Cost Per Acquisition Benchmarks for AI Chat Ad Campaigns

No official CPA benchmark exists yet for ads inside ChatGPT, and that gap is itself the finding worth reporting. AI chat ad spending is real money now, but it runs a distant second to Google and Meta, and the volume hasn't built up to where anyone can publish a trustworthy average. What follows builds a working framework anyway: what legacy channels cost today, what OpenAI's platform charges, why a conversational prompt behaves differently than a keyword, and what the early conversion numbers actually suggest, as opposed to what they're being sold as.

Read the growth number for what it is. A billion dollars in spend, spread across a handful of large advertisers and a long tail of small ones, does not generate the standardized, cross-advertiser volume that produces a reliable CPA figure. Google Ads has had two decades and millions of campaigns to settle on a reliable average figure. ChatGPT ads have had months. Anyone quoting a firm CPA for this channel right now is quoting a guess dressed up as data, and that includes the estimates built later in this piece.

What legacy channel CPAs actually look like across search and social, and why that's the number the new channel has to beat

Start with what's actually known, because it's the only solid ground on offer. Google Ads in 2026 runs a well-documented average CPA across all industries, per benchmark data from Search Engine Land and similar sources. That number swings hard by vertical: finance and legal advertisers on Google pay substantially above the average per acquisition, which is what happens when a keyword auction gets fought over by companies with high customer lifetime value and nowhere else to spend.

Meta tells a similar story with a wider spread. Its median CPA sits at $38.19 across industries in 2026, but the vertical extremes blow past that median by a wide margin: Legal Services averages $187.60, Insurance $198.42. These are exactly the categories where a channel offering better-qualified traffic, even at a premium CPM, has the most room to prove itself. If ChatGPT ads are going to beat anything, this is where the case gets made first.

None of these numbers sat still to get measured. CAC climbed meaningfully across channels in recent years, driven by auction saturation and rising CPMs on platforms that have been mature for a long time. Staying put in legacy channels has gotten more expensive, and that's the actual reason marketers in high-CPA verticals have a reason to look elsewhere. Google and Meta's figures carry weight because they're built from years of auction history and standardized tracking across millions of campaigns. Early AI chat numbers don't have that pedigree, and won't for a while yet.

What ChatGPT's ad platform costs today, and what the dashboard won't tell an advertiser

OpenAI's official bid guidance puts CPC at $3 to $5 per click, with a listed default max CPM bid of $60. Digiday has reported some advertisers seeing CPMs as low as $25, so the realistic operating range runs $25 to $60 depending on how crowded the category is. Early adopters in less competitive verticals report CPCs coming in under $4.

The auction runs on a second-price, relevance-weighted model, which means the highest bid does not automatically win the impression. A tightly matched, contextually relevant ad on a lower bid can beat a generic ad backed by more money. That's a genuinely different mechanic from keyword auctions, where relevance gets scored through match types and Quality Score, systems advertisers have spent twenty years learning to game. Here, relevance comes out of the conversation itself, and nobody has built the playbook for that yet.

Access opened up fast, and that speed matters. Minimum ad spend commitments dropped from $200,000 to $50,000 to zero over roughly three months, as the platform moved out of a limited beta and toward something closer to self-serve. That's the mechanism that will eventually produce the broader advertiser pool real benchmarks need to exist.

Here's the part that should worry anyone reading a ChatGPT ads dashboard at face value: it reports cost per click and cost per acquisition, but revenue per conversion is invisible unless the advertiser builds first-party tracking and manually ties revenue back to spend. In practice, dashboard-reported CPA is probably an overstatement of true cost, because conversions that happen later, off-platform, through research behavior the ad started, never get counted. That reflects how people actually use a chat interface: they research there and buy somewhere else, often days later.

Why conversational intent is structurally different from a search keyword, and what that means for CPA

A search query compresses intent into one to four words. A conversational prompt carries context, constraints, comparison criteria, and purchase stage, often across full sentences. "Best project management software" is a keyword. "I'm managing a remote team of eight, we've outgrown spreadsheets, and I need something under $20 per seat that integrates with Slack" is a brief. The second one contains targeting signals no keyword match type was ever built to hold.

