E-COMMERCE 101 FOR CHATGPT

ChatGPT Ads: is it for your brand?

For the right brand, it is the most important channel you are not on yet. For the wrong one, it is a quick way to lose money. This guide tells you which you are, then shows you how to do it properly.

PART 1 · 10 MINUTES

Decide

Is this for you or not? Start here. Leave if the answer is no.

PART 2 · THE FULL HOW-TO

Do it

Only if Part 1 said yes. Ordered by what matters most.

By Porcellia · Updated 12 October 2026 · Based on OpenAI's own documentation and early advertiser reports

1 · THE VERDICT

For some brands this is the best move available. For most, it is a waste of money.

There is no middle. ChatGPT ads reward a specific kind of product and punish everything else. The rest of this part tells you which side you are on.

IF IT FITS YOUR BRAND

Not being on it is the mistake.

Your buyers are already asking ChatGPT what to buy. The auction is young, few brands are in it, and the ones learning now will know how to win it when everyone else arrives. You reach new customers at the exact moment they are deciding.

IF IT DOES NOT

You will burn money and weeks.

Clicks here are expensive for what a cheap, impulse product can earn back. There is no age or interest targeting to rescue you, and no way to turn a ₹600 product into a good bet. You will spend, see little, and blame the platform.

FROM OUR ACCOUNTS

We are not guessing. Porcellia is running ChatGPT ads for 7 to 8 brands today, with close to ₹15 lakh spent and a blended ROAS of about 2× across them. Some brands are far above that. Some would have been better off not starting. That is why Part 1 exists.

2 · WHY IT BEHAVES THIS WAY

Three facts that decide who wins

Understand these and the rest of this guide follows.

NOT A FEED

1. The shopper is in a conversation

Nobody scrolls ChatGPT for fun. They are asking for help with a decision, often across many messages. They are researching, comparing and weighing up. That is a consideration moment, not an impulse moment.

NO AUDIENCE TARGETING

2. You buy relevance, not people

There is no age, gender or interest targeting. You describe the conversation you want to appear in. Relevance is scarce and the floor is ₹725 a day per campaign. US starting CPCs are reported at roughly $3 to $5. India has no published benchmark yet.

HARD TO MEASURE

3. The click is rarely the decision

Much of the influence happens before the click and after it. Early advertiser analysis suggests roughly 60% of influenced conversions are not traceable to a click. This channel is read in the whole, not in a single dashboard.

WHAT FOLLOWS FROM THIS

Expensive clicks plus a long, multi-step path to purchase means the channel only pays when one customer is worth a lot and takes some winning. Illustration only: if a click costs ₹100 and 2 in 100 visitors buy, every order costs ₹5,000 in clicks. That works for a ₹40,000 sofa. It does not work for a ₹799 T-shirt.

2B · WHAT THE SHOPPER SEES

Who sees the ad, and what it looks like

Before you spend, know exactly who your ad reaches and what it looks like when it appears.

Who sees it

Your ad reaches people on the free and low-cost tiers, not everyone using ChatGPT. Anyone who pays for Plus or above never sees it. Ads are matched to the topic of the current conversation and, if the user has personalisation on, signals from their wider ChatGPT use.

What it looks like

• It appears below ChatGPT's answer, never inside it. It is labelled “Sponsored” (or “Ad”) and boxed off from the answer.
• A standard ad shows the advertiser name, logo, a headline, a short description, an image and a link to your page. A feed campaign shows the product image, name, price and rating from your feed, and can show several of your products side by side.
• OpenAI decides whether a shopper sees a single product or a row of products. You do not choose.
• Ads do not change ChatGPT's answer, and advertisers never see chats, memories, names, emails or location. You only get totals such as impressions and clicks.
• Shoppers can tap “About this ad” to see who paid for it and why it appeared, hide it, or delete their ad data. ChatGPT cannot see the ad unless the shopper chooses “Ask ChatGPT” from the ad's menu.

A real example

This is the example OpenAI used to launch ads on 9 February 2026. The person was planning a potluck and asked ChatGPT for ideas. ChatGPT answered. Then, underneath, separate from the answer:

ChatGPT's answer ends here…

Heirloom Groceries

La Mesa Roja Enchilada Kit

$14.99 · ready in 25 to 35 minutes

Suggested follow-up: “Topping and garnish ideas”

Notice what happened. The shopper did not search for a kit. They were planning a meal, and a relevant product appeared at the right moment. That is the whole channel. In India the same format has been seen with labelled “Ad” cards, for example a headline and description from SerpApi under a search-related answer.

