A small store should expect to pay at least $25 a day per ChatGPT ads campaign, with OpenAI suggesting a starting maximum bid of $3 to $5 per click. You can bid per click or per thousand views. A sensible first test is about $400 over sixteen days, with written rules for when to stop.
The catch is arithmetic, not access. A $4 click only pays for itself if enough visitors buy, and most small-store margins cannot carry that, so treat the first test as a learning purchase. It buys roughly a hundred visits, enough to see whether shoppers on ChatGPT behave like your other traffic, not enough to prove profit.
What OpenAI publishes and what only advertisers report
Separate two kinds of figure before you plan anything. Some numbers come from OpenAI's help centre: the $25 minimum daily budget per campaign, the three campaign objectives, the recommended $3 to $5 starting maximum bid for clicks and the relevance-weighted, second-price auction. Other numbers come from agencies and marketers describing what they saw. OpenAI has not published a rate card, and its own advertiser FAQ says the platform does not yet have performance benchmarks across advertisers.
The self-serve Ads Manager is still in beta, and the pricing moved quickly. OpenAI describes a focused pilot from February to April; trade press reported that it sold impressions only, at $60 per thousand, to advertisers who committed a large sum up front. Click bidding arrived later, and self-serve opened with a $25 daily floor instead of a commitment. Since then the spread has been wide: a September 2026 round-up of advertiser results in Search Engine Journal found clicks reported from under $3 to more than $20 depending on market and sector, and one business-to-business campaign averaging about $64 per thousand views. Treat those as other people's results until your own account shows you a real price.
Your account is the only place a price becomes real. The auction is relevance-weighted and second-price, which means a higher bid does not guarantee a win and you usually pay less than your maximum. Your report shows what each click or thousand views cost on the days you actually ran.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowCPC or CPM: the sum that decides it
CPC (cost per click) means you pay only when someone taps through to your store. CPM (cost per mille, a thousand views) means you pay for being shown, whether or not anyone clicks. For a store with no history on this channel, the risk sits in different places, and one short sum shows why.
Say an impression-priced campaign clears at $25 per thousand views, and say 0.5% of viewers click (this click rate is an assumption for illustration, since OpenAI publishes no benchmark). A thousand views produce five clicks, so each click costs $5. If the click rate is 0.25%, the same $25 buys 2.5 clicks and each costs $10. At the $60 per thousand reported for the pilot, 0.5% gives $12 a click.
| Bid style | You pay for | Your risk | Fits a small store when |
|---|---|---|---|
| Clicks (CPC) | Each valid click | Bid too low and the ad barely shows | You want a ceiling on cost per visit |
| Views (CPM) | Every thousand views | A weak ad gets seen but not clicked, so the click costs balloon | You are testing brand reach with a strong image |
| Conversions | Clicks or views, steered towards a sale event | Needs tracking in place and enough sales for the system to learn from | You already have some sales events recorded |
For a first test I would choose clicks. You know the most you will pay for a visitor before you start, and the report tells you quickly whether the visitors are worth it. OpenAI says plainly that advertisers never pay per conversion, even when the system is optimising towards one, so a conversion bid style is a steering choice, not a pay-on-results deal.
Break-even cost per click for three kinds of shop
Before you set a daily budget, work out the most a click can cost. The formula: average order value times gross margin gives the margin per order; multiply by your conversion rate (the share of visitors who buy) to get the margin each visit is worth. That figure is your break-even click. All the inputs below are invented for illustration, so replace them with your own.
| Illustrative shop | Average order | Margin | Margin per order | Conversion rate | Break-even click |
|---|---|---|---|---|---|
| Online gift shop, candle sets | $48 | 55% | $26.40 | 2.5% | $0.66 |
| Garden centre selling raised-bed kits | $140 | 45% | $63.00 | 3% | $1.89 |
| Sports club kit store, bulk team orders | $420 | 40% | $168.00 | 1.5% | $2.52 |
Look at the right-hand column against a suggested starting bid of $3 to $5. None of the three breaks even on first-order margin at those prices. That does not make the test pointless; it changes its purpose. The gift shop would need repeat buyers worth several orders each, and the sports club would need a conversion rate of about 2.4% or better to break even at a $4 click. If your numbers look like the first row, ChatGPT ads are probably a poor first channel and you should say so out loud before spending.
