If you are looking for a ChatGPT Ads Manager tutorial that does not invent CPMs and does not promise an early-mover lift, this is one. On May 5, 2026, OpenAI announced a beta self-serve Ads Manager with CPC bidding for ChatGPT ads. Your first campaign is a controlled test, not proof of channel quality.
What ChatGPT Ads Manager actually is right now
The product picture is narrower than most write-ups suggest. OpenAI’s testing post from February 9, 2026 introduced ads to logged-in adult users on the Free and Go tiers only — Plus, Pro, Business, Enterprise and Education tiers were excluded. The May 5, 2026 announcement layered three things on top of that: a beta self-serve buying portal, CPC bidding alongside the existing CPM option, and measurement primitives (Conversions API and pixel-based events). That is the verified surface area; everything else circulating online is interpretation.
Three boundaries to keep in front of you before you spend a dollar:
- It is still labelled beta. The self-serve flow is rolling out — OpenAI’s own framing is “easier for companies of all sizes, from SMBs and startups to global brands”. That is not the same as “open to everyone everywhere”.
- The minimum-spend picture changed but is not “zero”. Axios reported that advertisers previously needed at least $50,000 to test ChatGPT Ads, and that threshold was being removed with the May 2026 self-serve rollout. “Removed” is the verified word; “no minimum” is not.
- Ads stay separate from ChatGPT answers and from user conversations. OpenAI states advertisers do not get access to individual conversations and receive aggregated performance insights. That privacy boundary is a real product constraint, not a wish list.
What to prepare before you open the campaign builder
Almost every avoidable mistake in a first ChatGPT Ads campaign happens before login. A short pre-flight reduces the surface area meaningfully.
1. The landing page
Have one URL per ad variant, each with a single dominant call to action, fast load and stable mobile rendering. If the page redirects, test the final URL — a tracking template that breaks the chain is the most common silent failure in any new platform.
2. The offer and the message
Decide what specifically you want a click or impression to mean. If you can’t write the conversion event in one sentence (“Submitted a demo request for the GEO audit”), don’t launch yet.
3. UTM naming and conversion event
Lock the UTM schema and the conversion definition in your analytics tool first. For most stacks that is GA4 plus a server-side endpoint; if you have BigQuery, you can later export GA4 raw events to BigQuery for SQL-grade analysis. Doing this after launch costs you a comparable baseline.
4. Privacy and consent review
Whatever you wire into Conversions API or your pixel must respect your existing consent stack. OpenAI’s privacy line is firm: aggregated insights, no individual conversations. Your responsibility is the half you own — what events you fire, with what identifiers, under what consent state.
5. Creative variants and budget guardrails
Have at least two distinct messages and at least two visual treatments before you launch. Set a daily or campaign-level cap that you would be comfortable losing if the campaign ended overnight. Do not “let it learn” without a stop number on the wall.
Step-by-step launch workflow
The actual click-path inside the portal will change as the beta evolves, so this stays at the level of the decision behind each step, not screenshots of UI labels.
Step 1 — Advertiser account and payment
OpenAI’s framing of the self-serve flow is that businesses register as advertisers, add payment information, set budgets, bids and pacing, upload ads, launch and manage campaigns, and view performance in the portal. Treat account approval as a step that may not be instantaneous and plan a buffer.
Step 2 — Campaign objective
OpenAI surfaces two objective shapes — Reach (bought on CPM) and Clicks (bought on CPC). Choose the one that matches the page you’ve prepared. If your conversion is one click away (form, signup, install), Clicks is the more honest signal. If it’s a brand or awareness moment, Reach is more honest than pretending a soft signal is a click.
Step 3 — Buying model and bids
From OpenAI’s Help Center, the current defaults are concrete: CPM campaigns have a default max bid of $60 CPM, and CPC campaigns have a recommended starting max bid of $3–$5 USD per click. Two things to understand about these numbers: they are platform defaults and recommendations from OpenAI, not benchmarks of what other advertisers are paying. Use them as a starting position; revisit after you have your own delivery data.
Step 4 — Creative upload and destination
Upload your variants, attach the prepared destination URL, and confirm the tracking template renders the final URL correctly. Open the rendered URL in a clean session — different network, different device — and verify both the page and the conversion event fire.
Step 5 — Measurement wiring
OpenAI says Conversions API and pixel-based measurement are available to help advertisers understand post-ad actions such as purchase, lead or sign-up. Pick one as primary; running both without a deduplication plan creates noisy reporting. Server-side via Conversions API is typically more durable; pixel is faster to install if your team owns the front end but not the back end.
Step 6 — Final launch checklist
Before clicking launch: every variant points to a working URL; the conversion event fires from a real session; daily and lifetime caps are set; reporting access is granted to the right people; you’ve agreed in writing what would make you pause the campaign.
A safe first-test structure
You don’t need a benchmark to design a first test; you need a clear unknown. Pick one thing the campaign should resolve.
