Your ChatGPT ads test plan: Aldergrove Furniture Co

• Your website with a page people land on from the ad, and a page they see once they've bought.
• Admin access to your store (Shopify), or someone who has it.
• Business details: your business's legal name, ABN, address and a payment card. Billing in Australian dollars is what puts the minimum at A$25 a day.
• A privacy policy that mentions OpenAI. The OpenAI Pixel sends details from your site to OpenAI, a US company (section 3.3). Update your policy before you switch it on.
• Your break-even numbers from section 1, written down.

1. Go to ads.openai.com and sign in with an OpenAI account (the one you use for ChatGPT is fine), then create your business's advertiser account. Ads Manager is still in beta.
2. Choose Australia, Australian dollars and your time zone. Yours is (GMT+11:00) Australia/Melbourne, from your postcode: pick the entry that names Melbourne. OpenAI doesn't let you change any of the three once the account exists, and the time zone lines up your budget days with your own records.
3. Add your payment card. OpenAI bills after you spend: it charges the card when spend reaches a threshold it sets, or at the end of the month.
4. Sign up now: OpenAI's checks can take up to 7 business days. It reviews every advertiser at sign-up and may verify your business. Your test starts on Friday 6 November 2026, so they have time to finish.
5. Let OpenAI's ad crawler read your site. An ad whose landing page OpenAI can't review can't run. Allow OAI-AdsBot (and OAI-SearchBot) in your robots.txt and in any firewall, CDN or bot protection. If someone else runs your site or hosting, send them this step.
6. Switch off creative enhancements. Since Friday 30 October 2026 OpenAI has turned them on automatically, for new and existing campaigns. Left on, they let OpenAI rewrite your title and copy, translate them, resize your image and change your landing page, without you reviewing each version. A test needs one fixed ad you've checked, and you're responsible for whatever runs. Turn them off in Manage settings at account level if it's offered, and on the campaign itself every time (section 4).

• One real event for each real result. Send the sale itself from the page people see once it's done: the order confirmation page. Never count page views, clicks or adds to cart as a sale.
• Browser and server, one shared ID. In the visitor's browser, the OpenAI Pixel sends the events. A server connection sends them from your store or website, and catches the sales that browsers and cookie banners block. If you use both, give each event the same ID, such as the order number, so OpenAI counts it once.
• A reload can count twice. Someone who refreshes the confirmation page, or comes back to it from an email, can fire the event again. The same shared ID stops that, so use it even if you only send from the browser.

For your store:

• Send the sale as order_created, with the order total and currency, and the order number as the event ID. OpenAI wants amounts in cents: your A$395 Oak bedside is sent as 39500.
• On Shopify: OpenAI's own app, ChatGPT Ads for Shopify, isn't available to Australian stores yet: on 2 Nov 2026 its Shopify App Store listing said "This app isn't compatible with your store" for Australian stores, although OpenAI said in September it would reach ad markets outside the US. This plan doesn't depend on it: add the OpenAI Pixel as a Shopify custom pixel (below). If the app appears for your store later, don't run both: each sale would count twice.
• Using the Conversions API as well? It must send the same order number as its id, or each sale can count twice.

Shopify: the OpenAI Pixel as a custom pixel

These steps follow Shopify's help page "Managing custom pixels". If you have a developer, give them this section.

1. In Shopify admin, go to Settings > Customer events.
2. Give the pixel a name, such as "OpenAI Pixel", then click Add pixel.
3. Customer privacy: set Permission to Required and choose marketing, so the pixel waits for each shopper's consent.
4. Paste the code into the Code window. Shopify takes JavaScript only, so copy the OpenAI Pixel code from Ads Manager (Conversions, your data source) without its <script> tags, then add the purchase event below it:

analytics.subscribe('checkout_completed', (event) => {
  const c = event.data.checkout;
  oaiq('measure', 'order_created', {
    type: 'contents',
    amount: Math.round(Number(c.totalPrice?.amount || 0) * 100),
    currency: c.currencyCode
  }, { event_id: String(c.order?.id || '') });
});

5. Click Save, then Connect pixel.
6. Remove any other OpenAI Pixel code from your theme, so each sale is sent once.
7. Check it on your first real order: your developer can switch on the Pixel's debug option and watch the browser console, then switch it off. Debug mode doesn't keep test events out of your reports, so don't place test orders or start test checkouts to check it.