SHOPPING INSIDE CHATGPT

Two things are in play, and they are not the same. First, OpenAI is testing Sponsored Agents: after clicking an ad, a shopper can start a conversation with a business-sponsored agent. It is clearly labelled and kept separate from ChatGPT's own answers. It is being tested with select US advertisers. Second, OpenAI has built buy-in-chat shopping with Shopify and Etsy. Whether a ChatGPT ad can complete a purchase inside the chat in India is not confirmed in OpenAI's ad documentation. Today, plan for the click to land on your own product page, and check Ads Manager for what is live for your account.

3 · WHO IT IS FOR

It is for brands that need the shopper to think

Four profiles. Most strong fits match two or more.

PROFILE 1

High order value

Typically an average order of ₹2,500 or more, and the higher the better. A big basket can carry an expensive click and a long journey. This is a working rule of thumb, not an OpenAI figure.

Furniture, appliances, premium electronics, jewellery, luxury skincare, premium footwear, gifting sets.

PROFILE 2

Complex category

Products people cannot judge from a photo. They need to understand specs, ingredients, materials, fit or compatibility before they commit. A conversation is exactly where that questioning happens.

Skincare by skin type, ergonomic furniture, audio gear, mattresses, kitchen appliances.

PROFILE 3

New category

If customers do not yet know to search for what you sell, Google cannot find them. In ChatGPT they describe a problem and are introduced to the solution. You win by being the answer to a question they did not know had your product in it.

A new type of device, a new ingredient, a new way of doing something.

PROFILE 4

Several touchpoints needed

If a sale needs education, repeated exposure and a strong page to convert a cold visitor, this channel opens the relationship early. It suits brands willing to do deep conversion work on the product page, not just buy a click.

Anything where the first ad rarely closes the sale.

FROM OUR ACCOUNTS

Complex category: one of our tech-category brands is running at roughly 3× ROAS on ChatGPT. High order value: a US brand of ours selling a very high-ticket product is at almost 5× ROAS.

PORCELLIA · GROWTH SPECIALISTS

We are already running this. Not theorising about it.

Porcellia is team of growth specialists who moves early on new channels. We started email marketing for our brands in 2021; Reddit in 2020; Meta in 2017; CRO in 2022. ChatGPT ads are already live for 7 to 8 of our brands right now.

If Part 1 made you think “this is me”, you should speak to one of our growth consultants.

7–8

brands running ChatGPT ads with us

~₹15L

spent on ChatGPT so far

2×

blended ROAS across those brands

Results are from Porcellia client accounts, vary by brand, and are not a guarantee.

4 · WHO IT IS NOT FOR

If this is you, stop here

Better to know in three minutes than three weeks.

THE BIG ONE

Low order value

Typically anything under ₹2,500 is a poor fit. The click costs more than the order can bear. A discount code or a bundle rarely fixes the maths for long.

CONVERTS ON FIRST SIGHT

Impulse purchases

If people buy the first time an ad appears, you do not need a conversation. You need a feed with a thumb-stopping image. Meta does that better and cheaper.

ALREADY KNOWN

Commodity products

If shoppers already know what they want and search by name, you are paying to intercept demand that was already coming to you.

NOT ALLOWED IN INDIA

Restricted categories

Dietary supplements, over-the-counter medicines, health or treatment claims, some period-care and maternity products, alcohol, gambling and adult. OpenAI's policy blocks these today, and also reviews your landing page.

WRONG EXPECTATION

Needs cheap, instantly traceable sales

If your reporting needs every sale tied to a click within a day, you will think it is failing when it is working. The platform is not built for that.

WRONG ROLE

Hoping to replace Meta or Google

It adds a new top-of-funnel layer. It does not replace the channels that capture demand you already have.