ChatGPT takes in 2.5 billion prompts a day, and a real share of that volume is people working through a decision out loud before they buy anything. That signal is hard to overstate, and most advertisers aren't touching it yet. eMarketer's 2026 data shows more than 80% of AI advertising still runs adjacent to AI content rather than inside the conversation itself, which is the gap platforms like Thrad, a programmatic ad infrastructure built for LLM chat environments, are structured around. Most of the money is buying placement next to the conversation rather than a seat inside it, which cuts against the whole argument for this channel resting on intent quality.

The implication for CPA follows directly. A user who has already stated their problem, their budget, and their readiness to buy sits further down the funnel than someone who typed three words into a search box, and in principle that means fewer impressions and fewer clicks are needed to close a conversion. Whether that principle survives contact with real campaigns is a conversion-rate question, not a CPA question, which is exactly why conversion rate deserves more scrutiny than CPA at this stage of the channel's life. It's the metric that tests the premise everything else rests on.

Meta makes for a useful contrast. It uses AI conversation as a signal that informs targeting for ads served elsewhere, in the feed, in Stories, wherever the algorithm decides to place them. The ad never shows up inside the intent moment itself. That's a different product from an ad that appears inside the answer to the exact question a user just asked.

Early conversion rate signals, and where the first-mover data runs thin

The headline numbers look strong, then they get complicated fast. Analysis from trylapis.com puts a good conversion rate for ChatGPT ads in 2026 at roughly 4% to 7% for commercial verticals, with top performers clearing 8%. First Page Sage's 2026 report, covering 19 industries, finds a much wider spread: 0.2% to 5.8%. These aren't contradictory so much as they're measuring different populations: the narrower range describes well-optimized campaigns in commercial categories, and the wider one includes poorly matched placements and early-funnel traffic that was never going to convert fast.

Set those against the legacy comparisons. Google Search conversion rates in matched verticals typically run 2% to 4%. Google Shopping runs 1.5% to 3.5%. Meta runs 1% to 3%. If the 4% to 7% range for AI chat holds up as the channel matures, that's a real lift, and it's the single strongest argument in this channel's favor.

Click-through rate says the opposite, and the contradiction is worth sitting with rather than smoothing over. Early advertiser data puts ChatGPT ad CTR at 0.3% to 1.5%, well under Google Search's 6.64% and Meta's 1.71%. But Ad inventory inside a chat interface is structurally limited compared to a search results page stacked with ad slots or a feed that refreshes every few seconds. CTR isn't measuring the same thing across these platforms, and treating it as comparable produces a false picture. Conversion rate is the number that carries meaning here.

One case study is worth naming, with the caveat attached up front. Opascope, an early advertiser, reported a meaningful return on roughly $60,000 in spend in June 2026. That's a useful data point and nothing more; it's a sample size of one, and it should carry exactly that much weight.

The thinness underneath all of this needs to be said plainly. Every figure available comes from early adopters who self-selected into a new channel, so survivorship bias is close to guaranteed: campaigns that failed don't get written up. No standardized definition of "conversion" exists across these reports either. A purchase, a lead form, a trial signup, and a page visit are all getting called the same thing in different places. Nobody knows the category mix among early advertisers, and if that first wave skews toward high-intent commercial verticals, the aggregate numbers will look better than they will once a messier, broader advertiser base shows up. One signal is harder to dismiss: visitors arriving from ChatGPT spend 60% to 80% more time on-site than visitors from social platforms. That's not a conversion rate, but it lines up with the argument that the traffic carries more intent walking in.

Building a working CPA estimate when no official benchmark exists

Three inputs are available, and they combine into a rough estimate, as long as that estimate gets treated as illustrative and not authoritative. The cost range: $3 to $5 CPC, or $25 to $60 CPM. The conversion range: 0.2% to 5.8% across industries broadly, 4% to 7% for well-matched commercial verticals. And the legacy benchmarks as anchors: Google's all-industry average near $59, Meta's all-industry median at $38.19.

Run the math at different points in the conversion range, holding CPC at a midpoint of the stated range. At a 4% conversion rate, implied CPA lands above Meta's median, above Meta's median but competitive with Google in mid-tier verticals. At 7%, implied CPA drops to a figure near Google's all-industry average, close to Google's all-industry average and genuinely attractive for high-intent categories. At 1.5%, representative of a poorly matched or early-funnel placement, implied CPA climbs past $260, uncompetitive against nearly every legacy channel available. These are constructions from available inputs, not published figures, and they should carry that label every time someone cites them.

