ChatGPT Ads Now Documents Why You Cannot Lower a Daily Budget — Your Highest Ad Group Bid Is the Floor

OpenAI updated the ChatGPT Ads Daily Budgets help article on 2026-09-03 with a new FAQ entry that had not appeared in any previous capture: Why can't I lower my daily budget?

The answer documents a constraint that plenty of advertisers had already hit as a save error with no explanation attached. On a campaign using a daily budget together with fixed bidding, the daily budget cannot be set below the highest effective per-event bid across your applicable non-archived ad groups — audience bid adjustments included.

That is a materially different kind of floor from the country minimums OpenAI already publishes. A country minimum is a fixed number you can look up. This one is derived from your own account configuration, it moves whenever anyone touches a bid, and until this update it was not written down anywhere.

Last verified: 2026-09-03, against help.openai.com/en/articles/20001413-daily-budgets and the developer documentation at developers.openai.com/ads/api-reference/campaigns and developers.openai.com/ads/api-reference/ad-groups.

What The Official Text Now Says

The new FAQ, quoted in full from the Daily Budgets article:

Why can't I lower my daily budget?

The minimum daily campaign budget varies by market and currency. The standard minimum for the US in USD is $25.

For campaigns that use a daily budget and fixed bidding, the daily budget must also be at least the highest effective per-event bid across applicable non-archived ad groups, including relevant audience bid adjustments. This bid-based constraint does not apply to automated bidding or lifetime-budget campaigns.

To set a daily budget below an existing ad-group bid or adjusted effective bid, first lower each incompatible bid cap or applicable adjustment to the desired daily budget or below. Then save the campaign's new daily budget.

For example, a $250.50 effective ad-group bid prevents you from setting that campaign's daily budget below $250.50 until you lower the bid. This is not a universal $250 platform minimum.

Nothing equivalent appeared in the previous captured version of the article. The rest of the page — the seven-day average, the 2× daily and 7× weekly ceilings, the mid-week proration rule, the catch-up spend of unused budget — is unchanged from the rollover update on 2026-08-25.

Three Things This Actually Establishes

1. There are two independent floors, and they are not the same rule. The market minimum ($25 in the US) applies to every daily-budget campaign. The bid floor applies only to daily-budget campaigns on fixed bidding, and its value is whatever your highest effective bid happens to be. Your true minimum is the greater of the two. Most advertisers running sane CPM bids will never see the second one; anyone running CPA bids on conversion campaigns will see it constantly, because a $100 CPA bid is four times the US floor on its own.

2. "Non-archived" is doing real work in that sentence. The check runs across applicable non-archived ad groups. A paused ad group is not archived. So a bid you set months ago on a paused ad group, and have not thought about since, is still setting the floor on the campaign's budget today. The API confirms these are distinct states — POST /ad_groups/{id}/pause and POST /ad_groups/{id}/archive are separate endpoints, and OpenAI warns that archiving is not reversible. That makes the remedy asymmetric: pausing is cheap and does not clear the floor, archiving clears the floor and cannot be undone. Lowering the bid is usually the right move over archiving.

3. OpenAI is pre-empting a specific misreading. The final sentence — "This is not a universal $250 platform minimum" — is not the sort of line a documentation team writes speculatively. It is the shape of a sentence written after enough people hit a $250-ish floor, compared notes, and concluded the platform had quietly raised its minimum. Worth remembering the next time a number in an error message looks like a policy.

The Audience Multiplier Is the Trap

The help article says the floor includes "relevant audience bid adjustments" and leaves it there. The developer documentation is where that phrase becomes concrete, and the arithmetic is worth spelling out because nobody has spelled it out.

In the Ads API, an ad group's bidding configuration carries bidding_config.custom_audience_bid_multipliers, and each entry has a bid_multiplier_micros documented with a range of 100000 to 10000000, representing 0.1× to 10×.

So the effective bid is not the number in your bid field. It is that number times whatever multiplier applies to the matched audience. Which means:

• A $25.05 base bid with a 10× multiplier on one audience produces a $250.50 effective bid.
• That is, to the cent, the example OpenAI chose to publish.

Whether that was deliberate or coincidental, the mechanism is the point: you can be blocked from lowering a daily budget by a figure you never typed anywhere. You typed $25.05, which is barely above the US minimum. The multiplier did the rest. And because the help article describes the outcome in terms of an "effective ad-group bid" rather than in terms of a multiplier, the number in the error message will not obviously map to anything visible in the campaign.

This is the operational takeaway of the whole update: when a budget save is refused, check the multipliers before you check the bids.

The API Surface Says Something the Help Centre Does Not

The two official surfaces do not contradict each other here, but they describe different worlds, and the gap is informative.

Search the documented POST /campaigns payload at developers.openai.com and there is no daily budget field at all. The only budget field is:

Every campaign created through the Ads API is therefore a lifetime-budget campaign. And the new help text states plainly that the bid-based constraint "does not apply to automated bidding or lifetime-budget campaigns."

Put those two facts together and the conclusion is clean: campaigns created through the Ads API are structurally exempt from this floor. Not exempted by a carve-out — exempted because the API does not offer the budget type the rule governs.

