Paying to reach curious readers in 2026: ChatGPT ads, feed ads, paid newsletter recommendations, and the labelling rules

Advertising inside AI assistants went from announcement to a working market within this year. It is a real place to buy attention, but not yet a cheap or well-measured one, and the older channels still undercut it. Reading the prices side by side, and the rules that now govern what an ad may pretend to be, shows where a curious reader can actually be bought, and on what terms.

1ChatGPT became an ad platform in 2026, with thin inventory

The timeline is short. OpenAI announced them Jan 16, 2026, and they went live Feb 9, 2026.TL;DR Buying started expensive and closed: Ads launched with a $60 CPM and a reported six-figure minimum.Key findings CPM is the price of impressions, the standard unit for ad space. Within months the door opened: Self-serve opened May 5 with no minimum spend.TL;DR By the end of summer the business was large: ChatGPT ads reached a $1B annualized run rate "in less than 200 days after launch," according to OpenAI's August 31, 2026 announcement.TL;DR

The format is unlike a search ad. The ads are sponsored cards below answers, targeted by plain-language "context hints" rather than keywords.TL;DR The rules around them are strict: No political ads. Health and finance outside the US generally prohibited. No ads in Temporary Chats, to logged-out users, to under-18s or to paid tiers.Details And the early experience was rough: Early advertisers complained they couldn't prove results or even spend their budgets, because inventory was thin.Key findings

The audience is mostly not shopping. OpenAI's head of global ads solutions said that roughly 20% of queries in ChatGPT have a direct commercial intentKey findings, which the research reads plainly: So most conversations are curiosity, not shopping.Key findings For a Canadian buyer there is an extra wrinkle: Canada has had ads since spring 2026, but whether self-serve buying works for a Canadian account is reported inconsistentlyTL;DR.

2Perplexity left advertising; Google and Microsoft are moving in

Not every AI answer engine wants the money. Perplexity stopped taking advertisers in October 2025 and confirmed in February 2026 that it was out.Key findings An executive explained why to the Financial Times: A user needs to believe this is the best possible answer... The challenge with ads is that a user would just start doubting everything.Key findings

The larger companies made the opposite call. Google (AI Mode formats announced May 20, 2026, testing in the US) and Microsoft (Copilot ads via Microsoft Advertising; AI Max for Search pilot April 21, 2026) are going the other way.Key findings Their formats are closer to search advertising than to content, and the performance claims are their own: Microsoft claims 73% higher CTR than traditional search (first-party, unverified)Details

3What a curious reader costs on each platform

Against the feeds, the assistant slot is expensive. The research's paid inventory map sets the prices side by side:

Meta is the reference point: Gupta Media's tracker put Meta at a $6.59 CPM and $0.68 per link click (October 2025).Key findings On the same measure ChatGPT's $25–60 CPM puts it in LinkedIn territory.Key findings Reddit sits at the cheap end with an audience that suits ideas: Reddit reportedly runs a $3–12 CPM with a median CPC around $1.25 (agency data), with an older, more technical audience.Key findings Its users are also the least forgiving of advertising that reads like advertising: Promoted posts must read like native posts. Community tone is unforgiving; give the full answer in the post.Details

Prices move through the year, and the figures are not audited. CPM seasonality: Q4 spikes of up to about 66%.Details Most cost benchmarks come from agencies or vendors (Gupta, Stackmatix, Newsletrix, SparkLoop, Billo). Treat them as ranges, not promises. ChatGPT ad pricing changes monthly.Caveats

4For a newsletter, the measure is a reader who keeps opening it

Impressions and clicks matter less to a newsletter than the price of a subscriber who reads. One operator acquired engaged subscribers on Meta at $0.65–$1.60.Key findings Another reports large B2C newsletters happily paying $3–$5 at scale, while some clients need under $1.25 to break even on day one.Key findings

The other route is to pay other newsletters to recommend yours. Beehiiv lists a $1.63 average cost per subscriber for Boosts (early-access data) and a $50 minimum.Key findings A vendor analysis puts SparkLoop at $4–7, but with 35–45% first-30-day open rates versus 22–30% for Meta lead ads.Key findings Recommendations can also pay the other way, which is how some publishers offset their ad spend: One customer says "we're actually making more money from paid recs than we're spending on ads."Details That's a vendor testimonial, but the mechanism is straightforward.Details

The measurement trap is the cheap signup. The research's rule: Judge by cost per engaged subscriber, not raw signups.Details The reason: One operator found the cheapest signups were people with no interest in newsletters, and that existing subscribers clicking ads made CPA look artificially low.Details Hard numbers are scarce in any case: Public, audited payback numbers from media companies are rare. Most figures come from operators and vendors.Details

