ChatGPT Ads in India: Lessons From Our Own Test
An ongoing test, updated every week. Latest: week 2, 8 to 13 November 2026, right below this note. The eight lessons from week 1 (2 to 7 November) follow it.
Week 2 update: we stopped, then kept only what worked
On 11 November we stopped all our ChatGPT ads. Across the two weeks, each qualified enquiry had cost us about four and a half times our target, and none had turned into a sale yet.
Then we split the results by ad topic. In our main campaign, the ads for founders asking about a Shopify store took 22% of the spend and brought every qualified enquiry that campaign produced. The ads about what an ecommerce website costs, and the ones for sellers who take orders in DMs, took the other 78% and brought none.
So on 13 November we switched back on only the Shopify ads, at a lower daily budget, and kept the rest off.
We also changed when they run: Monday to Friday, 9 am to 7 pm IST, off at night and on weekends. Late-night enquiries were the weakest in our data (lesson 8), and we want to see whether daytime-only cuts the enquiries from students. ChatGPT's Ads Manager has no scheduling, so we built it ourselves.
How the schedule works. A small script on our own server runs every 15 minutes and calls the ChatGPT ads API. Outside 9 am to 7 pm on weekdays, it pauses every campaign that is running. When the window opens, it switches back on only the campaigns it paused itself, so a campaign we pause by hand stays off. After each change it reads the account again to confirm the campaign really is on or off. The first Monday restart, on 16 November at 9 am, ran without anyone touching it. In the first hour we saw no spending rush (72 impressions, no clicks), but one morning proves little, so we keep watching.
We also refreshed the list of everyone who has already enquired or become a client, and exclude it from the ads, so we don't pay for their clicks again.
Next week will show whether this narrower setup brings qualified enquiries at a cost we can live with.
Founders now ask ChatGPT the questions they used to type into Google: what an online store costs in India, who can build it, how long it takes. Some of those founders were already finding us through ChatGPT without any ads. So when ChatGPT ads opened to Indian advertisers, I put our own money in to see what happens when you pay to be part of that answer.
We ran them from 2 to 7 November 2026, for India only. This post shares percentages, not rupee figures, but every number comes from our ad account or our CRM. One week is a small sample, and I'll say where it's too small to trust. I also run an agency, so read me with that in mind.
What we set up
- Two landing pages, tested against each other: our get-started page and our Shopify development page.
- Two ways of bidding: campaigns that chase clicks, and campaigns that chase a "qualified enquiry". Our website tells ChatGPT when a form passes our budget and timeline questions, from the server as well as the browser.
- Small tests for single services: store migration, headless builds, mobile apps, conversion work, and selling online for small sellers.
- Ads written around the questions founders ask: what a store costs, who builds Shopify stores in India, how to stop selling through DMs.
I went in with eight assumptions. Most of them were wrong.
1. The daily budget is an average, not a ceiling
I set a daily budget and expected spending to stop there. It didn't. On day one the account spent 2× the daily budget, most of it in about an hour. Day two: 1.75×. Later in the week, every one of our small test campaigns spent about double its daily budget in a single day.
That isn't a fault. It's the rule, and it's written down. OpenAI's billing and payment help calls the daily budget the average you want to spend per day over seven days, not a same-day limit: a single day can bill up to twice the daily budget, and seven days together never more than seven times it. Their help article on budget pacing adds that spend is paced through the day, so some days spend more and some spend less. Our 2× days were the system working as designed.
What to do: plan by the week, not by the day. Seven times your daily budget is the number you need to be comfortable with. If one day at double your budget would hurt, your daily budget is too high.
2. A stopped campaign comes back in a rush
Every time our ads stopped (a failed payment, a verification check) and came back in the middle of a day, the day's unspent budget went out fast. One evening, after a payment cleared, the account spent close to two days' worth of budget in about 40 minutes.
Those were also our most expensive days. Cost per click nearly doubled from day one to day two, and by day four it was 2.3× day one. Click-through rate barely moved (1.24%, then 1.27%, then 1.39%), so people weren't ignoring the ads. We were simply paying more for each view: the cost of 1,000 impressions rose about 2.5×. I can't prove the rush caused the higher prices, because OpenAI doesn't publish how pacing bids. But the rush days were the expensive days.
What to do: switch campaigns back on at the start of a day, so pacing has the whole day to work with. And fix the payment side first (lesson 6), because a stop is what sets up the rush.
