Leadership1 distinct publisher3 min readUpdated
OpenAI's sponsored cards are open to almost any advertiser, but the inventory sits on free and Go accounts and the only dials are country and context.
The Board Room · Leadership desk

Compiled by The Board RoomSomething wrong?How this is made
OpenAI has turned ChatGPT into an advertising surface, with the assistant rendering a sponsored card alongside the answer it generates [1]. Since May, according to an Entrepreneur contributor account, its Ads Manager has been open to more or less any business willing to name a price per click or impression, which makes this the first genuinely new buyable acquisition channel in years [10].
The unit itself is modest: a card with a headline, copy, an image and a link, sitting below the answer rather than inside it [2]. OpenAI has been firm that ads do not change what the model tells the user [3]. Treat that as the product boundary it is. You are not buying a place in the recommendation; you are buying a slot next to one.
The inventory is narrower than the ChatGPT user base implies. Ads appear to free-plan users and to the cheaper Go tier, while Plus, Pro and higher plans see none, and OpenAI says only a small portion of free users are shown ads at all [4]. So the reachable audience is the unmonetised tier, not the power users, and the volume is small enough that the author frames it as a few motivated users rather than millions of impressions [4].
Then the part that breaks imported playbooks. Targeting is by country or by "context," meaning the general subject or need, and not by keyword or demographic [5]. You can upload first-party data such as a customer list to target or suppress those people, and that is roughly the extent of the levers [6]. OpenAI does not hand conversations to advertisers [7], which leaves measurement entirely on your side of the click. A team whose competence lives in query-level bidding and audience layering has nothing to do here; the work moves back to the offer and the creative. The same account argues the channel favours a clear offer over a vague one, and fits best when someone is comparing options or researching a decision [8][14].
Cost of entry has collapsed. Reports cited in the piece say OpenAI initially demanded a $200,000 minimum to access its ad test, which excluded most small businesses [9]; the floor now described is about $25 a day [11]. That old minimum equals roughly 8,000 days of spend at today's floor [1]. Demand signals look real if second-hand: the same article cites industry news that the test reached a $100 million annual run rate after six weeks [12], and says Omnicom and Publicis have built ChatGPT ad strategies for clients [13]. At the $25 daily floor, a $100 million annualised figure would take on the order of 11,000 advertisers spending the minimum year-round, so the reported revenue is being carried by larger budgets, not the long tail [2].
One caution on provenance: all of the above comes from a single contributor piece whose author says they ran ads for their own clients, with the $200,000 and $100 million figures relayed from other reporting rather than from OpenAI [16][9][12].
What to watch: whether targeting granularity widens beyond country and context, whether ads ever reach paid tiers, and whether OpenAI publishes reach or impression data that would let a buyer model this channel rather than guess at it [5][4].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
OpenAI turned ChatGPT into an advertising tool, with the system showing a sponsor message card whenever it generates an answer to a user's question.
The sponsored message resembles a business card containing a headline, copy, an image and a link, presented as an additional option below the answer.
OpenAI has been firm that ads will never change what ChatGPT actually tells the user; the answer to the prompt is not impacted by the ad.
Ads are shown to ChatGPT free-plan users and to users of the cheaper Go version; Plus, Pro and better plans see no ads, and OpenAI says just a small portion of free users see the ads, so the piece describes it as a few motivated users rather than millions of impressions.
The platform allows targeting either by country or by 'context', meaning the general subject or need, but not by particular keywords and not by demographics.
Advertisers have very limited optimisation options; first-party data such as a customer list can be uploaded to target those customers or exclude them from seeing ads, and that is roughly the extent of it.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single practitioner column; key numbers unattributed
One publisher, one contributor opinion piece. The product-mechanics claims (card format, free/Go-only inventory, country-or-context targeting, first-party-list-only optimisation, no conversation sharing, $25/day floor) are stated first-hand and internally consistent, but the two headline commercial figures — the $100M annualised level and the $200,000 test minimum — are sourced only to unnamed 'industry news' and 'reports', and OpenAI's assurances are relayed without documentation or independent verification.
Generally available and being trialled, but shallow inventory and weak reported results
There are several concrete adoption signals: self-serve availability to nearly all businesses since May, live serving on free and Go tiers, agency programmes, and the author's own client campaigns. Depth is limited: exposure is confined to a small share of free-tier users, the only volume figure is second-hand, and the one first-hand performance report says traffic under-converted, so this reads as broad availability with early, unproven usage.
Mildly overstated: platform framing outruns verified reach and results
The framing that ChatGPT 'is an advertising platform now', anchored by an unattributed $100M annualised figure and agency-programme name-drops, sits above what the piece itself evidences: ads on free and Go accounts only, seen by a small portion of free users, with no keyword or demographic targeting, no conversation data for measurement, and the author's own campaigns under-converting. The article does partly self-correct ('not millions of impressions... a few motivated users'), which keeps the gap modest rather than large.
Disclosed practitioner and contributor-platform incentives
The author is a marketing practitioner who discloses running ChatGPT ads for clients, so they benefit professionally from advertiser attention to a new channel they already operate in; the piece also runs under Entrepreneur's contributor programme, whose bylines commonly serve author visibility. Disclosure is explicit, and the column includes unflattering detail (poor conversion, thin targeting), which tempers but does not remove the incentive.
Low: one publisher, no corroboration of the load-bearing numbers
Confidence is limited by a single-source cluster with no cross-publisher check, several load-bearing figures attributed to unnamed third parties, and an author with a disclosed commercial stake. Confidence is not lower because the operational mechanics are specific, internally coherent, fresh (published 2026-08-19) and partly grounded in first-hand campaign experience.
build
OpenAI Adds CPC Bidding to ChatGPT Ads, and There Are No Keywords to Bid On1 distinct publisher
science
Text watermarks land on 2 December. The detection they imply does not.1 distinct publisher
invest
A Connecticut judge just priced prompt injection: no fine, no e-filing2 distinct publishers
science
Claude's watermark is a compliance artefact, not a cheating detector1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 19, 2026