Advertising on AI platforms: what a Shopify brand can actually buy today, and whether it is worth testing yet
You run Meta and Google for a Shopify brand and someone told you again this week that AI is the next channel. Nobody could say which assistants sell ads, what is announced versus buyable, what is organic versus paid, or what a test would cost. This page does: every fact dated and taken from that platform's own page, and a blank cell wherever that page confirms nothing.
EDITORIAL Datadrew does not report on, manage or optimise ads on any AI platform. This is market education from a team that runs Meta and Google against Shopify orders.
Which AI assistants sell ads today, who can buy them, and what it costs to start
One row per surface, filled only from that company's own page on the date shown. Where that page confirms nothing, the cell says so. A blank beats a rumour.
VERIFIED AS OF 10 SEPTEMBER 2026 · RE-VERIFIED MONTHLY| AI surface | Is there a paid unit? | Who can buy it, and where | Cost to start and minimum | As of | Source |
|---|---|---|---|---|---|
| ChatGPT (OpenAI) | Yes, in beta. One labelled sponsored unit below a response, never inside it, shown to adults on the Free and Go plans and in some logged-out sessions; Plus, Pro, Business, Enterprise and Edu accounts see none. Chat card, one square image, no video. | Self-serve in 52 countries including the US, UK, EU and India (India live since 27 August 2026; self-serve Ads Manager reached India, Europe, the Middle East and North Africa on 31 August 2026). One API key per ad account; no manager account. | Daily minimum by billing currency: $25, £15, €15, ₹725, release-configured. CPM, CPC or conversion-optimised CPC; no CPA or ROAS target. No benchmarks, by OpenAI's own statement. | Verified 10 Sep 2026 | OpenAI FAQ; availability; create campaigns; Advertising Terms (25 Aug 2026); our dated fact box |
| Google AI Overviews and AI Mode | No separate unit. Text and Shopping ads from existing Search, Shopping and Performance Max campaigns are eligible within AI Overviews. No opt-out, no direct targeting. | Google Ads advertisers using broad match, AI Max, Performance Max, Shopping or Dynamic Search Ads. English only, in the US, Canada, Australia, India and eight other countries the help page lists. | No separate minimum; existing bids and budgets. No segmented reporting. Ads in AI Mode: Google describes them as in testing (blog.google, 21 May 2025 and 11 Feb 2026). A separate AI-Mode buy or eligibility rule: not verified as of 10 September 2026. | 10 Sep 2026 | Google Ads Help; Google blog, 21 May 2025; 11 Feb 2026 |
| Microsoft Copilot | Yes, as an automatic extension. Ads are built from existing campaign assets and shown in Copilot's responses; eligible campaigns are opted in, no opt-out, display not guaranteed. | Microsoft Advertising accounts running product ads, search ads with a logo extension, or multimedia ads. Showroom ads piloted with select clients from April 2025 and Dynamic Filters piloting in English-language markets (March 2025 post); Offer Highlights for retail in English-speaking markets via Microsoft Merchant Center, April 2026. Microsoft positions Performance Max as the route to Copilot placements. | No separate minimum; existing bids. Negative keywords apply. No Copilot-specific metrics yet, per Microsoft. | 10 Sep 2026 | Microsoft Advertising help; Microsoft blog, 5 Mar 2025; 21 Apr 2026; Performance Max |
| Amazon Rufus | Yes, as an extension of existing campaigns. Sponsored Products and Sponsored Brands prompts appear in results and on product pages and may open a Rufus dialog. | US Amazon advertisers running Sponsored Products and Sponsored Brands. Enrolled automatically. | No separate budget. Charged within existing CPC billing since US general availability on 25 March 2026. Pause controls per prompt and a prompt-level Prompts report (impressions, clicks, CTR, CPC, orders) exist in the Ads Console. | 10 Sep 2026 | Amazon Ads |
| Meta AI | not announced as of 10 September 2026 (no Meta newsroom statement found). Meta's 28 January 2026 post on AI and ads describes tools for running Facebook and Instagram campaigns, including a business assistant for advertisers, and no paid unit inside the Meta AI assistant. | Your Meta ads run where they always did: Facebook and Instagram. | Nothing to price on Meta's own pages today. | 10 Sep 2026 | Meta newsroom |
| Perplexity | An experiment, announced on Perplexity's own blog: “Ads will appear in the US to start and will be formatted as sponsored follow-up questions and paid media positioned to the side of an answer.” Current programme status and any self-serve buy: not verified as of 10 September 2026. | US to start, with named launch partners (Indeed, Whole Foods Market, Universal McCann, PMG). No self-serve sign-up verified. | Not published. “This is an experiment and formats may evolve” (Perplexity). No minimum verified. | 10 Sep 2026 | Perplexity blog (undated post) |
Floors, tiers and eligibility rules are release-configured and change; the date column is part of every fact. Google, Microsoft, Amazon and Perplexity rows are market context. Datadrew manages Meta and Google only.
