Someone will ask you for a forecast. "What CTR should we expect?", "what will a lead cost?". And you will run into this: OpenAI publishes no performance benchmarks at all. No CTR, no conversion rate, no CPA, and none by industry. It is not an oversight: the platform itself acknowledges it does not yet have benchmarks across advertisers, industries or campaign types (reported by Choice OMG and by the Headcore manual, both in 2026).
Everything in circulation, then, is somebody's sample: one agency's account, one tool's panel, one market, one month. This lesson is about reading those samples properly, and about building the only benchmark that will actually decide anything for you — your own.
The one thing OpenAI does guide: the starting bid
A results benchmark and a starting bid are not the same thing. For clicks campaigns a recommended maximum of $3-5 CPC is cited, and for reach a default $60 CPM (reported by Choice OMG, guide updated 26/08/2026). For conversions, not even that: there is no recommended amount.
⚠️ Trap: a recommended bid tells you what to enter the auction with, not what a result will cost you. Confusing the two is what turns a forecast into a promise you will not keep.
Six questions before using someone else's number
- Who measured it? The platform, an agency with its own account, or a tool with a panel of users?
- How? A real account's spend is not the same as a panel estimate (monitored people, extrapolated outwards).
- How big is the sample? Impressions, clicks, accounts, industries. With 58 clicks nobody gets to state "the CTR of ChatGPT Ads".
- Which period? In a channel that changes monthly, a May figure describes a different platform.
- Which market? US data comes from campaigns with personalisation; in Europe the ad is chosen from the conversation, with no history. They are not comparable.
- Who benefits from the number? Whoever sells training, audits or a tool has an interest in a striking figure.
Three real figures, read with those questions
| Figure | Who and when | What you cannot conclude |
|---|---|---|
| Average CTR 0.68% (top quartile 1%, best brands 1.57%) | Similarweb, published by Search Engine Roundtable on 05/05/2026; a panel estimate, with no methodology detailed in the article | That your CTR will be that. The split by industry or market is unknown |
| 0.65% CTR from 8,940 impressions and 58 clicks; CPCs of CAD 4.42 and 4.88 | Agency Choice OMG's own test, 8-25 June 2026, Canada | Nothing general: one account, three weeks, another currency |
| 1.30% CTR from 97,000 impressions and 1,263 clicks | SE Ranking test, August 2026, campaigns in the US, Canada, Australia and New Zealand (reported by Choice OMG) | That it applies to Europe: those are personalised markets |
Notice the point: two honest samples give 0.65% and 1.30%. Double. Either one, out of context, is a headline.
⚠️ Trap: the same goes for business headlines. "$1bn annualised in under 200 days" (OpenAI, 31/08/2026) measures OpenAI's growth, not your profitability.
What if they demand a forecast?
Tell the truth and turn it into a plan: "there are no benchmarks for this channel, so month one is the measurement". Put on the table how much you will spend, over how many days, and which number decides whether it continues. A bounded learning budget gets approved; an invented forecast gets remembered when it fails.
Build your own benchmark in 30 days
It is quicker than it sounds, and the result is actually usable.
- Week 0 · Measurement first. Without it, the 30 days are worthless. And filter bots: crawlers and link checkers sneak in as people.
- One campaign, one objective, few ads. Change three things at once and after 30 days you will not know which moved the number.
- A budget that lasts the whole month. A little a day for 30 days beats a lot for 4.
- Log five numbers daily: impressions, clicks, spend, real sign-ups or leads, and cost per one of them.
- Touch nothing for two weeks. Every change resets your sample.
- At 30 days pull YOUR numbers and compare them against yourself: your cost per lead in Google Ads, in social, in organic.
That internal comparison answers the only question that matters: is this channel cheaper or dearer than the ones I already have? You do not need an industry average for that.
📌 We measured this ourselves: our position is to publish our own numbers with a date, even when they are small or bad. On 17/09/2026, the first day our campaign served, they were 4,024 impressions, 0 clicks and €0 spent. Not a flattering figure, but verifiable and ours (see "Diary of our campaign, day by day").
What to take away
- OpenAI publishes no performance benchmarks: it says so, and the serious industry guides report it.
- What it does guide are starting bids ($3-5 CPC, $60 CPM): they are not forecasts of cost per result.
- Interrogate every figure in circulation: who, how, how much sample, which period, which market and who benefits.
- Honest samples of the same channel give 0.65% and 1.30% CTR: which is why a loose average decides nothing.
- Your benchmark is built in 30 days and compared against yourself, not against other people's averages.