One adjustment to the maths is fair: first orders are not the whole story. If a customer buys again, the margin per visitor is higher. Use a repeat rate only if you have it from your own orders, say that 30% of buyers order again within six months, and write the assumption beside the result so nobody mistakes it for a fact.
A sixteen-day, $400 test for an online gift shop
Here is the main worked example, using the first shop above. It sells candle sets, average order $48, margin $26.40. The owner wants to learn three things: what a click really costs, whether visitors from ChatGPT behave like visitors from elsewhere, and whether a second round is justified.
One campaign with the Clicks objective, a $25 daily budget, a maximum bid of $4 and sixteen days, so about $400. New ad groups default to a bid strategy called Maximize results, which sets bids for you, so switch the ad group to Manual: Max bid to hold the $4 ceiling. OpenAI treats a daily budget as an average over a Sunday-to-Saturday budget week: one day may spend up to twice the figure and the week up to seven times it, so a $48 day is normal and a full week still caps out at $175.
At a $4 average click that budget buys about 100 clicks. At the shop's 2.5% conversion rate, that is two or three orders, about $96 to $144 in sales and roughly $66 in margin against $400 in spend. The owner already knows the test will lose money. The real output is a column of facts: actual cost per click, actual click-through rate, bounce rate, add-to-basket rate.
- Days 1 to 3: check delivery only. If the ad shows fewer than a dozen times a day, the bid is too low or the ad was rejected; do not touch the creative yet.
- Days 4 to 9: compare cost per click with the $4 bid. If clicks clear at $3, the budget stretches to about 130 visits.
- Days 10 to 16: read store analytics for the ChatGPT-tagged visitors: pages per visit, basket adds, orders.
Day 17 is a decision meeting with three outcomes, which the stop rules in the next section define in advance.
Stop rules to write down before the first click
People rarely stop a test they have emotionally invested in, so decide the rules while the budget is still unspent. Mine are conservative and they are my own, not OpenAI's, with limits set from the shop's own break-even rather than from anyone else's benchmark.
- Hard cap. Either give the campaign a campaign-total budget of $400, which OpenAI calls the stricter control on total spend, or keep the $25 daily budget and set an end date sixteen days out. On a first campaign the date fields may only appear after publishing, so add the end date straight after launch. No top-ups during the test, whatever day six looks like.
- Delivery rule. If fewer than 200 views arrive in the first 72 hours, raise the bid once, by $1. If delivery is still thin after another 48 hours, stop and note that a $5 bid did not win auctions.
- Price rule. If you switch to Maximize results or to views at any point, check the average click cost daily and pause if it runs above $6, one and a half times the $4 ceiling, for three days running. Without a manual ceiling the system aims for volume, and OpenAI says Maximize results does not guarantee any cost-per-click target.
- Quality rule. If 40 clicks have arrived and fewer than 20 visitors viewed a second page, the traffic is not shopping. Pause and fix the landing page or the ad's promise before spending more.
- Win rule. A second round needs at least two orders, or a basket-add rate within a third of your other channels. One order is noise.
- Review date. Put it in the calendar now: day 17, 15 minutes, one person.
The numbers in rules 2 to 5 should shift with your own store. A shop with a $600 average order sees few orders in any 100 clicks, so its win rule should look at enquiries or basket adds, not sales.
Writing the ad with ChatGPT and catching its habits
You can use the assistant itself to draft ad text, as long as you check what it gives you. Here is a prompt that gives it your facts and your limits, then a sample of what comes back.
You are drafting a short ad for a ChatGPT ads campaign for a small online gift shop.
Product: hand-poured soy candle gift set, three scents, ships in 2 working days.
Price: $48. Free returns for 30 days.
Landing page: the candle sets collection.
Write 4 title and copy pairs. Titles 24 characters or fewer, copy 48 characters or fewer.
Rules: no health claims, no "best" or "No. 1", no claims about burn time I haven't given you.
Say what the product is, who it suits and what happens after a click.An illustrative reply might read: "Soy candle gift sets / Hand-poured, 3 scents, ships in 2 days." That passes the rules and fits the lengths. A second pair might say "The best gift for stressed friends", which breaks two: "best" and an implied wellbeing claim. Delete it, do not edit it into shape, because the limit exists to keep claims honest.