- A separate buy with its own account, floor and creative: ChatGPT Ads. The one surface you decide to fund.
- An automatic extension of what you already buy: Google's AI Overviews, Microsoft Copilot, Amazon's prompts. Google and Microsoft give no opt-out and no separate reporting line today; Amazon's prompts ride on your existing Sponsored Products and Sponsored Brands budgets but can be paused one by one and carry a prompt-level report.
- Announced or absent, with nothing to buy from their own pages today: Perplexity (an advertising experiment announced, current status not verified) and Meta AI (no announcement found). That is a fact about the sources, and nothing more.
So the budget question collapses to one: does ChatGPT Ads deserve a test from your account, and would you know if it worked?
Start with the question you actually have about advertising on AI platforms
Five questions any operator asks about a new channel, in this order: should it get any budget, what should be advertised, how should it be configured, can the results be trusted, and what happens next. Coming means being written; each Coming card links to the nearest answer that exists today.
A new channel competes for the same marginal dollar as everything you already run
The newness of a surface says nothing about where your next ad dollar should go. That dollar has three homes: the Meta campaign still returning above your margin line, the Google non-brand campaign that is budget-constrained, or a test on something new. The store's economics pick, and the number that picks is marginal ROAS: what the last slice of spend returned rather than what the average did. The calculator lives on the marginal ROAS post; the eight budget states it feeds live on scaling paid ads profitably.
“You need to be able to survive, I think, a 1.5X ROAS.”Sean Frank, CEO, Ridge, on the Up Next in Commerce podcast (verified)
A channel with no benchmarks is where that survival math bites hardest: you are buying clicks at a price nobody has published, so the only defensible envelope is a fixed amount your margin can lose entirely. Never a percentage of the Meta budget, which grows with the account and turns a test into a line item nobody approved.
What is different about buying on an AI surface, taken from the one you can buy today
ChatGPT Ads as OpenAI documents it, verified 10 September 2026. One surface that exists; no generalisation about every assistant.
| Lever | Meta | ChatGPT Ads (beta) |
|---|---|---|
| Targeting | Demographics, interests, lookalikes, placements | Country, platform, uploaded audiences (no lookalikes) and free-form context hints: descriptive need-phrases rather than exact-match keywords |
| Bidding target | CPM, CPC, CPA or ROAS targets, cost caps | CPM, CPC or conversion-optimised CPC, fixed at creation (to change objective you create a new campaign). No CPA or ROAS target exists; efficiency is governed by budget and bid caps |
| Term-level report | Placement and audience breakdowns | No search-terms or conversation report exists; hint quality can be judged only by delivery and click-through rate |
| Click window | 7-day click, 1-day view by default | 30-day click; a fixed 1-day view excluded from conversions, CPA and bidding |
| Budget pacing | Daily or lifetime | Daily budget is a 7-day average: any single day up to 2×, the week up to 7× |
| Stopping | Takes effect on save | A change or cancellation can take up to one business day to take effect; fees continue for ads delivered meanwhile |
| Benchmarks | Years of them, mostly vendor-published | None. “As an early platform, ChatGPT Ads does not yet have performance benchmarks across advertisers, industries, or campaign types.” (OpenAI FAQ, verbatim, 10 September 2026) |
| Creative | Image, video, carousel, catalog | Chat card (title, body up to 100 characters, one square image) or a product-feed template. No video |
Two rows change how you operate. Stop latency first. Meta trained a generation of founders on instant control:
“Meta is the only marketing channel where the founder can take direct action in 30 seconds…”Wittelsbach, advisory content on founder burnout from Meta ads (verified)
On the surface you can buy today, a Friday-afternoon pause can still be spending on Monday, so the stop rule is a date and a budget cap set before launch. Benchmarks second. The only starting number anyone has is OpenAI's suggested $3 to $5 starting CPC, with a warning that bids much under $3 may not deliver, per Search Engine Land's reporting of that guidance as carried on the ChatGPT Ads guide (checked 3 September 2026; not published on OpenAI's help pages as of 10 September 2026). Treat it the way an experienced buyer treats any seller's suggested price:
“The defaults are revenue-optimized for the channel, not performance-optimized for the advertiser… The entity building the machine shouldn’t be the one telling you when it’s ready.”Jay Stampfl, Blackbird PPC, in Search Engine Land (verified)
Organic discovery in ChatGPT is free and already on for US Shopify stores. Paid ads are separate plumbing.