Shopify fires checkout_completed once per order, on the thank-you page (or on the first upsell page, if you use one).

Privacy first. The OpenAI Pixel automatically hashes customer details it finds in your forms, such as an email address, and sends them to OpenAI with each event. Before you switch it on, make sure your privacy policy says you share website and order details with advertising platforms, including OpenAI, a US company.

Keep tests out of your results. Test the Conversions API with its validate-only option, which checks events without saving them. Pixel debug mode doesn't keep test events out of your reports, so don't rely on it to test with real orders.

Don't place a test order or start a test checkout, ever: on your live store either one also fires your other ad platforms' live pixels and your email platform's events. Check the tag with the method above, then confirm the sale arrives on your first real order. Never test by clicking your own ads: you pay for the click, and it muddies the numbers.

Tag every ad link, so your own reports can see these visits without relying on OpenAI. Add these to the end of each landing page link:

?utm_source=chatgpt&utm_medium=cpc&utm_campaign=aldergrove-furniture-co-chatgpt-test-nov26

The campaign tag is this plan's campaign name in lowercase, with hyphens for spaces: give any later campaign its own. If OpenAI also adds its own tracking code to links, leave that on: the UTM tags are for your reports, not theirs.

Where you'll see them:

• Shopify: Analytics, then Reports. Open a sessions or sales report and break it down by UTM source and campaign.
• Google Analytics 4: Reports, then Acquisition, then Traffic acquisition. Set the table to "Session source / medium".

Your results live in Shopify. That's where the test is judged.

For this test, every ad links to your page for Oak bedside table, with its tags: https://aldergrove.example/products/oak-bedside-table?utm_source=chatgpt&utm_medium=cpc&utm_campaign=aldergrove-furniture-co-chatgpt-test-nov26. OpenAI also adds its own click reference (oppref) to each link. If your site redirects visitors, make sure the full link survives, tags and all.

Before the test starts, note your usual week in Shopify, from the last four weeks before the test, leaving out any sale weeks: total sales, number of orders, and total ad spend on every platform. Total sales ÷ total ad spend is your usual MER. Take nothing out of it: section 8's store check reads your store's total sales against this whole figure, and once you've written it down it replaces the A$1,400 from your answers. Section 8 compares the test weeks with this.

Once a week, check OpenAI's count against your own. For the same dates, put OpenAI's conversions beside the orders that Shopify shows from these UTM-tagged visits.

• Expect some gap. OpenAI counts by its own attribution rules, and Shopify by its own.
• More conversions than real results? Look for a doubled event: the same sale sent twice, or a confirmation page that fires on reload. Fix it before you judge anything (section 11).
• Real results, but none in OpenAI? The event isn't firing, or a cookie banner is blocking it (section 11).

One campaign, billed per click, with a fixed total budget. Set it up in one sitting, then leave it alone until your first weekly check.