5 · THE GREY ZONE

When it is not obvious

• Under ₹2,500, but customers reorder or subscribe. Lifetime value can justify a first-order loss, but only if you have proof of repeat in your own data, not hope.
• Under ₹2,500, but sold as a high-value bundle or kit. If the ad leads with the ₹3,500 kit and not the ₹700 item, the maths can change. Advertise the bundle, not the single item.
• High AOV, but customers already search for you by name. You may be paying for people who were coming anyway. Use ChatGPT to reach people who do not yet know you.
• A strong fit, but a messy catalogue. You are a fit, and you are not ready. Fix the feed first. That is the next section.

6 · WHAT DECIDES SUCCESS

If you are a fit, three things decide whether it works

In this order. The first one is not negotiable.

1 · MAKE OR BREAK

Your product feed.

If you sell products online, the feed is the ad. Titles, descriptions, prices, images and links are pulled straight from it. A wrong price, a missing field, a stale file or a bad link does not warn you. It just quietly removes your products, or sends shoppers to a page that contradicts the ad. Get the feed wrong and nothing else you do will save the campaign.

2 · MAKE OR BREAK

Tracking you can trust.

Without Pixel and server-side events in place before launch, you are guessing. Many conversions will never show as a click, so you need every signal you can get.

3 · DRIVES PERFORMANCE

A page that converts a cold visitor.

The shopper arrives having just had a thoughtful conversation. If your page does not continue it, you lose them. Everything else (hints, bids, creative) is tuning on top of these three.

7 · YOUR VERDICT

Who is this for?

Count the statements that are true for your brand, then read the result.

THE CHECKLIST

Who is this for?

Read each statement and count how many are true for your brand today. Be honest. Nobody is checking.

Signs it is for you

• Our average order value is ₹2,500 or more.
• Buyers research, compare or ask questions before they buy.
• The product is complex: it needs explaining (ingredients, specs, fit, materials, how it works).
• The category is new or unfamiliar. Most customers would not know to search for it.
• A sale usually takes several touchpoints or several days.
• Our product pages are built to convert a cold, first-time visitor.
• We have a catalogue or a hero range whose product data we can get clean and keep clean.

Signs it is not

• Our average order value is mostly under ₹2,500.
• It is an impulse purchase. People buy the first time they see it.
• People already know what they want and search for it by name.
• We need cheap clicks, or a sale we can trace to a click within a day.
• We expect it to replace Meta or Google rather than add a new layer above them.
• Our main products are in a category OpenAI does not allow in India (supplements, over-the-counter medicines, health or treatment claims, alcohol, gambling, adult).

How to read your count

• Any statement in the last box on the right is true: stop. OpenAI's policy rules your category out in India today.
• 5 or more on the left, and at most 1 on the right: this is for you. Go to Part 2 and start with the product feed.
• 3 or 4 on the left, and at most 2 on the right: probably for you. Run a capped test on your strongest product only, and get the feed right first.
• 3 or more on the right: this is not for you. Put the budget in channels that suit your product.
• Anything else: not clear yet. Wait until your price, product or category changes.

PORCELLIA · COMPLEX CATEGORIES

One of our tech brands is at roughly 3× ROAS on ChatGPT.

Considered purchases, products that need explaining, buyers who compare. That is where the channel works best, and where we have the clearest results. Let us tell you if your brand is in that zone.

Results are from Porcellia client accounts, vary by brand, and are not a guarantee.

PART 2

How to do it properly

Only read on if Part 1 said this is for you. Sections are in order of importance. The first three are make or break. Everything else is tuning on top of them.

PART 2 · STEP 1

The product feed

For an ecommerce brand this is the whole game. Every ad is built from it. If it is wrong, everything downstream fails, and often without a single error message.

WHY THIS IS THE ONE THING

In a feed campaign you do not write ads. OpenAI generates each one at the moment it is shown, using your title, description, image, price and link. So your feed is your copywriter, your designer and your price tag at once. A malformed required field can silently disqualify a product. A stale price shows a wrong number to a shopper who has just been told to trust the ad.

1. When you need a feed (and when you do not)

A feed campaign carries an extra layer of structure that a standard one does not:

Feed → Campaign → Ad group → Product set → Product-ad template → The ad that shows

• Feed: the catalogue file you supply.
• Product set: a filtered slice of the feed that one ad group is allowed to use, such as one brand, one category, one price range or one custom label.
• Product-ad template: how a selected product's title, description, price and image are laid out in the ad.