5An ad that gives the whole answer selects its readers

The common practice is to pay only for what has already worked. This is the "test free, pay for proven" loop newsletter operators use.Details The research's argument for making the ad itself useful is about who it attracts. Giving the complete answer in the feed, rather than teasing it, reverses the curiosity-gap clickbait patternDetails, and People who subscribe after being fully satisfied are self-selecting for the product, which is why their open rates hold.Details That is reasoning, not a measured result. The honest version of a reason to come back is a promise with a date: Clickbait withholds; open loops promise a dated resolution and deliver it.Details

The economics are tight either way. The loop breaks when revenue per subscriber (often only $1–4 a year from sponsorship alone) can't repay CAC before churn.TL;DR CAC is what it cost to acquire the reader. Sponsorship needs a list of some size before it pays at all: Sponsorship also requires scale: sponsors rarely buy lists under a few thousand subscribers.Details Readers bought with ads tend to leave sooner: Paid cohorts churn faster than organic ones.Details though the research flags that rule as a heuristic rather than a study. Paid tiers convert slowly and lose readers steadily: 0.62% median free→paid conversion (beehiiv)Details and beehiiv reports monthly churn of 5.06% to 16.67% depending on vertical.Details

6The 31 percent figure measures minutes of scrolling, not people

A common shorthand says a large share of scrollers have attention problems. The study behind it measured something narrower: Allcott, Gentzkow and Song ("Digital Addiction," AER 2022; about 2,000 US adults, 2020)Key findings estimate that self-control problems account for an average of 31 percent (48 minutes per day)Key findings of social-media time. It is a share of minutes, not a share of people, and it isn't a clinical attention diagnosis.Key findings The effect is also uneven: for 22% of participants it adds less than 10 minutes a dayKey findings. For anyone buying attention, the distinction is the difference between reaching curious people and harvesting compulsive minutes.

7The label stays; the rules target fake people and hidden sponsors

Every paid placement carries its label, and the research treats that as no obstacle to useful content. Deception starts when you hide who's speaking or why: fake users, fake reviews, undisclosed paid relationships.Details In the United States the FTC's rule bans reviews and testimonials by people who don't exist, "such as AI-generated fake reviews."Details It also covers fake social-media indicators such as bought followers or views.Details Civil penalties run up to $53,088 per violation.Details Enforcement has begun: The FTC sent its first warning letters to 10 companies on Dec 22, 2025.Details

Canada's rule is framed around disclosure: under the Competition Act anyone promoting a business interest must disclose material connections "as visible as possible," and reviews must reflect real experience.Details The advertising standards body extends that to synthetic presenters: An AI influencer still has to disclose brand connections and can't give testimonials about experiences it can't have.Details Europe's rules reach the same point by another route: Deepfake audio and video must be disclosed as AI-generated when shown to EU users.Details

The platforms enforce their own versions. Meta auto-labels AI-edited ads with "AI info."Details OpenAI prohibits ads that imitate ChatGPT's interface or voice, or use false endorsements.Details Readers want the disclosure even when they cannot spot the difference. In an Ipsos test, only 25% could confidently tell which ads were AI-made, but 79% said companies should disclose AI useKey findings. Canadian results run the same way: In Canada, the Ad Standards/Caddle survey found 78% say AI influencers must disclose their identity, and 57% say AI reduces their likelihood to buy.Key findings Not everyone minds: Klaviyo's March 2026 survey of 7,998 consumers in 8 countries found that 32% say brand use of AI-generated content makes them trust brands less, while 61% are neutral.Details

8Synthetic crowds work as test rigs, not as audiences

Cheap AI-made content invites a shortcut: fill the room with synthetic people. The experiments so far argue against it. The agent-only network Moltbook was a sharp lesson: It went viral partly on posts that turned out to be humans posing as agents through an insecure API.Key findings A research version built to study such networks found they reproduce the worst of human ones. With only posting, reposting and following, it reproduced echo chambers, elite concentration and amplification of polarized voices, and the fixes they tested helped only modestly.Key findings

Feeding machines their own output degrades them as well. A study in Nature found that training on recursively generated data causes irreversible defects… tails of the original content distribution disappear.Details The research draws the consequence: An AI-only feed homogenizes, and human-origin data becomes more valuable.Details Synthetic panels can still help rank ideas before money is spent. Argyle et al. report correlations around 0.90–0.94 on ANES itemsDetails, but they show less variance than real people, and regression relationships can differ or flip sign.Details The research's verdict: Synthetic societies are useful as mirrors and test rigs, not as fake crowds.Key findings

9AI-assistant ads sell a reader mid-question, at a premium

It explains why the new ad market has not displaced the old one. What an assistant sells is a reader caught in the middle of a question, and it charges accordingly, with thin inventory and weak measurement while it learns. The cheap curious readers are still in the feeds and in other newsletters' recommendations, and the cost that matters is the one for a reader who keeps opening. The research's own advice for a small publisher follows from the prices: Use them as a small, high-intent probe, not the main channel.Bottom line

The rest of the finding is about honesty rather than price. An ad can be the whole of something worth reading and still say it is an ad. The rules, the platforms and the readers all draw the line in the same place: at fake people, invented testimony and hidden sponsors, not at ads that are useful.