OpenAI has replaced Max Bid with Target CPA for new conversion campaigns in ChatGPT Ads and added Target CPC for new click campaigns, according to its bidding help article. Both strategies became available for new campaigns in Ads Manager on November 9, 2026. On November 18, 2026, the API followed: OpenAI now recommends the dedicated Target CPA and Target CPC API fields, and legacy Max Bid or Max CPC values used to create a campaign or change a bid type are already treated as average-cost targets, not caps. Legacy Max Bid and Max CPC controls are scheduled for deprecation on January 1, 2027.
Kopa does not run ChatGPT Ads. It is the AI Google Ads media buyer on the other side of most ecommerce stacks, where the same discipline applies: conversion tracking guardrails and targets built from margin rather than guesses. This checklist covers what to fix on the ChatGPT side before the deadline, and how to keep the two automated systems from being judged on one blended number.
ChatGPT Ads bidding change at a glance
| Update | Effective date | Default behavior | Risk for online stores | What to do |
|---|---|---|---|---|
| Target CPA replaces Max Bid | Nov 9, 2026 | New conversion campaigns bid toward an average cost per action | Bids follow whatever event you report as a conversion | Confirm the purchase event fires once per real order |
| Target CPC introduced | Nov 9, 2026 | New click campaigns bid toward an average cost per click | Cheap clicks can still be low-intent clicks | Use for tests, judge on orders |
| API target fields recommended | Nov 18, 2026 | Legacy API bid values on new campaigns or bid-type changes become average-cost targets | An old ceiling silently becomes a target | Switch scripts and connectors to the target fields now |
| Max Bid and Max CPC retire | Jan 1, 2027 | Legacy bid controls scheduled for deprecation | Existing campaigns and workflows on old fields need rework | Migrate well before the holiday freeze |
- Targets in Ads Manager
Target CPA and Target CPC on new campaigns
- API milestone
Legacy API bid values read as average-cost targets
- Your move
Verify tracking, set margin targets, update API calls
- Legacy controls retire
Max Bid and Max CPC deprecated
1. Target CPA replaces Max Bid on conversion campaigns
Under Max Bid, you told ChatGPT Ads the most you would pay in an auction. Under Target CPA, you tell it the average you want to pay per conversion, and the system adjusts each auction bid toward that goal. An average is a different promise than a ceiling, so expect individual costs to move around your number.
On new conversion campaigns, Target CPA is now the cost-target option, alongside untargeted Intelligent bidding, formerly Maximize Results, per OpenAI's help article. Max Bid is no longer the choice for new builds.
The practical shift is where your control sits. You stop tuning bids and start tuning two inputs: the conversion event the system learns from, and the target you give it. If either is wrong, the automation will optimize precisely toward the wrong outcome.
What to do this week:
- List every existing ChatGPT conversion campaign and note whether it still runs on Max Bid.
- On new conversion campaigns, choose Target CPA over untargeted Intelligent bidding when you have a margin-based number to hold.
- Write down the conversion event each campaign optimizes toward, and confirm it is a completed purchase.
- Record current cost per purchase on Max Bid campaigns so you have a baseline before you migrate.
2. Target CPC arrives for click campaigns
OpenAI also introduced Target CPC for new click campaigns. It works the same way: you set an average cost per click, and auction bids adjust toward it.
Click campaigns are useful for early tests when you do not yet have enough purchase data. The risk for a store is judging them on cost per click alone. A low CPC says nothing about whether those visitors bought.
What to do:
- Use Target CPC for discovery tests with a fixed budget and end date.
- Track orders and revenue from those clicks in your store analytics, not just clicks in the ad platform.
- Move proven products or audiences to Target CPA once purchase tracking is verified.
3. The API changed on November 18, and Max Bid retires January 1
For API-managed campaigns, OpenAI set a separate milestone. Starting November 18, 2026, it recommends the dedicated Target CPA and Target CPC API fields. Legacy Max Bid and Max CPC API fields stay supported through December 31, but using them to create a campaign or change an ad group's bid type already selects Target CPA or Target CPC and treats the submitted number as an average-cost target.
That is the quiet risk. A script that still sends an old bid ceiling will not fail. It will hand the system a target that was never meant to be one, and the system will try to hit it on average.
