adding another Meta campaign?
An incremental Meta campaign is more predictable and less interesting. ChatGPT Ads is a cheap option on a channel that may matter in twelve months. Treat it as an option purchase, not a reallocation, and the comparison resolves itself.
Where Linkrunner fits
Gate one is the only gate this framework treats as absolute, and it is the one a measurement platform decides.
Linkrunner is the first MMP in India to go live with ChatGPT Ads, and attributes those installs alongside Meta, Google and TikTok in a single dashboard, without an enterprise contract or a minimum volume commitment. On a channel most teams will test at ₹725 a day, a minimum commitment is the thing most likely to stop the test happening at all. The activation-rate comparison that this framework hangs on is a same-screen comparison rather than a spreadsheet exercise, which matters because a comparison that takes an afternoon stops being run by week three.
OpenAI's mobile measurement partner setup article is the primary source for how the channel behaves, and Adthena's UK placement findings are worth reading directly rather than through this summary.
Key takeaways
- Adthena found zero paid ChatGPT placements across 169,560 UK queries in June 2026, against 7,378 advertisers and 97x placement growth. A live market is not a delivering market.
- Entry cost in India is ₹725 a day, competition is thin, and 60.1% of advertisers are in the US.
- No cost API means ROAS has to be assembled manually. No India performance data exists from any source.
- Gate one is absolute: if you cannot attribute click-based installs today, the test produces no evidence.
- Judge the channel on activation rate against your existing channels, not on install count.
- Write kill criteria and pass criteria before launch, with a date attached.
What to do this week
If you passed all six gates, launch one campaign at the floor with your attribution link in place and your kill criteria written down. Total setup time is under an hour.
If you failed gate one, fix that instead. It is worth more than this channel, because it applies to every channel you already run.
And if you passed the gates but the timing is wrong, put a calendar reminder for December to re-read the evidence. India performance data will exist by then, most likely including your competitors', and the decision gets easier for the price of a quarter of higher CPMs.
To check whether your setup clears gate one, request a demo from Linkrunner.