Check each line against OpenAI's ad policies, which are the final word. The lengths in the prompt follow OpenAI's recommended ranges (16 to 24 characters for a title, 32 to 48 for the copy); the hard maximums are 50 and 100, but short lines suit an ad unit that sits below an answer. Confirm the limits Ads Manager shows on the day, as the beta keeps changing.
The landing page matters as much as the bid
A click that arrives at a store's home page and finds nothing about what the person asked costs the same as one that lands on exactly the right product. Because you pay per visit, a weak landing page quietly doubles your cost per order. Here is a before and after for the garden centre's raised-bed kit, with every figure illustrative.
| Element | Before | After |
|---|---|---|
| Landing page | Shop home page | Raised-bed kits collection, filtered to the 120 cm size the ad mentions |
| First screen | Seasonal banner, no price | Kit photo, price, delivery date, "fits a small patio" line |
| Clicks in a test week | 32 | 32 |
| Visitors who viewed a second page | 9 | 19 |
| Orders | 0 | 1 |
The ad did not change, and neither did the bid. The only difference was where the click landed, and visitors who looked beyond the first page roughly doubled. One order is not proof, but the second-page figure is exactly what the quality rule above is built to catch.
A test log you can fill in each morning
Keep one row per day and fill it in each morning for the day before: OpenAI says spend figures can lag by seven to eight hours, while views and clicks refresh about every 15 minutes, so a same-day spend of $0 does not mean nothing was charged. Copy the columns below into a spreadsheet; the sample rows are invented to show how a first week can look when things are drifting.
| Day | Spend | Views | Clicks | Cost per click | Orders | Note |
|---|---|---|---|---|---|---|
| 1 | $11 | 240 | 3 | $3.67 | 0 | Ad approved at noon, half a day of delivery |
| 2 | $24 | 610 | 6 | $4.00 | 0 | At the bid ceiling, so the ad is winning only some auctions |
| 3 | $47 | 980 | 12 | $3.92 | 0 | Nearly double the daily budget, allowed under the weekly average |
| 4 | $18 | 420 | 5 | $3.60 | 1 | First order, a repeat buyer from the mailing list |
Day 4's order deserves a caution. A customer who already knew the shop was probably going to buy anyway, so the log needs one more column in practice: new or existing customer. Match order emails to your customer list at the end of the test and count only the new buyers when you judge the ads.
The log also shows why a four-day total of $100 is no basis for a decision. Four days is a view of the auction at a particular moment, and the early days are often skewed by the review delay and by the system working out who responds. Make judgements at the end of the planned period, and only change one thing at a time, so you know which change moved the result.
Where the tracking goes wrong
OpenAI supports a conversion API, a tracking pixel and your own UTM parameters, with aggregated reporting in Ads Manager. It reports views, clicks and conversions, and it never shares individual conversations. Put UTM tags on every ad link so your store analytics can name the source, and set up the conversion event before day one. The guide on setting up conversion tracking before an AI spends your budget covers the steps.
One quirk catches owners out. OpenAI's ad-review crawler visits submitted landing pages with your tracking parameters attached, so your analytics may log a few visits tagged as ChatGPT ads around approval time. OpenAI says these are not billable clicks. Leave them out of the test log, or your first day's figures will look better than they were.
A realistic mistake: the shop's site blocked unknown crawlers through a security plugin. OpenAI says its OAI-AdsBot crawler must be able to reach the landing page for ad review, and it recommends allowing OAI-SearchBot too. A firewall or bot-protection service can block the crawler even when the robots.txt file allows it. The ads sat unapproved for four days, and the owner thought the account was broken. Test the landing page from outside, and read the rejection reason in Ads Manager before you change anything else.
The reporting gap that matters most is assisted sales. Someone sees the ad, leaves, and buys from a branded search tomorrow. You will not see that in click reports, so compare total orders on test days against the previous fortnight too, and expect the answer to be fuzzy.
Choosing the right product for the first test
Pick one product with a clear answer to a question people already ask the assistant. A garden centre has a natural one: "what fits a north-facing balcony?" The ad leads to the planters and trailing plants that actually do. A product that needs no explanation, such as a plain mug, gives the ad little to say.
Other points to check before you commit:
- Audience fit. Ads show only to Free and Go users, so if your buyers are professionals who pay for ChatGPT at work, you may be fishing in the wrong pool. Compare plans in the ChatGPT subscription comparison table.