WORTH KNOWING

Beta limits. Products in your feed are eligible for paid placements only. They do not enter ChatGPT's organic answers through this feed. Bulk campaign upload does not support feed campaigns, so you build them in the guided flow. Since the 4 September update, shopping-style placements require a genuine product feed.

2. Decide what goes in. Do not advertise everything.

Your fit verdict in Part 1 applies to the feed too. Putting your entire catalogue in is the most common way to waste the budget, because the platform will spend on whatever you allow it to.

• Include products with a high order value, a good margin, a story that needs explaining, and reliable stock.
• Leave out low-ticket items, items that are often out of stock, high-return SKUs, and anything in a restricted category (supplements, medicines, health claims and so on).
• Control this with the is_ads_eligible field. Set true only for products you want processed for ads. Feeds created inside Ads Manager treat products as ads-eligible by default unless an item is explicitly set to false, so confirm what your feed is doing.
• Tier your products with labels so you can bid and budget by group. Use custom_label_0 to custom_label_4 (or ads_metadata keys that OpenAI confirms for your integration), for example hero, high-margin, gifting.

3. The file: what every row must contain

One row per purchasable item or variant. A black size-10 shoe and a black size-9 shoe are two rows. UTF-8 text. Absolute URLs. Keep identifiers as text so leading zeros survive. Quote any cell containing commas, quotes or line breaks.

The fields that make a real difference to ad quality and control:

An example row (illustrative values):

CONFIRM INDIA DETAILS WITH OPENAI

The published feed examples use US dollars and US targeting. Practitioners note that INR and India targeting should be confirmed for your account. Check the currency and country fields in Ads Manager's feed setup and upload a small test file before you build the whole catalogue.

4. If you already have a Google Merchant Center feed

OpenAI accepts a Google-compatible format, but only after OpenAI confirms it for your registered feed. Required columns are id, title, description, link, image_link, availability, price and brand. Differences to watch: use preorder (not pre_order), unknown is not accepted, the seller name comes from your registered merchant display name, and ad eligibility comes from registered feed settings, not from uploaded flags. Files can be .csv, .txt or .tsv, optionally gzipped. This is the quickest route for brands already running Google Shopping.

5. Getting it into Ads Manager

• In Ads Manager, open Feeds and choose the feed row's menu to upload a file, set a hosted URL or connect SFTP.
• Wait for processing. Anywhere from a few minutes to a few hours, depending on size.
• Open Upload History. Check how many rows were accepted and rejected, and read every diagnostic.
• Create a product-feed campaign in the guided flow and link the feed.
• In each ad group, build a product set with filters (brand, category, price range, item ID, custom labels).
• Create the product-ad template. One per ad group. Macros fill in brand, title, body and price from each product. The image comes from the product.
• Check the sample preview. Confirm the product count matches what you intended and that titles, descriptions, images and landing pages look right.

Reading Upload History

Common diagnostics: missing_required_column, invalid_value (usually price or currency format), unsupported_file_type, invalid_sftp_directory_layout.

6. Keeping it alive

• Name an owner. A feed is operations, not a one-off task. Someone is responsible for it every week.
• Use a hosted URL or SFTP so the file refreshes on its own and avoids the two-week expiry.
• Sync stock and price constantly. To stop an item serving, set availability to out_of_stock explicitly. When it returns, send in_stock with current details. For frequent price and stock changes, OpenAI offers a partial-update (delta) API. Note the trap: CSV prices are in major units (12999.00 INR) but the API uses integer minor units.
• Run a sale? Update sale_price before the sale starts and remove it the moment it ends.
• Alert on failures. If a refresh fails, you want to hear about it the same day, not in two weeks.

7. India-specific details that decide whether the click converts

• Show the same GST-inclusive price in the feed and on the page. Use price for MRP and sale_price for the selling price.
• Make pincode serviceability, delivery estimate, shipping fee and COD availability visible on the product page the feed points to.
• Variant URLs should open the exact size and colour in the feed.
• State warranty, returns and installation where relevant. These are the objections the shopper just raised in the chat.