Since the week of 24 March 2026, every US-selling Shopify store with eligible products has been discoverable in ChatGPT by default through Shopify Catalog (Shopify-side facts in this section verified 3 September 2026). No fee. Checkout on your own store. Orders arrive as ordinary online-store orders tagged with a ChatGPT referral in Shopify, and Shopify's admin already reports sales per AI channel. Non-US stores have no syndication path today; from India or the UK, that door is shut.
Paid is different plumbing. There is no Shopify sales channel for ChatGPT Ads. Product-feed ads take a Google-compatible file (CSV, hosted URL or SFTP) whose items expire after 14 days unless refreshed. Your Google Shopping feed is the natural starting shape, and a stale feed on a sold-out product is the same leak it is everywhere else: ad spend on out-of-stock products.
Separate the two before you judge a test, or the test will “work” on orders you were already getting free. One more split, stated once: being cited in an AI answer is a different job from buying an ad beneath one. This page sells neither. The organic side, and what “Shopify MCP” means there, is untangled in the Shopify MCP guide.
Three labels for every number from an AI-platform test: platform-reported, Shopify-observed, blended ratio
You know this argument from the channels you run: Meta reports one ROAS, GA4 another, Shopify shows what reached the bank. A fourth voice makes it louder.
“They’re tracking everything and measuring nothing.” … “So they make budget decisions based on bad data. They over-invest in whatever platform tells the prettiest story.”Curtis Howland, DTC growth operator, curtishowland.substack.com (verified)
What OpenAI, Meta or Google says it delivered, subject to that platform's windows, matching and modelling. ChatGPT reports on a 30-day click window, Meta on seven by default. Side by side on one dashboard, the comparison is meaningless.
Orders and revenue that exist, with the referrer Shopify recorded. No causal claim. The one number you can bank; every other number is a story about it.
Shopify revenue divided by all ad spend for a scope you declare. An efficiency ratio that credits every platform with all revenue. Useful, daily, and never attribution.
The mechanics, in one line: a paid click tagged utm_source=chatgpt.com&utm_medium=cpc lands in GA4's Paid Other channel; organic citations arrive as referral or the GA4 AI Assistant channel Google added on 13 May 2026. Keep chatgpt.com as the source for both and let the medium carry the split.
If you already use Triple Whale, it shipped a ChatGPT Ads reporting connector on 28 July 2026 and is listed on OpenAI's measurement-partner page, so the platform's numbers can sit next to your others there.
Sizing a double-count is arithmetic: add up what each platform claims and set the sum beside the Shopify order count. Deciding which platform deserves the credit is a different job, and this page does not do it.
Method: blended ROAS across Meta, Google and Shopify. Limits: what blended ROAS can't tell you. Shopify as source of truth: which ROAS number you can trust. The UTM template is on the ChatGPT Ads guide.
The channel-agnostic readiness check, with every input yours
No benchmarks exist, so nothing here is borrowed. Six checks, in order; the first that fails ends the conversation until it passes.
- Unit economics survive an assumed CPC at a plausible conversion rate. Contribution per order is order value times the margin you know before ad spend; cost per order is CPC divided by conversion rate. If the second exceeds the first, no configuration rescues it, unless repeat behaviour by product covers the gap, read from your own product cohorts and never per campaign.
- A ring-fenced, fixed test budget. Small against your existing Meta and Google spend, large enough to buy clicks you can read, written down before launch with the stop rule. Never a percentage carve-out.
- One hero product or a clean feed. Current availability, a square image, a landing page the platform's crawler can read, an ad that says what the page says.
- Inventory to cover the test at the higher run rate. Days of cover on the hero product itself.