1. Create one campaign in OpenAI Ads Manager with the Clicks objective, billed per click. Leave the Conversions objective for later: it needs a purchase event that's already recording, and you can't change a campaign's objective once it exists.
2. Name it Aldergrove Furniture | ChatGPT test | Nov 2026.
3. Budget: a campaign total of A$750, not a daily budget. OpenAI calls the campaign-total budget the stricter control on total spend. A daily budget is only an average over the week, and one day can spend up to twice it. Don't switch the total to a daily budget later: OpenAI won't let you switch back.
4. Dates: Friday 6 November 2026 to Thursday 19 November 2026 (14 days), judged on Saturday 21 November 2026. Section 1 says why these dates. Start at 12:00 am on a weekday, so the first day isn't a part day. If Ads Manager doesn't show start and end dates on your first campaign, don't publish to find them: OpenAI warns against it. Save it as a draft if Ads Manager offers that, and set the dates before you publish.
5. Maximum bid: A$6.39 a click, set on the ad group. Never higher, even if delivery is slow. That's your break-even line (section 2.2). Enter it in Australian dollars: bids use your account's currency. Day-7 delivery check: if by then it has spent less than about A$94, a quarter of its budget to date, ChatGPT won't sell you clicks at your bid yet. Pause it and try again next month.
6. Location: Australia.
7. Creative enhancements: off. Open Manage settings on the campaign and switch them off (section 3.2). OpenAI has switched them on by default for new and existing campaigns since 30 October 2026, so check the setting is still off after you publish.
8. Context hints: paste the hints from section 6.
9. Ads: add the three ads from section 5, each linking to https://aldergrove.example/products/oak-bedside-table?utm_source=chatgpt&utm_medium=cpc&utm_campaign=aldergrove-furniture-co-chatgpt-test-nov26, your page for Oak bedside table with the tags from section 3.5.
10. Attribution: a 7-day click window, with the view window off. That keeps OpenAI's count closest to what your UTM report can see.
11. Optional: leave out past customers. OpenAI lets you upload a customer list, but it can't target one: that needs 25,000 matched people, so a small business's list can only exclude. Do it only if your privacy policy says you share customer details with advertising platforms, including OpenAI, a US company, and only with customers who agreed to marketing. Otherwise skip it: the test works without it.
12. Launch it. Check every setting against this list once more first. Then write the start date, the total and the stop line in your weekly log.

Your stop rule: Pause the campaign as soon as it has spent A$426 (twice your break-even) with no sale in your own records. Check its spend once a day until the first one arrives. Otherwise it stops by itself at A$750, its campaign total. On day 7, pause it too if it has spent under about A$94, a quarter of its budget to date: ChatGPT isn't selling clicks at your bid yet (section 4, step 5). Write it down now, before you launch, and follow it as written.

Delivery may be thin. Your maximum bid sits inside OpenAI's suggested range, so you may win fewer auctions. Slow spend is the cap working, not a fault: don't raise the bid above A$6.39. The day-7 delivery check (step 5) and the first row of your decision table (section 8) say when to stop.

• The landing page matches the ad: the same product, the same price, the same terms.
• The price on the page itself, with the delivery cost beside it.
• The page loads fast on a phone. A ChatGPT visitor arrives mid-conversation and leaves just as fast.
• One clear next step above the fold on a phone: the Add to cart button.
• Who you are: business name, ABN, and a way to contact you.
• A privacy policy page that says how you use the details people give you when they buy.
• The OpenAI Pixel on the confirmation or thank-you page (section 3.3).

• Dimensions in cm (W × D × H) on every product, with a diagram for furniture.
• A plain product shot and a room shot for every product; every colourway; a scale reference.
• Materials and care, named exactly: solid timber or veneer, fabric composition, linen weight, and how to care for it.
• Delivery cost before checkout (a postcode calculator for bulky items), and lead times, including made-to-order.
• Assembly information and what's in the box.
• Returns in plain words, including who pays return freight on bulky items, and that faulty items are always covered (section 10).
• Swatches or samples, if you sell fabric or colour-critical pieces.
• Tip-over warnings and anchoring for tall furniture (section 10).
• Real reviews with customers' own photos.
• Buy now, pay later near the price on big pieces, if offered.
• Mobile speed: room galleries and autoplay video sized for phones.
• Business name, ABN, contact details and returns address consistent across footer, terms and policies.