3. The cheapest clicks found the wrong people
Our "stop selling in DMs" ads, written for small Instagram and WhatsApp sellers, had by far the cheapest clicks: roughly a quarter to a third of what our Shopify and store-cost ads paid. They brought 71% of all our leads. Not one of them was qualified. One was a student doing research. Another was a development agency looking to buy hours. Neither was the founder the ad was written for.
The Shopify and "what does a store cost" ads paid far more per click and brought 21% of our leads. Two-thirds of those were qualified. A small store-migration test brought a single lead, and it was qualified too.
The numbers are small, so don't treat this as a law. But it matched what our form kept telling us: someone asking ChatGPT how to sell in DMs is usually still learning, not hiring.
What to do: judge every topic by qualified enquiries, never by cost per click. We paused the cheap topics and kept the expensive ones, knowing our average cost per click would go up.
4. "Optimise for conversions" still means paying per click
I thought conversion bidding meant we would pay when someone enquired. It doesn't, and OpenAI's billing page says so plainly: ads are not billed per conversion. You still pay for every click; ChatGPT only tries to show the ad to people who are more likely to convert. On day one, our conversion campaigns did pay about 30% less per click than our click campaigns. That's one day, but it's a good sign.
There are two catches, both in OpenAI's conversion tracking docs. A conversion campaign optimises for exactly one event, chosen when you create it, and it can't be changed afterwards. We tried to add "deal won" tracking to our running campaigns, and the system refused. And it needs conversions to learn from. With a handful in a week, it learns very little.
What to do: send only the enquiries you would actually want as your conversion event. If you send every form fill, students included, you are teaching the system to find you more students.
5. Almost nobody started the form
Fewer than 1 in 100 clicks even reached the first step of our enquiry form. Speed wasn't the problem: our main landing page shows its largest element in about 1.8 seconds on a slow 4G test. The first screen was the problem. Someone who had asked ChatGPT about Shopify costs landed on a headline written for Instagram sellers.
It's a leaky bucket. We were paying to fill it while the first screen let 99% run straight out. Patch first, fill second.
What to do: make the first screen of your landing page answer the exact question the ad answered. That will move your results more than any change to cost per click.
6. Indian cards and automatic payments don't mix easily
ChatGPT ads are postpaid. OpenAI charges your card automatically each time your unpaid spend crosses a threshold, and again at month end (billing and payment help). New advertisers start with a low threshold, so the charges come often, and nobody, not even their support team, can change it by hand. Under RBI rules, an automatic charge on an Indian card needs an active e-mandate (Stripe explains how that works). Our automatic charges failed again and again, and every failed charge stopped every ad. Adding a new card didn't fix it either. Each card we added placed a temporary authorisation hold (money held, not taken) that took days to come back.
In our first six days, the ads stopped at some point on every single day, for billing or for verification.
What worked: when a charge fails, open Billing in Ads Manager, find the unpaid invoice under Documents and pay it there with an OTP. Keep one card with one active e-mandate, and don't cancel that mandate at your bank. Don't keep adding cards; each one is another hold.
7. There's a spend limit until you're verified
On day two, every campaign stopped for an identity verification step. Ads Manager later explained why: the account had reached its verification spend limit. Our first attempt was declined, there was no retry button, and only OpenAI's support team could open a new review. It took about two days.
OpenAI's onboarding troubleshooting page explains both halves. Some accounts can run ads before the extra business verification, and reaching a spend threshold can trigger it. And a decline usually means they couldn't confirm the business's identity, ownership or affiliations, or couldn't match the advertiser to the website's domain.
What to do: finish business and identity verification before you launch, and make sure your brand name, your legal company name and your website clearly belong together. Allow a few days for it.
8. Most founders enquire in working hours
My belief going in: founders are busy all day and look into things like this at night, at home. And students, who make up about half of ChatGPT's users in India according to OpenAI, would fill the evenings.
ChatGPT's ad reports don't break down by hour. So I looked at every enquiry form submitted on our website since late July, from all sources, including the junk we had already deleted:
| When the form was sent (IST) | Share of all enquiries | Junk rate | Share of deals won |
|---|---|---|---|
| 10 am–7 pm | 58% | 27% | 67% |
| 7 pm–midnight | 23% | 38% | 20% |
| Midnight–10 am | 19% | 28% | 13% |
Most enquiries came in during working hours, and so did two-thirds of the deals we won. For a business owner, a website is work, and many look into it between customers. Evenings had more junk (38% against 27%), which fits the student theory, but on this sample the gap could still be chance. Late night was the weakest window. Sunday was our quietest day, with 8% of the week's enquiries.