8. How feeds fail, and what to do

CATALOGUE SIZE: SOURCES DISAGREE

The Advertiser Guide cites a minimum of 1,000 products and a maximum of 2 million. Trade press reported a starter sample of 100 products before a full catalogue, and OpenAI's own feed guidance recommends starting with about 100 items. Confirm the live minimum in Ads Manager. If you only have a handful of products, run standard campaigns instead.

PORCELLIA · FEEDS THAT WORK

The feed is where most brands fail. It is also where we start.

Product data, eligibility, tracking and the first campaign structure, set up properly before a rupee is spent. A US brand of ours selling a very high-ticket product is generating almost 5× ROAS on ChatGPT.

Results are from Porcellia client accounts, vary by brand, and are not a guarantee.

PART 2 · STEP 2

Tracking, installed before launch

Many conversions never show as a click. You cannot afford to lose the ones that do.

BROWSER

Measurement Pixel

A snippet in your site's <head> tied to a Pixel ID from Ads Manager's Conversions section. It auto-fires page views and purchases. Blockers and browser limits cause misses.

SERVER

Conversions API

Sends events from your backend or a server-side GTM container. More reliable. Run both together.

• Share an event_id across pixel and server events so one purchase is not counted twice.
• Capture oppref, OpenAI's click identifier on your landing-page URL, into a first-party cookie. It ties a later conversion to the original ad click. Preserve it through redirects.
• Send order_created with value, currency, order ID and product details. Turn on automatic advanced matching.
• Standard events include contents_viewed, items_added, checkout_started and order_created. Custom events work for reporting but cannot be the optimisation event.
• Add UTMs to every ad URL and keep them identical across campaigns so GA4 can group the traffic.

BECAUSE ADS MANAGER WILL UNDERCOUNT

Triangulate with GA4 by UTM, a post-purchase survey with “ChatGPT” as an option, branded search and direct traffic trends, and pulsing spend on and off to see whether new-customer orders move. For COD brands, judge on delivered, kept orders, not orders created.

PART 2 · STEP 3

A page that converts a cold visitor

The shopper just had a considered conversation. Your page has to continue it, not reset it.

• Open on the exact product or collection in the ad. Never the homepage. Do not block OpenAI's crawlers (OAI-AdsBot, OAI-SearchBot).
• Answer the question the conversation was probably about: size, ingredients, compatibility, dimensions, comparison, who it is for.
• Price (GST-inclusive), offer, delivery date for their pincode, COD and returns, all visible without scrolling.
• Reviews and real photos next to the buy button. Proof matters most to someone who has never heard of you.
• Fast on mid-range Android over 4G. No pop-up wall, no forced login.
• Help one tap away, for example WhatsApp or chat.

If your fit is based on a high-ticket or complex product, the page is where most of the money is made or lost. Brands that need several touchpoints should plan to capture email and keep the conversation going on their own channels after the first visit.

PART 2 · STEP 4

Context hints: how you target

Instead of keywords you describe the conversations your product belongs in.

A context hint is a short, natural-language description at ad-group level covering what the product is, who it helps and when it is useful. Write one clear idea per hint, and write several per ad group.

• Additional: adds detail the ad and page do not already carry.
• Specific: a concrete feature, need or situation.
• Relevant and accurate: a real connection to a real use case. Wanting a group does not make you relevant to it.

WORTH KNOWING

Hints inform relevance. They are not targeting rules and cannot enforce location, schedule or exclusions. Split ad groups when needs call for different messages or landing pages, for example shopper (comparing specific products at a price) versus browser(new to the category, asking what to look for).

PART 2 · STEP 5

Campaign structure, objectives and budget

• A conversion campaign optimises for one standard event, cannot be changed afterwards, and cannot use a custom event.
• Minimum ₹725 per day per campaign. A daily budget works as a seven-day average, so a strong day can spend up to about double.
• Give a new campaign a full week before changing bids or budgets.

PART 2 · STEP 6

Creative for standard ads

For feed campaigns your feed is the creative. For standard campaigns, short and specific wins.

• Title up to 50 characters (shorter displays most cleanly). Body up to 100. One image, square, PNG or JPG, no larger than 1200 × 1200, at a public URL.
• Make each variant a different angle, not a rewording. Cover the need, the constraint, the proof and the objection.
• Be accurate. Ads that contradict the landing page are rejected.
• You can opt in to AI text customisation, which adapts copy to the conversation. Test it on one ad group first.
• Confirm limits in the live Ads Manager. They change in beta.