- A category the platform allows. On the surface buyable today, financial, health and legal are case-by-case and US-only; wellness copy with health claims is prohibited everywhere (OpenAI ad policy v1.5, verified 3 September 2026).
- Measurement you already trust. If Meta and Google still disagree with Shopify by an amount nobody has explained, a fourth voice will not settle it.
And one calendar rule: never launch a new channel in the middle of a sale. The sale moves conversion rate, order value and stock at once; you will not be able to tell the channel from the promotion.
A $3M GMV Shopify brand spending $60,000 a month across Meta and Google. Order value $85; the margin the founder knows before ad spend, 45%. Assume a $3.50 CPC, inside the $3 to $5 range OpenAI has suggested (reported by Search Engine Land, 3 September 2026), and a 2% conversion rate, generous for cold traffic from a new surface.
| Contribution per order ($85 × 45%) | $38.25 |
| Cost per order at $3.50 CPC and 2% conversion | $175.00 |
| Break-even CPC at 2% conversion | $0.77 |
| Break-even conversion rate at $3.50 CPC | 9.2% |
| 21-day envelope at $30 a day (about 1% of monthly spend) | $630 |
| Clicks bought, orders expected, contribution returned | 180 · 3.6 · $137.70 |
Verdict: not ready on unit economics. The envelope is fine; the store is not. A $38 contribution cannot buy a $175 order, and no hint, bid or creative closes that gap. The honest options: a hero product with contribution near $175, a repeat-driver whose 12-month product cohort covers the difference, or no test. The novelty is not a reason. The margin is. The same arithmetic, run on the Meta and Google spend that brand already has and the margins it shares, is what Drew's Budget Recommendations rest on when you ask for them.
New-channel readiness check
Any channel, any currency. Every input is yours because no benchmarks exist.
No benchmarks are used or estimated: there are none. Margins are the ones you enter. Product-level ROAS elsewhere on this site is approximate; repeat behaviour is read at product or cohort level, never per campaign or ad. Datadrew does not report on or manage ads on any AI platform. The canonical ChatGPT calculator is on the ChatGPT Ads guide.
You just ran this on assumptions. Drew works on real numbers, daily, for the two channels you already buy: Meta and Google spend reconciled against Shopify orders, platform figures labelled alongside.
Connect Shopify, Meta and Google freeWhat Datadrew does on advertising channels today, and what it does not do on AI platforms
Status words are exact here, so nothing reads as live that is not.
Diagnose Performance across Meta, Google and Shopify
Why CPA, ROAS, revenue or conversion rate moved, in one pass across Meta, Google and your Shopify orders, with platform-reported figures labelled alongside. Checks are daily, not real-time.
Budget Recommendations with the business in view
What to scale, reduce, pause or investigate on Meta and Google, with the reasoning, using the margins you share with Drew and the product-level repeat behaviour it reads from Shopify. Ask it to check stock on the products a campaign sells and it will; it does not watch for that on its own yet. Product-level ROAS is approximate; we say so.
Ads on AI platforms
Datadrew does not report on or manage advertising on AI platforms. That covers ChatGPT Ads and every other surface on the board. Drew does not host feeds for them, write hints for them or optimise them.
Your Meta, Google and Shopify data inside Claude or ChatGPT
Datadrew MCP is read-only and free to connect on every plan, including Free. The tools that read your own connected accounts work on any plan; the analyst tools whose answers Datadrew computes (cohort LTV, creative intelligence, competitor tracking) need AI Intelligence or AI Ads CoPilot. A surface for the channels you already run; no reporting on AI-platform ads.
Drew manages Meta and Google, and reconciles what they spend against your Shopify revenue into one blended efficiency number, daily, with platform-reported figures labelled alongside. It does not report on or manage advertising on AI platforms. Whatever you decide to test, the number you bank is Shopify orders — and that is the number Drew already keeps honest.
Pricing is published: a Free plan with no credit card, AI Intelligence from $99/mo, AI Ads CoPilot from $249/mo by early-access request. The free store health check runs on connected data and shows the reconciled picture first.
Advertising on AI platforms, answered
Written by Sumit Bansal, co-founder of Datadrew. Published 10 September 2026. Platform facts carry their verification date inline; the status board is re-verified monthly.
Bank the number that counts, on the channels you already buy
Connect Shopify, Meta and Google and see spend reconciled against orders, daily, with platform-reported figures labelled alongside. Free plan, no credit card required.
Book a demo