One thing that confused me at first: more enquiries in the day doesn't mean cheaper clicks. Click prices come from how many advertisers are bidding at that hour, and working hours are usually the busiest. What daytime gives you is more people who actually fill the form.
What to do: ChatGPT's Ads Manager has no ad scheduling, so running only in working hours means switching campaigns off yourself. Since 13 November we do it with a small script on our own server that calls the ads API every 15 minutes (see the week 2 update above). Expect a rush of spend each time a campaign comes back on (lesson 2). On our first Monday restart we saw none in the first hour, but one morning proves little.
Three small things that helped
- Excluding people we already know. We upload our existing enquiries and clients as a hashed exclusion list, so we don't pay to show them ads.
- Writing the targeting hints in the buyer's words. "I want to hire someone in India to build my online store", not developer words like "theme" or "development company". It's too early to measure, but the wording now matches the person we want.
- Showing the price in the ad. It probably costs us some clicks. The clicks it costs are mostly people who were never going to pay that price.
When not to run ChatGPT ads
Skip them for now if:
- you can't trace which enquiry came from which ad, all the way into your CRM;
- a single day at double your budget would hurt;
- your card can't carry an e-mandate for overseas charges;
- the first screen of your landing page doesn't answer the question your ad answers.
If your buyer is the kind of person who asks ChatGPT before hiring someone, a small test is worth it. Judge it only on qualified enquiries.
This is one week of observation, not a verdict. I'll update this post every week with whatever new we learn, and with the final result when the test ends: which landing page won, and how ChatGPT compared with our other channels. If you're planning ChatGPT ads for your own store and want a second pair of eyes on the setup, book a strategy call. Thirty minutes, no sales pitch.
Frequently asked questions
Can Indian businesses run ChatGPT ads?
Yes. We run ours from India, target India and are billed in rupees. Expect a business verification step, with a spend limit until it's done, and set up your card's e-mandate before you launch.
Does ChatGPT stick to my daily budget?
No. OpenAI treats the daily budget as a seven-day average. A single day can bill up to twice the daily budget, and seven days never more than seven times it. In our first week, that is exactly what we saw. Plan your spend by the week.
Do you pay per conversion with conversion bidding?
No. You still pay per click. Conversion bidding changes who ChatGPT shows the ad to. On our first day, it paid about 30% less per click than click bidding.
Why do automatic payments fail on Indian cards?
Indian cards need an active e-mandate for automatic overseas charges. When a charge fails, every ad stops. Pay the open invoice inside Ads Manager, under Billing and then Documents, with an OTP.
Can I run ChatGPT ads only during working hours?
Not in Ads Manager: we found no ad scheduling and no hour-by-hour report. The ads API can do it. Since 13 November a small script on our own server checks every 15 minutes. Outside 9 am to 7 pm on weekdays it pauses our campaign, and inside that window it switches it back on. It only restarts campaigns it paused itself, so one we pause by hand stays off. Expect a spending rush right after a restart, though our first Monday restart showed none in the first hour.
How much do ChatGPT ads cost in India?
You set a daily budget, and the minimum is ₹500 a day per campaign. You pay per click, even with conversion bidding, and a single day can bill up to twice the daily budget. We share our results in percentages, not rupees: in our first two weeks, each qualified enquiry cost about four and a half times our target.
How do I start ChatGPT ads in India?
Create an advertiser account in ChatGPT's Ads Manager, finish the business verification, and add a card with an active e-mandate for overseas charges. Then create a campaign with a daily budget and an end date. We chose conversion bidding and send ChatGPT our qualified enquiries from the website, so it learns from leads, not just clicks.
The service for this problem
Growth
Performance Marketing
Meta & Google Ads with real ROI tracking
- Digital Marketing for EcommerceYou’re spending ₹30K/day profitably but every time you scale to ₹1L/day, ROAS tanks.
- Digital Marketing for Ecommerce
Festival Ad Strategy — Meta + Google, Diwali 2026
Ad costs climb around Diwali. How to phase your Meta and Google festive budget — from pre-season audience building to post-Diwali retention. - Digital Marketing for Ecommerce
UGC Marketing for Indian D2C — Free Content That Sells
UGC marketing for Indian D2C: how I collect customer photos and videos, get permission, and turn them into ads that often beat studio creative.
Apply this note to your own store.
A 30-minute call — we'll map what works for your store, not a generic playbook. No sales pitch.
Calls Monday to Friday, in Telugu, English, Hindi or Kannada.