Assume a small first test with these parameters (this is an explicit hypothetical scenario, not an OpenAI recommendation or industry benchmark):
- Budget: a sum you are comfortable losing entirely while learning the platform.
- Variants: 2 messages × 2 audiences = 4 ad sets, with the same destination per pair so the variable is the message.
- Window: at least two weeks, so weekly seasonality is in the data.
- Success definition: agreed before launch, written down, signed off by the person who owns the budget.
Resist the urge to swap variants mid-flight. Half the value of the test is the discipline of letting one cycle complete.
What to check at each setup area (table)
The table below is the checklist I use the night before a new-platform launch.
| Setup area | What to check | What can go wrong | Safe action before launch |
|---|---|---|---|
| Advertiser account | Business profile, billing, user access | Approval delays; wrong user has spend permission | Add the launch window with buffer; restrict spend role to one person until first review |
| Buying model | CPM vs CPC matches the page you built | Reach objective forced onto a conversion page that doesn’t measure reach value | Decide on paper before opening the portal; document the reason |
| Bid setting | Default vs recommended starting bid (CPM $60 max, CPC $3–$5 start) understood as defaults, not benchmarks | Treating OpenAI defaults as fair-market price for your account | Set conservative caps; revisit after first complete cycle, not first day |
| Tracking | UTM, destination, conversion event in one clean session | Tracking template breaks the redirect; pixel double-fires | Test in incognito; if Conversions API + pixel both fire, set deduplication |
| Creative variants | 2 messages × 2 visuals minimum, distinct enough to read | Variants too similar — no learning from the spend | Pre-launch sanity check with someone outside the project |
| Caps and stops | Daily, lifetime, pause threshold | “Let it learn” without a written stop number | Stop number lives in the campaign brief, not in someone’s head |
Measurement and reporting without false precision
The portal will show clicks, impressions and any conversions you have wired. Two principles keep that from misleading you in the first weeks.
- Do not compare raw CPC or CPA against Google Ads in week one. CPC defaults on ChatGPT Ads are OpenAI’s recommendations, not auction-clearing prices in your category. Comparing platforms before you have at least one full delivery cycle is a guaranteed wrong conclusion.
- Use aggregated insights for what they are. OpenAI’s own framing is that advertisers receive aggregated performance insights and do not get access to individual conversations. That is good for privacy, and it is also why your downstream behaviour data (GA4, your CRM, your server logs) is the source of truth on what the click actually became.
For higher-fidelity behaviour analysis, your existing analytics stack carries more weight than the in-portal view. If you already export GA4 to BigQuery, that’s where post-click attribution conversations happen — not in the ad platform UI.
What not to conclude after the first campaign
The single most common mistake at this stage is over-reading early signal. A first-week ChatGPT Ads campaign on a beta self-serve product is one observation, in a moving market, on a single creative set.
- Do not extrapolate “channel performance” from one campaign.
- Do not scale spend without conversion validation from outside the ad platform.
- Do not declare a winning creative on a sample size that wouldn’t pass a basic significance check.
- Do not assume the bid defaults you started on are the bids you should still be running in week three.
- Do not present aggregated insights to leadership as proof of intent quality; the privacy boundary cuts both ways.
If you want a second pair of eyes on the test design, the launch wiring or the measurement plan, that is exactly the work my marketing audit and consulting engagement is built for — we walk the campaign brief, the conversion plumbing and the read-out before money moves. Longer-form notes on AI ad platforms live in Insights, and the prompts and skills I lean on during these reviews are collected in AI Lab.
Frequently Asked Questions
Is ChatGPT Ads Manager available to every business today?
No. OpenAI’s own framing on May 5, 2026 was a beta self-serve rollout that makes ads “easier for companies of all sizes, from SMBs and startups to global brands”. That is a directional rollout, not a guarantee that any specific account anywhere can transact today.
Is there a minimum spend on the new self-serve flow?
The verifiable statement is that the previous $50,000 minimum was being removed with the May 2026 self-serve rollout (Axios). “Removed” is not the same as “zero floor at every scale of access” — treat that as an open question until OpenAI’s own documentation confirms it for your account type.
Should I run CPM or CPC for a first campaign?
Pick the buying model that matches what the destination page actually measures. If your page has a one-click conversion, CPC is the more honest signal. If you’re measuring brand reach against an awareness page, CPM is more honest than pretending the campaign is direct-response.
Can the ads platform see individual ChatGPT conversations?
No. OpenAI states advertisers receive aggregated performance insights and do not have access to individual conversations. The privacy boundary is built into the product and is also the reason your own analytics stack — not the in-portal view — is the better source of truth on downstream behaviour.
How long should I wait before judging the campaign?
At least one complete delivery cycle including a weekend, with reporting numbers cross-checked against your own analytics. One week is rarely enough to read CPC, CPA or any cohort signal from a brand-new beta platform.