- Policy fit. No adult content, dating, alcohol, tobacco, gambling, recreational drugs, political content or counterfeit goods. Individual job and housing listings are prohibited too.
- Eligibility. The business entity you bill must sit in one of the countries OpenAI lists; read the current list in Ads Manager, and note that new accounts may have restrictions at first.
- Capacity. If the ad works, can you ship? A stock-out wastes budget; the orders, stock and returns tutorial helps here.
What to do on day 17
The review is three questions. What did a click actually cost, compared with the $4 bid? What did visitors do, compared with your other traffic? What would the numbers need to be for the channel to pay?
Fill the answers into a small table and keep it with the account. An illustrative result: 96 clicks, $3.60 average click cost, $346 spent, 2 orders, $96 in sales, 41 second-page views. Decision by the win rule: two orders and a basket-add rate close to other channels, so a second round is allowed, with a smaller budget and a new product, because the gift shop's margin still cannot carry $3.60 clicks. The honest option beside it is to stop and put the $400 elsewhere, and a result like that is perfectly legitimate.
New advertisers have seen an offer of $500 in ad credit when they spend $500, shown on OpenAI's sign-up page with an end date of 31 December and no year. If the offer is live when you start, it changes the sum, because a $500 spend turns into $1,000 of delivery, but you would need to spend $500 first. Do not size the test around a promotion that may have changed.
For a wider view of what to expect before judging any AI-run campaign, read how much to spend on AI-run ads before judging results, and for the buying steps, how to buy ads on ChatGPT for a small business. If the numbers point to Meta or Google instead, smart campaigns or Performance Max covers that choice.
Questions small stores ask before spending
Is there a minimum spend to start ChatGPT ads?
OpenAI sets the floor at campaign level: a daily budget must be at least $25 for an account billed in dollars, and other billing currencies have their own minimums. Two campaigns on daily budgets therefore need at least $50 a day between them. Check the figure in Ads Manager before you commit, as the platform is still in beta.
Do ChatGPT ads show to people on paid plans?
No. OpenAI says ads appear for Free and Go users only. Plus, Pro, Business, Enterprise and Edu accounts see none, and neither do Temporary Chats or accounts identified as under 18. That narrows your audience to people on the lower tiers, which may suit a shop selling to consumers more than a firm selling to buyers at work.
Can I pay per sale instead of per click?
No. You can choose to be billed per impression or per valid click, and you can optimise delivery towards a conversion event, but OpenAI states that advertisers never pay per conversion. A sale only tells the system what to aim for. You still pay for the clicks or views that were needed to find it.
What if my product category is restricted?
OpenAI's ad policies disallow adult content, dating, alcohol, tobacco, gambling, recreational drugs, political content and counterfeit goods. Finance, health and legal ads face extra restrictions. If your products sit near one of those lines, read the policy page and ask OpenAI's sales team before you build a budget around it.
Further reads
- How to Advertise on ChatGPT: Approval, Budget and Reporting for Agencies — Approval, reporting and account structure from an agency point of view.
- Best AI Tools for a Small Ecommerce Store: 7 Worth Paying For — Where else a small store's AI budget could go first.
- ChatGPT Ads for Law Firms: Can You Advertise, and Is It Allowed? — An example of how a restricted category is treated.
- AI for Ecommerce Analytics: Shopify Reports, GA4 or ChatGPT First? — Shopify reports, GA4 or a spreadsheet queried with ChatGPT or Claude: what each shows a small store, what it costs, and the order to set them up.
- Are Etsy Ads Worth It for a Small Shop? Judge Them on Your Stats — Judge Etsy Ads on your own numbers: break-even return per listing, pause thresholds and a daily budget, with ChatGPT or Claude doing the sums.
- AI Tools for Ecommerce Websites: What a 50-Orders-a-Month Store Needs — A need-versus-skip checklist for a store shipping about 50 orders a month, with free tiers, break-even sums and the volume at which each skipped tool pays.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help centre articles on ads in ChatGPT, ads basics, creating and launching campaigns, Maximize results, the advertiser FAQ and crawler access; OpenAI's ad policies; the ads.openai.com sign-up page; OpenAI's ads developer docs; advertiser results reported by Search Engine Journal and PPC Land (labelled as reports, not OpenAI figures).