Tomorrow morning, the number a media buyer types into a new ChatGPT Ads campaign changes meaning. Same dollar sign, different contract. Until now that number told OpenAI’s auction where to stop. From Monday, November 9, it tells the system what you would like things to cost on average, and the system is explicitly allowed to go past it.
If your Q4 plan for ChatGPT relied on a manual cap to keep a test small, Sunday night is the time to find out which campaigns still have one, and to decide what replaces it, before Black Friday on November 27 and before OpenAI makes the call for you on January 1.
What changes Monday, in OpenAI’s words
OpenAI’s help article on bidding strategies and pricing, updated this week, lays out the new menu. Target CPA is “the average amount you’d like to pay for a conversion.” Target CPC is the same idea for clicks. Starting November 9, Target CPA “replaces Max Bid as the bidding option for advertisers creating new conversion campaigns who want to work toward a specific cost per conversion,” and Target CPC becomes available for new click campaigns. The fully automated option, Maximize results (autobid), stays and is now called intelligent bidding.
The email OpenAI sent advertisers, quoted in full by Barry Schwartz in both Search Engine Land and Search Engine Roundtable on November 5, is blunter on the click side: “While Max CPC caps individual bids, Target CPC lets our system bid above or below your target in each auction.” It also says that “starting Monday, November 9, you can optionally choose a cost target in Ads Manager when you create a campaign or update an existing one.”
Existing campaigns are untouched for now. Per the email, they “will continue to serve as Max Bid campaigns” or as maximum CPC campaigns until you change them. The help page says Max Bid and Max CPC “will retire on January 1, 2027 for existing campaigns,” with Target CPA replacing Max Bid in conversion campaigns and Target CPC replacing Max CPC in click campaigns.
What neither source says: which target value a capped campaign receives when it is converted on January 1, a gap Performance Marketing Hub also flagged on November 6. Until OpenAI documents it, assume nothing. A campaign you leave alone gets a number you did not write.
A target is a statement about the average
The help page is almost repetitive about what a target is not. “Your Target CPA or Target CPC is an average-cost goal that guides the system’s auction bids. It is not a fixed price per conversion or click, and actual average costs may be higher or lower than your target.” Its FAQ answers the question finance will ask, “Is a cost target a guaranteed maximum?”, with one word: “No.”
That gives you two misses to plan for. The tail: some clicks and conversions will cost more than your number by design. And the average itself, which OpenAI says “may be higher or lower” than your target. A Target CPC of $2 is a direction, and the invoice can land elsewhere.
Three smaller lines on the same page matter for anyone modeling spend.
The price rule did not change. “The final winning auction bid sets the price you pay,” and OpenAI says that applies across all strategies, including the legacy ones. What changes is who chooses the bid.
Bidding goal and billing event are separate settings. A campaign can optimize toward conversions while still being charged on clicks or impressions, depending on its billing setting. A Target CPA does not mean you only pay when someone buys.
Audience multipliers stack on the target. OpenAI’s own example: a $50 Target CPA with a 1.2x new-customer multiplier gives that segment a $60 target, which “remains an average-cost goal, not a guaranteed price per conversion.” If you built multipliers on top of capped bids, they now inflate an average instead of a ceiling.
Your budget is already an average too
This turns a bidding update into a governance question. In ChatGPT Ads, the daily budget was never a daily cap. OpenAI’s Daily budgets help article defines it as “the average amount you want your campaign to spend per day over a 7-day period,” with a budget week running Sunday through Saturday in the account’s time zone. Maximum spend on any single day is twice the daily budget. Maximum spend across the week is seven times it. PPC Land documented the switch to averaged budgets when it happened in July, and noted at the time that it came with no opt-out.
Put the two averages together. On a campaign you move to Target CPC, no individual click has a ceiling, and no individual day is held to the daily budget figure. The hard numbers left are the platform’s spend limits: twice the daily budget on any day, seven times it across the week. Black Friday falls on a Friday, near the end of the budget week that starts Sunday, November 22, so a campaign can spend up to twice its stated daily budget on the day you care most about, as long as the week stays inside seven times.
None of this is a scandal, and ChatGPT is not alone. PPC Land reported in August that Microsoft stopped accepting Max CPC on new non-portfolio Max Conversions, Max Conversion Value and Max Clicks campaigns from October 1. If your test plan says “we will not spend more than X on ChatGPT this week,” the field that enforces it is now the weekly budget, not the bid.
Leave the caps on through Black Friday, on purpose
Nothing forces you to switch an existing campaign on Monday. OpenAI’s email recommends starting the transition now, but the hard date is January 1. That leaves a real choice: move now, or keep the cap through the peak and migrate in December.
There are honest arguments for keeping a cap through November. The help page documents no learning period and no minimum conversion volume for the new strategies, so nothing published predicts how fast a switched campaign settles. Learning a new behavior during peak shopping weeks means learning it when mistakes cost most. And a cap is still the only per-auction limit you can hold, for seven more weeks.