Legacy controls are then scheduled for deprecation on January 1, 2027. From the November 18 API milestone to the cutoff is 44 days, and most of those days fall in peak selling season. Many stores freeze account changes during the busiest weeks, so the real migration window is shorter than the calendar suggests.
What to do:
- Find every script, rule, spreadsheet, or connector that writes Max Bid or Max CPC values, and assume any new campaign or bid-type change it made since November 18 is running on a target.
- Check those campaigns and confirm the number now acting as a target is one you would choose on purpose.
- Assign one owner to move each workflow to the dedicated target fields, and test it on a low-spend campaign before the holiday freeze.
- Migrate remaining Max Bid campaigns on a planned date rather than waiting for the deadline to force it.
4. Purchase-event tracking becomes the real bid
OpenAI says conversion campaigns need verified conversion signals through the OpenAI Pixel or Conversions API, warning that weak or missing signals can reduce delivery and leave budget unspent. For an online store, that means reliable purchase-event tracking. Once the system bids toward an average cost per conversion, the conversion signal effectively becomes your bid strategy.
A duplicated event makes the cost per purchase look lower than it is. A missing event makes it look higher. Either way, the target you set is measured against the wrong number.
The destination matters too. Eligible U.S. ecommerce and online-service businesses in the ChatGPT beta can use Business Agents as destinations for ChatGPT Ads. If you send traffic to an agent instead of a product page, confirm how a purchase that starts in that conversation gets recorded before you trust any CPA it reports.
What to do:
- Place a real test order from a ChatGPT ad click and confirm exactly one purchase event arrives.
- Check that the event carries order value and currency, not a placeholder.
- If you send both pixel and server events, confirm the same order is not counted twice.
- Make sure add to cart, checkout start, or newsletter signups are not set as the optimization event.
- Compare a week of ChatGPT-reported purchases with orders in your store backend that came from ChatGPT traffic.
5. Set the target CPA from contribution margin
A Max Bid was often a guessed ceiling. A target CPA deserves more care because the system will work to hit it on average, across every product you advertise. Start from what an order is actually worth to you after costs.
The formula is simple. Contribution per order equals average order value minus product cost, shipping, payment fees, and expected returns. That number is your break-even CPA, and your target CPA sits below it by the profit you need to keep per order.
What to do:
- Calculate contribution per order for the products you plan to advertise, not for the whole store.
- Recalculate when you run holiday discounts, since a lower price lowers your break-even CPA.
- Split campaigns by margin band when products differ sharply in profitability, so one target does not cover very different economics.
- Start the target near your measured Max Bid cost per purchase, then step toward your margin-based number rather than jumping.
6. Keep ChatGPT and Google Ads reporting separate
Most stores testing ChatGPT Ads also run Google Ads, and both now lean on automated bidding toward outcomes. Google is also moving ads into conversational surfaces, including multiturn ads in AI Mode in the U.S., so the same shopper can meet both platforms during research. Each system will report the conversions it sees, and the two can overlap.
If you blend the numbers, a weak channel hides behind a strong one and you lose the ability to adjust either target with confidence. Judge each platform on its own reported results, then reconcile both against store orders.
On the Google side, Kopa applies the same principles: tracking guardrails and profit-based ROAS reporting tied to your margins, with changes executed only after you approve them.
What to do:
- Report ChatGPT Ads and Google Ads in separate lines, each with its own cost, purchases, and CPA or ROAS.
- Add a third line for store-recorded orders so you can see how much the platforms double count.
- Use consistent UTM tagging so store analytics can attribute ChatGPT traffic on its own.
- Review each platform's targets on its own data, on a fixed weekly cadence.
What good looks like
- Tracking verified: one purchase event per real order, with correct value, confirmed by a test purchase.
- Margin-based targets: every Target CPA traces back to contribution per order, recalculated for holiday pricing.
- Legacy campaigns scheduled to switch: every existing Max Bid or Max CPC campaign has a planned migration date to Target CPA or Target CPC well before January 1.
- API workflows migrated: every script and connector writes the dedicated target fields, with a named owner, and no old ceiling is running as a target by accident.
- Click tests contained: Target CPC campaigns have budgets, end dates, and order-level evaluation.
- Separate reporting: ChatGPT Ads and Google Ads judged on their own numbers and reconciled against store orders.
- Done before the freeze: migration finished ahead of peak season, not on January 1.
If you want the Google Ads side handled with the same discipline, get started for free at kopa.ai.