• Ads must be true, and you must be able to prove them. OpenAI's ad policies say "ads must be truthful and not misleading", including about what you sell, your prices, outcomes, your affiliations and comparisons with other businesses. Exaggerated results and false endorsements are banned in every category. That's the same test the Australian Consumer Law applies to every ad (below).
• Your ad must look like an ad. OpenAI requires ads to be clearly distinct from ChatGPT itself, and can remove ads that imitate its look or voice. No images or wording that look like a ChatGPT answer or its screens.
• The whole path is reviewed. OpenAI checks your business, then the title, copy, image and landing page of every ad before it runs, and its rules apply to all of them. In OpenAI's words, "ads must be consistent end-to-end": an approved ad can't lead to a page that breaks the rules. OpenAI may also decline an ad for its own reasons.
• OpenAI must be able to read your landing page. A page its systems can't review can't run, and OpenAI says "you must allow OAI-AdsBot", its ad crawler, in your robots.txt and any firewall or bot protection (section 3.2).
• No ads near sensitive topics. OpenAI doesn't show ads in conversations about topics such as personal health, mental health, suicide and self-harm, and politics. Don't write ads or context hints that aim at those conversations.
• Describe what you sell, never the person. OpenAI's ad policies don't allow wellness claims or ads about sensitive topics, and OpenAI keeps ads out of sensitive conversations. So don't write ads or context hints that assume something about the reader: their health, money troubles, age, body, beliefs or other personal circumstances. "Solid oak dining tables, made to order" works; "Moving out after a break-up? Furnish your place with us" doesn't.
• Write for everyone. OpenAI doesn't show ads to accounts it identifies as under 18, and ads can also appear to people using ChatGPT without logging in, where they must suit all audiences. Never use content that appeals mainly to children.
• Some things can't be advertised at all. Among them: alcohol and tobacco, adult content, gambling, recreational drugs, political content, counterfeit goods, health and wellness claims, and individual job or housing listings.
• Only your own names and logos. Use trade marks you own or may use, and don't use a competitor's brand in your ad, image or context hints. Never imply OpenAI endorses you.
• If your ad is rejected, fix it and make a new ad, or use Request Review in Ads Manager if you think the reason doesn't apply. Never re-post it with small tweaks to slip through (section 11).

• The price in the ad must be the price on the page. Same product, same price, same delivery terms. If the ad mentions a sale, the sale has to be live on the page.
• Sale and "was" prices follow Australian law (below): a "was" price must be one customers actually paid, for a reasonable time before the sale.
• Show delivery costs before checkout. A free-shipping claim needs its conditions beside it, on the page.
• Images are claims. Colour, size and detail must match what you ship, AI-made images included.
• No health or wellness claims. OpenAI doesn't allow health ads outside the US, including supplements, and bans wellness claims everywhere. Never say a product treats, heals, detoxes or changes the body. That's why this plan isn't sold for products sold on a therapeutic claim.

Every online store: the price, sale, review, origin and returns rules are under "Rules for every online store", at the end of this section.

Product safety (mandatory standards; check productsafety.gov.au before you list)

• Tall furniture: the Toppling Furniture Information Standard 2024 (in force 4 May 2025) requires a tip-over and anchoring warning in your online listing (s 7), a permanent warning label on the furniture (s 9) and warnings in the instructions (s 10). It covers clothing storage units (chests of drawers, wardrobes) and bookcases 686 mm or taller, entertainment units of any height, and hall tables, display cabinets, buffets and sideboards 686 mm or taller. Resold second-hand furniture and furniture that can only be used fixed to a wall are excluded (Product Safety Australia, "Toppling furniture mandatory standard").
• Bean bags: a suffocation warning and child-resistant zips on every opening (Bean Bags Safety Standard 2014).
• Anything else you sell: search its product type on productsafety.gov.au for mandatory standards and bans before you list it.

Material claims

• "Solid timber" only when it isn't veneer or board.
• "100% linen", "French linen" and "organic cotton" only with supplier proof, and certification for "organic".
• Dimensions and colour in your ads must match the product you deliver.

Prices and sales

1. "Was" prices must be real: a price customers actually paid, for a reasonable period before the sale. There's no fixed number of days; it depends on the product and how often its price changes (ACCC, "Advertising and selling guide"). False price claims breach the Australian Consumer Law (ss 18, 29(1)(i)). An ugg retailer paid A$39,600 for "was" prices it hadn't charged (ACCC, June 2026); a bedding supplier was ordered to pay A$15 million after 58 of 74 products never sold at their strikethrough price (ACCC v Emma Sleep, Federal Court, 24 Apr 2026).
2. "Sitewide" means the whole site. A jeweller, a homewares store and an online hair and beauty retailer each paid A$19,800 over Black Friday "sitewide" sales with exclusions (ACCC, 11 Jun 2025).
3. Show delivery costs before checkout, including any minimum charge you know of (ACCC, "Advertising and selling guide", component pricing; Australian Consumer Law s 48), with "free shipping" conditions beside the claim. New drip-pricing rules and a ban on unfair trading practices start on 1 July 2027 (Competition and Consumer Amendment (Unfair Trading Practices) Act 2026). Get ready now: wherever you show a price, say whether extra charges apply and how much they are, or how they're worked out.