What the range makes obvious: conversion rate, not CPC, does almost all the work here. The cost-per-click band is narrow; three to five dollars isn't much spread at all. The conversion-rate band is enormous, and that gap means targeting precision and creative relevance move CPA on this channel far more than bid strategy does. Anyone building a media plan around bid optimization on this platform is optimizing the wrong lever.

Vertical context changes how any of these numbers should read. A finance advertiser whose Google CPA already runs well above the channel average should look at a moderately elevated AI chat CPA very differently than an e-commerce advertiser whose Google CPA sits at $30. The same number is a bargain in one case and a wash in the other. And every estimate here carries the same caveat established earlier: because the ChatGPT dashboard undercounts conversions and can't report ROAS, any CPA figure pulled straight off the platform is probably a ceiling on true cost, not the cost itself. First-party tracking isn't optional if these numbers are going to mean anything.

What a reliable measurement setup looks like before running the first campaign

State the core problem plainly: the platform can report what a conversion cost, but it cannot report what that conversion was worth. ROAS only exists if the advertiser builds it, tying revenue data back to the campaign by hand.

That starts with first-party tracking, and it starts before launch, not after. UTM parameters belong on every ad, because AI chat referral traffic has to be distinguishable from organic ChatGPT use and from every other source hitting the same landing pages. Conversion events need a definition set in advance, one that matches how the same event gets defined on Google and Meta campaigns; otherwise cross-channel CPA comparisons are comparing different things while pretending they're the same metric.

Attribution model choice matters more here than on most channels. AI chat clicks often start a research process that resolves later, through a direct visit or a branded search days afterward. Last-click attribution will systematically undervalue this channel; time-decay or data-driven attribution fits how the behavior actually plays out.

While conversion data stays thin, post-click quality signals work as a decent interim proxy. Time on site, benchmarked against that 60% to 80% lift over social traffic, is one. Pages per session, bounce rate, and trial-signup rate are others, anything showing a visitor is further along in the decision without needing a full purchase to prove it.

Budget size matters too. Conversion rate variance across the available data is wide, and a small test budget on a channel with that much variance produces noise, not signal. Enough spend to generate a few dozen conversions is roughly the floor for something usable, rather than anecdote. Once that data comes in, the comparison that matters isn't against the provisional estimates built earlier in this piece. It's against the same advertiser's own Google and Meta CPA in the same vertical, and against revenue per customer, because that last comparison is the only one that answers whether the number is acceptable, not just whether it's competitive.

How the benchmark picture is likely to develop over the next 12 to 18 months

The spending trajectory suggests this gap closes faster than it looks right now. Even a modest share of projected 2026 spend, spread across a wider advertiser base than exists today, will generate the cross-advertiser volume that's currently missing.

Whether that data goes public is a separate question, and it comes down to OpenAI. Google built its benchmark reports over years; Meta already offers its own benchmarking tools to advertisers. If OpenAI follows that path and publishes aggregated performance data, the vacuum closes on a predictable schedule. If it doesn't, the market falls back on third-party aggregators and agency disclosures, a slower and noisier route to the same destination.

The broader AI assistant landscape complicates things further. ChatGPT is the most visible advertising surface right now, but it won't stay the only one, and CPA on one conversational surface won't reliably predict CPA on another. Audience makeup, conversational context, and auction dynamics differ enough between platforms that treating them as interchangeable is a mistake worth avoiding now, before habits form around it.

The intent-quality argument underpinning most of the optimism here is going to meet its real test as the advertiser base widens. Early conversion rates come from self-selected advertisers in commercial verticals who had the most to gain from getting in early. As auction competition rises, CPCs will climb, and conversion rates will likely compress toward the middle of the observed range instead of the top. The 4% to 7% figure may turn out to be an early-adopter number rather than a steady-state one, and treating it as permanent would be the same mistake advertisers made with early Facebook CPMs a decade ago. The signal worth watching from here isn't aggregate ad spend growth. It's the first vertical-specific CPA benchmark, finance, travel, SaaS, published with enough sample size behind it to trust.