There are also arguments for moving a few now. January 1 converts whatever is left all at once, and you would rather learn how the system behaves on two quiet campaigns in mid-November than on every campaign on New Year’s Day. If you move one, anchor the target to reality. OpenAI’s own example: $400 in spend and 10 purchases is a $40 CPA, and “setting a $20 target does not mean the system can deliver purchases at half the observed cost.”
The reasonable middle is a written split. Name the campaigns that keep Max Bid or Max CPC through November 30, and the one or two with steady conversion data that move this week as controlled trials. Everything else gets a migration date in the first half of December, after the Cyber Monday data settles and well before January 1.
One more trap for new campaigns. OpenAI’s Maximize results help article says that strategy, now intelligent bidding, is “selected by default for eligible new ad groups.” That article, last updated in late October, still names “Manual: Max bid” as the alternative, so expect labels in Ads Manager to move after Monday. Click through a quick holiday launch and you get the option with no cost target at all. OpenAI’s wording is that it “prioritizes result volume and does not guarantee delivery against a specific CPA, CPC, ROAS, or other cost-efficiency target at this time.”
Signals before targets
A Target CPA is only as good as the conversions the system can see. OpenAI says so directly: for conversion campaigns, set up the OpenAI Pixel, the Conversions API, or both, select a supported event, and “verify that real events are arriving at the data source linked to the campaign. Setup alone does not establish that usable conversion signals are being received.”
Event Quality Score is useful and easy to misread. The bidding page says a high score “does not establish that there are enough signals to support scale.” OpenAI’s troubleshooting guide for conversion tracking adds the mirror image: a low or unavailable score “alone cannot establish whether tracking works.” The score describes the information in the events you send, not how many there are. Fix the warnings for missing matching information, delays and repeated events, then check separately whether the event happens often enough to steer by.
Two details from that troubleshooting guide deserve a line in your switch checklist. A conversion-optimized campaign’s selected goal “cannot be changed after creation.” If the campaign optimizes toward the wrong event, the fix is a new campaign, which on or after November 9 means a target or intelligent bidding, not a cap. And the Conversions column in Ads Manager blends click-through and eligible view-through conversions when the view-through window is one day. PPC Land reported in August that view-through conversions sit outside bidding and CPA calculation. If your “recent actual CPA” came from a column that includes view-throughs, the number you hand the system may not be the number it steers by. Pull your baseline with the window you intend to optimize against.
OpenAI’s warning is the part that will show up first in reporting: “Weak or missing conversion signals can reduce delivery and leave budget unspent.” A target set below real performance does the same. Underspend is how a bad target announces itself, so give someone the job of watching pacing daily in week one.
API buyers get a second deadline
If your ChatGPT Ads run through an agency tool, a bidding script or a partner seat, the change has its own calendar. The help page says OpenAI is adding dedicated Target CPA and Target CPC fields for API partners. Before November 18, the legacy Max Bid and Max CPC fields “retain their existing behavior.” From November 18 the legacy fields remain supported through December 31, but “using them to create a campaign or change an existing ad group’s bid type will select Target CPA or Target CPC,” and “submitted values will represent average-cost targets rather than maximum bids.” On January 1, 2027, those fields are deprecated.
Read that middle step slowly. From Wednesday, November 18, a script that has always sent a ceiling can create a campaign that bids to an average instead, from the same value it always sent. The Keywordless made the same point on November 5. Ask your tools when they move to the new fields, get the answer in writing, and freeze automated campaign creation on legacy fields from November 18 until you have it.
Same fence, new material
Earlier this week the argument here was that ChatGPT clicks need a written price fence before anyone scales them. Monday does not weaken that argument. It changes what the fence is made of. A Max CPC was a crude fence, but it was physically in the auction. A Target CPC is a request. The fence you can actually enforce after the switch is a combination of three things you control: a target set from your own recent actuals, a weekly budget you can afford to lose, and a holdout that tells you whether the spend did anything.
OpenAI’s bidding page makes no performance claim for the new strategies, and no source this week offered one. Easier language for the buyer is not evidence of better economics for the brand.
What to do before Monday morning
Inventory every live campaign. List objective, current bid strategy (Max Bid, Max CPC, intelligent bidding, or already on a target), recent actual CPA or CPC, and weekly budget limit.
Write the number before you create anything new. For any campaign launched on or after November 9, record the target from recent actual CPA or CPC for the same event and reporting window, the weekly budget limit, and a holdout. Treat the target as an average with a tail, not as a price.
Clean signals before switching a conversion campaign. Confirm the Pixel or Conversions API is sending the selected event to the linked data source, read Event Quality Score warnings, and check volume separately from score.
Decide your Black Friday split. Name which capped campaigns keep their cap through November 30, which one or two move now as trials, and the December date for the rest. Do not let January 1 pick the target.
Reset budget expectations for peak week. Budget weeks run Sunday to Saturday. If finance needs a hard daily number for November 27, remember the platform can spend twice the daily figure that day and build the weekly limit around what you can tolerate.
Calendar November 18 and December 31 if you buy through the API or a partner. Ask each tool when it adopts the new target fields, and stop legacy-field campaign creation until it does.
The control that looked like a ceiling is becoming a preference. That is fine, as long as someone on your side still owns the numbers that actually stop spend.
