The First Fourteen Days of a ChatGPT Ads Campaign, Decision by Decision
By Keshav Parsai 9 min read
In brief
A day-by-day operating plan for the first 14 days of a ChatGPT Ads campaign, including when to wait, diagnose or change course.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 October 2026
Three days in the first fortnight carry a decision: day one, day four and day fourteen. Day one confirms the campaign can serve. Day four confirms it is serving enough to read later. Day fourteen is the first point at which a performance number means anything. Everything in between is watching, and the main failure mode is acting.
This is about operating the account, not about running a project. Nothing here concerns kickoff meetings or reporting cadence.
Day 0: the checks that prevent a wasted fortnight
Before the campaign goes live, four things decide whether days one to fourteen produce data or an empty report.
OAI-AdsBot must be allowed on your site, or ads can be rejected outright. Your verification and billing must be complete, because the account is postpay and an incomplete business verification stops delivery rather than warning you. Campaign start and end dates must actually cover today. And your pixel must be firing before the first impression, because conversions are not backfilled and a pixel installed on day five makes days one to four permanently unreadable.
The pixel point is the one people get wrong. A campaign that runs four days without measurement has not saved four days of setup, it has spent four days of budget buying nothing you can use.
Day 1: does it serve at all
The only question on day one is whether impressions are accumulating. Not how many, not at what CTR. Whether the number is above zero.
A campaign showing zero impressions after a full day is a delivery problem, not a performance problem, and the two get confused constantly. Delivery problems have a short list of documented causes: review status, verification, dates, targeting that is too narrow, and a bid under the clearing price. None of them are fixed by rewriting creative.
If impressions are above zero, do nothing else. Do not read CTR on day one. A handful of impressions produces a CTR that will swing by a factor of five tomorrow, and the number will tempt you into a change you cannot undo cleanly.
Days 2 to 3: leave it alone
There is no action here and that is the point.
The argument for leaving it alone is not patience as a virtue, it is arithmetic. At the 25 USD daily minimum and reported CPC guidance of three to five dollars, a single-ad-group campaign is buying somewhere around five to eight clicks a day. After three days you have perhaps twenty clicks. Twenty clicks cannot distinguish a 2% CTR from a 4% CTR, and cannot distinguish a working ad from a failing one at any conversion rate a B2B advertiser would care about.
The one thing worth checking on day two is the reconciliation between OpenAI's click count and your own analytics sessions. Agencies reported mismatches to MediaPost in August 2026, including a case of 57 platform-reported clicks against fewer than 20 recorded site visits. If your gap is large in either direction, you want to know on day two, while it is cheap to investigate, not on day fourteen when it invalidates the whole read.
Day 4: is this going to be readable
Day four is the first genuine decision point, and the decision is about volume, not performance.
Take the spend so far and project it forward to day fourteen. Then ask whether that projected click count is enough to distinguish the outcomes you care about. If your campaign is pacing at four clicks a day, day fourteen arrives with 56 clicks, and 56 clicks will tell you almost nothing about conversion rate for a B2B offer converting at low single digits.
There are three responses and only three. Raise the budget so the test finishes readable. Consolidate ad groups so the same spend concentrates into fewer readable units. Or accept in advance that this is a delivery test and not a performance test, and write down what you will and will not conclude from it.
What you should not do is leave a campaign pacing below readability and plan to draw performance conclusions from it anyway. That is how a channel gets declared dead by a test that was never capable of showing life.
Days 5 to 10: the watching window
Two things are worth watching in this stretch and neither is CTR.
Spend pacing against your daily budget. A campaign consistently underspending its budget is telling you something about inventory or bid position. Delivery on this channel has been uneven: Digiday reported in 2026 that early pilot advertisers saw severe underdelivery, in one case 2,500 USD spent against a 250,000 USD commitment with roughly 200 impressions over four weeks, with fill rates improving 30 to 50 percent from launch levels by late April. Treat that as reported, and treat persistent underspend as a delivery signal rather than a performance one.
Ad review status on any new creative. Ads added mid-flight enter review. An ad group where one of two ads is stuck in review is not running the creative test you think it is running.
If something must be changed in this window, change budget. Budget moves are reversible and do not reset anything you are trying to read. Creative swaps and hint rewrites restart the clock on a sample that was not finished.
Day 14: the first real read, and what it can carry
At fourteen days you have a baseline. That is the correct word for it, and it is a smaller claim than it sounds.
What the fortnight can support: whether the channel serves for your category, roughly what you pay per click, whether clicks reach your site at a rate that reconciles with your analytics, and a first internal reference point for CTR that later ad groups can be compared against. OpenAI publishes no cross-advertiser benchmarks, so your own first fortnight is the only benchmark you will have.