That is worth stating precisely rather than loosely, because it is easy to over-read. This does not mean API campaigns escape budget constraints; the lifetime spend limit is a real cap with its own minimum. It does not mean the API is a workaround for a bid you should have lowered. What it means is narrower and more useful: this constraint is an Ads Manager behaviour attached to daily budgets, not a platform-wide bidding rule — and anyone reasoning about ChatGPT Ads purely from the API reference would have no reason to know it exists.

Two other API details sharpen the picture of what "effective per-event bid" means in practice:

• max_bid_micros is documented as per event, so a $60 CPM — $0.06 per impression — is passed as 60000. The floor is measured against the per-event figure, not against a daily or CPM-normalised one.
• For conversion-optimized (oCPC) campaigns, billing_event_type is click, but max_bid_micros is the CPA bid: 100000000 is a $100.00 CPA bid on a USD account. So an oCPC campaign's effective bid is a CPA-scale number, and CPA-scale numbers sit far above the $25 US minimum. If you run oCPC on daily budgets, expect this floor to be your binding constraint most of the time, not the country minimum.

What Did Not Change

Worth recording, because "the docs changed" is frequently reported as more than it is.

• The developer documentation surface did not change at all in this window. Zero of the captured developers.openai.com/ads/* pages differed from the previous run, llms.txt included. This is a help-centre-only edit.
• There was no OpenAI advertiser product-update email in this window. The most recent one is dated 2026-08-20, outside the rolling period, and it does not mention this. So the documentation diff is the sole source, and there is no announcement claiming this as a launch.
• The rest of the Daily Budgets mechanics are unchanged: 2× the daily budget on any single day, 7× across the seven-day period, mid-week changes prorated from the change through the following Sunday at midnight, and unused budget spendable later in the same period.
• The $25 US minimum is not new. It was already the published figure, and the country-by-country table remains the more complete reference for market minimums.

One thing the official text does not say, and which should therefore not be claimed: whether the underlying validation is new, or whether only its documentation is. OpenAI describes the constraint in the present tense without a changelog note. The honest reading is that this is newly documented, and advertisers who hit it before now have an explanation they previously lacked.

What To Do About It

Before you try to lower a budget. Pull the effective bid across every non-archived ad group in the campaign — including paused ones — and take the maximum. That number, or your market minimum, whichever is higher, is your real floor. Do this before you promise anyone a spend reduction, because discovering it at save time is how a planned Friday budget cut becomes a Monday one.

When a save is refused. Follow OpenAI's documented order: lower each incompatible bid cap or audience adjustment to the target budget or below first, then save the budget. The validation runs against the state at save time, so the reverse order simply fails again. Check multipliers before base bids — a 10× adjustment moves the effective figure ten times faster than editing the bid does.

If you are scaling a campaign down rather than off. This constraint makes aggressive down-scaling structurally awkward: cutting a budget by 80% may require touching every ad group's bidding first, which changes delivery economics as well as spend. Consider whether pausing ad groups and archiving the ones you genuinely will not revive is cleaner than fighting the floor — bearing in mind that archiving is irreversible and pausing does not help.

If the floor is a recurring nuisance. Two documented exits exist, and both are real decisions rather than tricks. Automated bidding — the Maximize results strategy — is exempt, because there is no fixed per-event bid to measure. Lifetime budgets are exempt, which is also what the API gives you by default. Neither should be adopted to dodge a validation rule; both should be adopted if they suit the campaign, with this as a secondary benefit.

If you run oCPC on daily budgets. Assume this floor binds. A CPA bid is denominated in the tens or hundreds of dollars, and it is measured against a daily budget denominated in the same units. That is a much tighter relationship than a CPM campaign ever experiences, and it means your minimum viable daily budget on a conversion campaign is effectively set by your CPA target. Plan the budget and the CPA bid as one decision, not two — the mechanics of which are covered in conversion-optimized oCPC campaigns.

The Short Version

A daily budget on ChatGPT Ads now has a documented second floor: the highest effective per-event bid across your non-archived ad groups, audience multipliers included. It applies only to fixed bidding on daily budgets, it is exempt on automated bidding and on lifetime budgets, and because the Ads API only offers lifetime budgets, API-created campaigns never meet it.

The practical risk is not the rule. It is that a 10× audience multiplier on a modest bid can manufacture a floor in the hundreds of dollars that appears nowhere in your campaign as a number — and that a paused ad group you forgot about still counts.

Sources

• OpenAI Help Centre — Daily Budgets (captured 2026-09-03; the Why can't I lower my daily budget? FAQ is new in this capture)
• OpenAI Developer Docs — Campaigns API reference (budget.lifetime_spend_limit_micros; no daily budget field)
• OpenAI Developer Docs — Ad Groups API reference (max_bid_micros per event; custom_audience_bid_multipliers at 0.1×–10×; oCPC CPA bid semantics)
• OpenAI Help Centre — Maximize results bid strategy
• OpenAI Help Centre — Billing & Payment

Facts above are attributed to OpenAI's official documentation as captured on 2026-09-03. The audience-multiplier arithmetic, the paused-versus-archived asymmetry, and the API-exemption conclusion are Soku analysis derived from those sources, not statements by OpenAI.