Reviews, origin and green claims

• No fake reviews, no hiding genuine bad ones, and disclose incentives (ACCC, "Online reviews: a guide for business and review platforms"). Paid or gifted creator posts must say so: a photo-book retailer paid A$39,600 over undisclosed paid influencer reviews (ACCC, March 2026).
• "Made in Australia" needs substantial transformation here (ACCC, "Country of origin claims and the Australian Consumer Law", 2022). Packing or relabelling imported goods here isn't enough. "Designed in Australia" is fine if true and nothing else (flags, maps, "Aussie made" wording) implies it's made here.
• No broad green words ("sustainable", "eco-friendly") unless qualified and provable (ACCC, "Making environmental claims: a guide for business", Dec 2023).

Returns, images and payment

• Never "no refunds", even on sale items: faulty goods are always covered. Honour your own change-of-mind policy (ACCC, consumer guarantees guidance).
• Images are claims, AI-made ones included: colour, size and detail must match what you ship.
• Buy now, pay later: the provider's wording, and never "interest-free" without its conditions (ASIC 25-193MR, 3 Sep 2025).

1. Still in review? OpenAI reviews every ad and its landing page before it runs, and publishes no typical time. If it's still pending after a few days, check the status note in Ads Manager.
2. Landing page blocked? OpenAI can't run an ad whose page it can't read. Check OAI-AdsBot is allowed (section 3.2).
3. Start date in the future, or the campaign paused? Check the dates and status.
4. Payment failed, or the business check isn't finished? Check billing and any messages in Ads Manager. Verification can take up to 7 business days.
5. Ad rejected? See below.
6. Bid too low to win? If your maximum bid is your break-even and it sits inside or below OpenAI's suggested range, thin delivery is the cap working. Don't raise it above A$6.39.
7. Context hints too narrow? Check they describe everyday conversations, not one exact question (section 6).

• Set a daily budget by mistake? A daily budget is an average over seven days, and one day can spend up to twice it. Switch to a campaign-total budget (section 4).
• Spend shows about 7–8 hours late, so today's figure keeps rising after you look. Check again tomorrow before you act.
• Hit your stop line? Pause the campaign now, whatever the date.

1. Wait 24–48 hours after the first sale. Conversions appear late.
2. Is the pixel on the confirmation or thank-you page, not just the home page? Re-check section 3.3.
3. Does your cookie banner hold marketing tags until people accept? Then events only arrive after they do.
4. Using the Conversions API? Check its events with the validate-only option, and that it sends the same event ID as the pixel.
5. Still nothing? Work through OpenAI's guide, "Troubleshooting conversion tracking for ChatGPT Ads", in its help centre.

1. Read the reason in Ads Manager.
2. Think the reason doesn't apply? Use Request Review or the appeal option in Ads Manager. Email ads-support@openai.com only if neither is offered, or an appeal has waited more than 7 days.
3. Otherwise fix it and make a new ad. Don't resubmit the same ad with small tweaks to slip it through.
4. Common reasons: a claim the page doesn't back up; a price or offer that isn't on the landing page; wording about health, the body, money troubles or other personal circumstances; an image that looks like part of ChatGPT; a landing page that's broken, slow or blocked to OpenAI's crawler.

• OpenAI counts more? Its count is last-touch inside its own window and can include modelled estimates. Check your attribution window is a 7-day click with the view window off (section 4).
• Your records show more? Some visitors block the pixel, and some come back later another way. Your UTM report still sees the first visit.
• Either way, your records decide (section 8).