What it cannot support: a conversion rate you should trust, a verdict on the channel, or a comparison against Google or Meta where those channels have years of accumulated learning behind them and this one has a fortnight.
The output of day fourteen is not a decision to scale or kill. It is a written baseline and a second test designed against it.
The reckoning: day four is the day people skip
The honest weakness in this schedule is that day four asks you to make a projection from three days of data, which is exactly the kind of small-sample reasoning the rest of the article warns against.
The defence is that the day four question is different in kind. It asks about pacing, which stabilises quickly because it is driven by budget and delivery rather than by user behaviour. Three days of spend data predicts fourteen days of spend reasonably well. Three days of CTR data predicts nothing. You are allowed to project the one and not the other, and the distinction is the difference between operating this channel competently and reacting to noise.
What we cannot tell you
- How many clicks this channel needs for a stable CTR. No advertiser has published a variance study, so every recommended sample size in circulation, ours included, is inference from general auction behaviour.
- How long ad review takes. OpenAI does not publish a review time, and reported experience varies.
- Whether delivery ramps over the first fortnight. There is no published pacing or learning-phase documentation for ChatGPT Ads. Whether a new campaign serves differently on day twelve than on day two is undocumented.
- What share of eligible users see ads. Ads serve to Free and Go plan users only, users can opt out, and take-up of that opt-out is unpublished.
- Whether OpenAI's click count reconciles with site analytics. Reported discrepancies exist in both directions and no reconciliation method is published.
Quick answers
What should I check on day one of a ChatGPT Ads campaign? Only whether impressions are above zero. Zero impressions is a delivery problem with a short list of causes: review status, verification, campaign dates, over-tight targeting or a bid below the clearing price. Do not read CTR on day one.
When can I first judge performance? Day fourteen at the earliest, and only as a baseline rather than a verdict. At the 25 USD floor and reported three to five dollar CPCs, a fortnight buys roughly 70 to 110 clicks, which is enough to estimate cost per click and not enough to estimate conversion rate.
Should I change context hints in the first two weeks? No. Hint changes are the least readable intervention available, because nothing is reported at hint level, and they restart a sample that was not finished. Change budget instead.
What if my campaign is underspending its daily budget? Treat it as a delivery signal, not a performance one. Check review status, verification, campaign dates and targeting breadth first, then bid. Early advertisers reported severe underdelivery in 2026 with fill rates improving over time.
When should the pixel be installed? Before the first impression. Conversions are not backfilled, so any day the campaign runs without the pixel is permanently unmeasurable.
Is fourteen days long enough to decide whether ChatGPT Ads works? No. It is long enough to decide whether your campaign is readable and what it costs to run. A verdict on the channel needs a sample large enough to measure conversion, which at the minimum budget takes considerably longer.
Sources
| Claim | Source | Tier |
|---|---|---|
| 25 USD minimum daily budget, no minimum lifetime commitment | OpenAI Ads Manager, 2026 | Confirmed, primary |
| Three to five dollar CPC starting bid guidance | Multiple agency write-ups citing OpenAI guidance, 2026 | Reported |
| OAI-AdsBot must be allowed or ads can be rejected | OpenAI advertiser documentation, 2026 | Confirmed, primary |
| Postpay billing, business verification required before serving | OpenAI Ads Manager account setup, 2026 | Confirmed, primary |
| Ads serve to Free and Go plan users; users can opt out | OpenAI, 2026 | Confirmed, primary |
| No published learning phase or pacing documentation | OpenAI documentation, by absence | Absent |
| 2,500 USD spent against a 250,000 USD commitment, roughly 200 impressions over four weeks; fill rates up 30 to 50 percent by late April | Digiday, 2026, anonymous advertiser and agency sources | Reported, third party |
| 57 platform clicks against fewer than 20 analytics visits | MediaPost, 31 August 2026, citing Nicholas Verity of Cleverly | Reported, third party |
| Click volumes implied by budget and CPC range | Arithmetic from the above | Inference, ours |
Related reading
- How to Optimise a ChatGPT Ads Campaign With No Search Terms Report
- Why Your ChatGPT Ads Campaign Is Not Spending
- Reading CTR and CPC When No Benchmarks Exist
- ChatGPT Ads Measurement: The Pixel and the Conversions API
- Budget Calculator
Changelog
12 September 2026, v1.0. First publication. Establishes a three-decision fortnight, separates the day four readability check from performance judgement, and defends projecting pacing while refusing to project CTR.
Field kit
Tools
Site
OpenAI (primary)
Keshav studies how AI systems retrieve, verify, and cite brand information. At InPromptAds, he leads source research and turns platform documentation into practical guidance for advertisers.