SPONSORED

Kasa Home

Compact desks for small flats

90 cm wide, built-in cable tray. Free assembly in 40 cities.

Title 29 chars · Body 60 chars

Context hint: Compact work-from-home desks for small apartments where floor space is limited.

SPONSORED

Nuvora Skin

Gel moisturiser, oily skin

Non-greasy hydration for humid weather. Fragrance-free. 30-day returns.

Title 26 chars · Body 71 chars

Context hint: Lightweight gel moisturiser for oily, acne-prone skin in hot, humid climates.

Fictional brands, for pattern only.

PORCELLIA · AHEAD OF THE CURVE

This channel changes almost weekly. We stay on top of it so you do not have to.

Beta features, new policies, new tracking options. We have close to ₹15 lakh of live spend teaching us what actually moves results across 7 to 8 brands.

Results are from Porcellia client accounts, vary by brand, and are not a guarantee.

PART 2 · STEP 7

Eligibility and policy

Early ads are limited mainly to consumer verticals: lifestyle and household goods, local services, travel, and digital products or education. Outside the US, the sensitive categories are much tighter. Review the live Ads Policies before you plan.

Ad title, copy, image and landing page are all reviewed. Ads are also never shown in sensitive contexts such as mental or personal health conversations.

PART 2 · STEP 8

Account setup

• Create an account at ads.openai.com with a work email.
• Submit business details and billing. Verification takes a few business days.
• Add a card. Billing runs as spend accrues, with no account-level minimum.
• Let OpenAI's crawlers reach your site.
• Install tracking, upload the feed, and build the campaign in the guided flow.
• Submit for review. Ads typically clear in about 24 hours once the account is approved.

PART 2 · STEP 9

Targeting and audiences

• You control country, platform (iOS, Android, web), custom audiences and context hints. There is no age, gender or interest targeting.
• Custom audiences come from emails, phone numbers (plain or SHA-256 hashed) or Google Advertising IDs. Inclusion and bid multipliers need at least 25,000 matched users (100,000 or more recommended). Exclusions have no minimum.
• Best use for most brands: exclude existing customers so spend goes to new-customer acquisition.
• No lookalikes. Only upload data you have consent to use under India's DPDP Act.

PART 2 · STEP 10

Reading results and when to change things

• Impressions, clicks and CTR refresh about every 15 minutes. Spend can lag 7 to 8 hours, so a displayed zero does not mean no charge.
• Do not judge on day 2. Give it a week, change one thing at a time, and read by ad group and by product set.
• If tracking breaks, pause, fix and re-test. If a click arrives but the page does not convert, fix the page before touching bids.
• If delivery only appears at bids above what your order value can carry, stop. Your product is not a fit for this channel.

THIS IS A SNAPSHOT

Prepared 12 October 2026. ChatGPT Ads is a beta that changes close to weekly. Where sources disagree, this guide follows OpenAI's own documentation. Quoted results from individual advertisers are not benchmarks or promises.

PORCELLIA · LET'S TALK

If this is for you, the next 30 days matter.

Early advertisers learn while the auction is young. Book a call and we will tell you honestly whether ChatGPT ads fit your brand, and what a first test should look like.

7–8

brands running ChatGPT ads with us

~₹15L

spent on ChatGPT so far

2×

blended ROAS across those brands

Results are from Porcellia client accounts, vary by brand, and are not a guarantee.

SOURCES

• OpenAI Developers: Ads product feeds, product file schema and file upload overview (developers.openai.com)
• OpenAI Help Center: Create Campaigns from Product Feeds, Create Campaigns, Context Hints, Create Ads, Conversion-optimised Campaigns, Custom Audiences, Launch Campaigns, Onboarding and Policy Troubleshooting
• OpenAI Ads Policies and Advertising Terms
• Razorpay Newsroom, 8 October 2026; MarketingTech News, May 2026
• Practitioner guides on ChatGPT Ads for Indian D2C and Shopify merchants, Sept to Oct 2026
• Advertising on ChatGPT: A Working Guide for Indian Advertisers, September 2026

Porcellia is an independent agency. This guide is not affiliated with or endorsed by OpenAI.