• Your ad text or landing page changed without you? That's creative enhancements. OpenAI has switched them on automatically since 30 October 2026, for new and existing campaigns. Switch them off on the campaign (section 4), check the ad is back to the one you wrote, and note the dates in your log: those days don't count as a clean test.
• OpenAI's Shopify app shows up in your Shopify admin? Keep your custom pixel until we update your plan. Running both would count each sale twice.
• OpenAI changes a rule, a minimum or a feature during your test? Finish the test on the settings you started with, unless the change stops the campaign running. Your plan's facts are re-checked at least monthly, and we'll update your plan if a change affects it.
• OpenAI pauses ads in Australia? Nothing to do: the campaign stops spending. Your tracking and UTM tags keep working for every other channel.
• The facts date in section 1 is more than a month old? Check the sources in the appendix before you spend.

OpenAI changes its advertiser help pages often. Check the links before you spend.

• Ads Manager (OpenAI): OpenAI's self-serve tool for buying ads in ChatGPT, still in beta. Where you build the campaign, set the budget and bid, and read reports.
• Aggregated data: totals only. You see how many people clicked, never who they were or what they asked.
• Attribution window: how long after a click (or a view) OpenAI will credit your ad with a sale. You choose 7, 14 or 30 days after a click, plus an optional 1 day after a view.
• Auction: how OpenAI picks which ad to show. Relevance counts as well as your bid, and in a second-price auction you can pay less than your maximum.
• Break-even cost per click: the most a click can cost before each sale it brings costs more than it's worth. Break-even cost per sale × the share of visitors who buy.
• Break-even cost per sale: what one sale keeps after GST, product cost, shipping and payment fees, before ads: pay more than this in ads for it and that sale loses money. This plan works it out on what your ads sell.
• Break-even ROAS: the sales each A$1 of ads must bring back before the ads stop losing money.
• Campaign-total budget: one fixed amount for the whole campaign. OpenAI calls it the stricter control on total spend.
• Context hints: short descriptions of the conversations your ad fits. Not keywords: they guide where your ad might show, and don't decide it.
• Contribution margin: what you keep from an order after product cost, shipping and payment fees, before ads.
• Conversions API: sends your sales to OpenAI from your server, so fewer are lost to ad blockers and privacy settings.
• Creative enhancements: an OpenAI setting, on by default from 30 October 2026, that lets OpenAI rewrite, translate or resize your ad and change its landing page without your review. This plan has you switch it off.
• CPC (cost per click): what you pay each time someone clicks your ad.
• CPM (cost per 1,000 impressions): what you pay each time your ad is shown 1,000 times.
• Custom pixel (Shopify): code you add in Shopify's Settings > Customer events, so a tracking pixel can see checkouts. How Australian Shopify stores add the OpenAI Pixel while OpenAI's Shopify app is US-only.
• Daily budget: an average over seven days. One day can spend up to twice it.
• Event ID: a number sent with each sale, such as the order number, so it isn't counted twice.
• Last-touch attribution: the whole sale is credited to the last ad clicked (or viewed) inside the window, even if other things helped.
• Maximum bid: the most you'll pay for one click. Your plan sets it at or below your break-even cost per click.
• MER (marketing efficiency ratio): total store sales ÷ total ad spend, from your store's own numbers.
• Modelled conversions: OpenAI's estimates of sales it couldn't see directly. They can make its count higher than your own.
• OpenAI Pixel: the code on your website that tells OpenAI when someone buys after clicking your ad.
• OAI-AdsBot: OpenAI's ad crawler. It reads your landing page to review your ad, so your site must let it in.
• Order_created: the event name OpenAI uses for a sale.
• Pilot: an early, limited version. Rules, budgets and features can change at short notice.
• Stop rule: the point, written down before you start, at which you pause the test whatever the dashboard says.
• Suggested starting bid: OpenAI's suggested maximum bid for click campaigns, US$3–5 as of this plan's facts date. A starting point, not an average price.
• UTM tags: labels added to the end of your ad links, so your own store or website reports can see which visits came from which ad.
• Validate-only: a Conversions API option that checks your events are set up correctly without saving them.
• Visitor rate: out of every 100 people who visit your site, how many buy.