ChatGPT Ad Clicks Are Worth About $0.26, and OpenAI Suggests Bidding $3

ChatGPT Ad Clicks Are Worth About $0.26, and OpenAI Suggests Bidding $3
In the paper's U.S. data, a paid search visit brought in $2.45. A paid ChatGPT visit brought in $0.89.

A paid ChatGPT ad visit earned $0.89 in tracked revenue in the U.S., against $2.45 for a paid Google Search visit, according to an October 2026 working paper covering 102 advertisers. At that ratio, a ChatGPT click is worth about $0.26 if you expect search-level returns. OpenAI suggests starting bids at $3 to $5, so most test budgets need a written kill threshold before launch.

The paper comes from Maximilian Kaiser (Vienna University of Economics and Business) and Christian Schulze (Frankfurt School of Finance & Management), and it's titled "Can Advertising Fund Free AI?" The data is Google Analytics e-commerce tracking from Grips Intelligence: 60 weeks, roughly 10 billion sessions, ending August 22. Matt Southern's write-up at Search Engine Journal pulls out the paid-media numbers, and they're blunter than the abstract. Of 87 advertisers that bought both ChatGPT ads and paid search for U.S. traffic, 84 earned less per ChatGPT visit than per search visit.

Three out of 87 beat search. That's the headline I'd put on a slide for whoever approved the test budget.

Revenue per visit is the number your test plan is missing

Most ChatGPT Ads tests I've seen discussed get judged on CTR, CPC, or a vague "the traffic looks engaged." None of those tell you whether the click paid for itself. Revenue per visit does, because you can divide it by what you paid for the visit and get revenue per ad dollar directly.

Run the paper's numbers through that and the gap gets uncomfortable fast. Paid search: $2.45 per visit on a median CPC of $0.72 (from 16 accounts with at least 80% U.S. search traffic), so about $3.40 of tracked revenue for every dollar spent. ChatGPT at the paper's $0.89 per visit:

  • At a $3 CPC, the low end of what OpenAI's ads basics page suggests as a starting max bid, you get roughly $0.30 back per dollar.
  • At $1.72, the CPC Opascope reported from one scaled account (the largest publicly known figure, per the paper), you get about $0.52.
  • To match search's $3.40 per dollar, you'd need to pay about $0.26 per click. The authors' resampled range on that parity price runs from $0.05 to $0.57.

Conversion rate tells the same story from the other side: 0.37% for paid ChatGPT visits, 1.17% for paid search. The authors say the gaps hold after controlling for device and week.

Now, the caveats, and they matter. This is last-click attribution only. It doesn't see prompts, impressions, or anyone who saw an ad in ChatGPT and searched the brand on Google two days later. The parity price ignores margin. And advertisers chose to buy these ads, so it's a description of how early buyers did, not proof that the channel can't work. Kaiser is also a scientific advisor to Grips, which the paper discloses. I don't think any of that flips the conclusion, but a 3x gap on last click could plausibly shrink to something less dramatic once you count assisted conversions. Shrink, not vanish.

The average hides one very lucky advertiser

This is the part that surprised me. Of the 102 advertiser sites with full sales tracking, 73 had zero ChatGPT-attributed sales across 18,726 visits combined. The remaining 14 that had any sales at all produced all 570 ChatGPT purchases. And a single advertiser accounted for 66% of those purchases and 76% of U.S. paid ChatGPT revenue.

Take that one out and the gaps get wider, per SEJ's summary.

So the "average" ChatGPT advertiser in this dataset is, in practice, one big winner plus a long tail of accounts that sold nothing. Which also squares with the louder success stories. OpenAI's August 31 post mentioned one ecommerce advertiser seeing 3x ROAS over 28 days, and DoubleVerify's attribution launch cited Weight Watchers at a cost per acquisition 15.3% below benchmark. Opascope's account posted a 1.49x blended ROAS on about $60K of spend with a 2.35% conversion rate, though its author says plainly it isn't a universal benchmark. Those results are probably real. They just look a lot like the top of a very skewed distribution, and case studies tend to come from the top.

Think of it like a slot machine bank that publishes its jackpot photos. Every photo is genuine. None of them tell you the payout rate.

A test budget that can actually reach a verdict

Here's what I'd do before spending another dollar on ChatGPT Ads, or before renewing a test that's been running on vibes.

1. Calculate your own parity CPC. Pull revenue per visit and average CPC for your non-brand Google Search campaigns over the last 90 days. Your ChatGPT parity CPC is your search CPC multiplied by (ChatGPT revenue per visit ÷ search revenue per visit). Before you have ChatGPT data, use the paper's ratio of about 0.36 as a placeholder. If your non-brand search CPC is $2.50, that puts parity near $0.90, not $3.

2. Size the test so zero sales means something. At a 0.37% conversion rate, 250 visits produces less than one expected purchase, which is roughly what the 73 zero-sale advertisers averaged (18,726 visits split 73 ways is about 256 each). You need something like 2,700 visits to expect around 10 purchases. At Opascope's $1.72 CPC that's about $4,650, which lines up with the $5K to $10K test budget Opascope recommends. If you can't fund that, I honestly wouldn't run the test at all. A $1,000 test mostly buys you an inconclusive dashboard.

3. Write the kill rule down before launch. Something like: after 2,700 paid ChatGPT visits, if revenue per ad dollar is below 50% of non-brand search, pause and reallocate. Pick your own percentage, but pick it in advance, while nobody is emotionally invested in the channel yet.

4. Track it outside the ChatGPT dashboard. Opascope's author notes the native dashboard reports CPC and CPA but not ROAS, and seemed to undercount conversions. Use GA4 or your own revenue data with consistent UTMs. And watch for ad-review crawler visits carrying your tracking parameters, which the paper flags as a source of fake "paid" sessions. If you've been bitten by attribution weirdness in this channel already, our piece on Delta buying ChatGPT ads through Amazon DSP covers how messy the click path gets.

Where I think ChatGPT ad pricing goes from here

There's a reasonable counterargument: this is a 2026 snapshot of a channel that started in February, and Google's search ads were not exactly efficient in 2002 either. Fair. OpenAI now says the product covers 89 countries and more than 100,000 advertisers have run campaigns, which is a lot of signal for its systems to learn from. And honestly, the organic side of the paper was kinder to OpenAI than I expected: ChatGPT referral volume fell an estimated 6% to 8% after ads launched, but the people who still arrived converted exactly as before (2.21% baseline, basically no change).

But the paid gap is roughly 3x, and closing 3x through learning alone seems like a stretch. My guess: by mid-2027, effective ChatGPT CPCs for most retail advertisers settle under $1, either because underperforming buyers pull budget and soften the auction or because OpenAI shifts more spend toward outcome-based bidding. That fits with Google's own admission that AI Mode ads earn less than classic search. Conversational ad inventory, from every vendor, seems to be monetizing below the old search results page so far.

None of this means skip the channel. It means a ChatGPT test should be priced like an experiment with a known downside, and the downside right now is about 70 cents of every dollar if you bid what OpenAI suggests. If your own revenue-per-visit number comes back better than $0.89, good, you might be one of the 14. I'd just rather find that out on a $5K test with a kill rule than a $50K one without.