ChatGPT Ads (OpenAI Ads) in 2026: How They Work, What They Cost & How to Run Them

A ChatGPT ad is a single sponsored card that appears below the assistant's response. It sits visually separated from the organic answer and carries a "sponsored" label. The card resembles a product or service listing, showing an item name, a price, and the sponsored tag. It can feature one or more items from a single advertiser. OpenAI keeps ads away from sensitive topics like health, mental health, and politics, and it suppresses them in Temporary Chats, logged-out sessions, and the ChatGPT Atlas browser.

Only logged-in adults on the Free and Go tiers see ads. Plus, Pro, Business, Enterprise, and Education users stay ad-free. Your reach skews toward people who haven't paid to remove the experience. OpenAI began testing ads in the U.S. on February 9, 2026, expanded to Canada, Australia, and New Zealand in March. It then announced rollout to the UK, Mexico, Brazil, Japan, and South Korea in May.

The pilot ran on an impression-based CPM model with a flat rate near $60 and minimum commitments around $200,000. When multiple advertisers compete for the same conversation, OpenAI runs a relevance-weighted second-price auction and shows the ad most relevant to the chat first. You pay just above the next-highest effective bid, and relevance acts as a multiplier rather than a tiebreaker. A tightly written ad on a $3 bid can outrank a generic ad on a $7 bid.

CPC bidding launched May 5, 2026 alongside self-serve access at ads.openai.com, with no minimum spend, a Conversions API, and pixel-based measurement. The minimum dropped in stages from $200,000 a month down to zero.

Named pilot brands include Target, Ford, Adobe, and Williams-Sonoma, with agency partners Dentsu, Omnicom, WPP, and Publicis building integration paths. The platform crossed $100 million in annualized revenue within six weeks and onboarded more than 600 advertisers. OpenAI is targeting $2.5 billion in ad revenue for 2026, so the spend is real and the buyer pool is growing fast.

Start at ads.openai.com, which opened to all US businesses on May 5, 2026 with no minimum spend. The advertiser registers the account in their own name. Agencies get invited in after the account exists, which trips up agencies used to standing up accounts on a client's behalf.

The ownership rule deserves a hard stop before you touch anything. If you are an agency, do not create the account. Have the client register it first, then add your team as a user. Account ownership cannot be transferred retroactively. A misstep here means tearing down and rebuilding from scratch.

Build the account structure

The hierarchy follows the pattern every paid search manager already knows. Campaigns sit at the top, ad groups underneath, and individual ads inside each group. Context hints attach at the ad group level, so plan your groups around distinct buyer situations rather than products. One group per landing page works cleanly because the system reads landing page content as part of its matching.

Pick a bidding objective

OpenAI gives you two objectives. Reach optimizes for impressions and bills on CPM, which suits awareness work where you want the sponsored card seen by people in the right conversation. Clicks optimizes for traffic and bills on CPC, which fits pipeline and signup goals. The CPC objective launched May 5, 2026 alongside self-serve access. Do not run both objectives in the same ad group.

Wire up measurement

Install the conversion pixel on your site and connect the Conversions API for server-side event tracking. The pixel captures post-click events like signups and purchases. CAPI backs it up when browser tracking degrades. OpenAI describes platform-reported conversions as a leading indicator only, so treat your CRM as the truth layer.

Tag every click with a consistent UTM scheme. Use utm_source=chatgpt, utm_medium=cpc or cpm, utm_campaign for the campaign name, and utm_content for the ad variant. These parameters persist through the click and give you attribution your analytics can read.

What you cannot connect yet

The Ads Manager runs as a standalone tool today. Integrations with Search Ads 360, Skai, and major MMM vendors are not available. Adobe and Criteo offer integration paths, and CRM connectors for Salesforce and HubSpot are slated for later in 2026. Until those ship, you manage OpenAI Ads outside your usual stack or pull the data through a tool that already supports it.

Synter's own live OpenAI Ads campaigns give the clearest picture of what these ads actually cost, because most published numbers are platform estimates rather than observed spend. Synter runs five separate geo campaigns from a single account (US, UK, Canada, Australia, New Zealand), each on impression-based CPM bidding with conversion events layered on top. The numbers below come from real per-day delivery windows in early-to-mid 2026.

CPM, CPC, and CTR in Synter's campaigns

CPM in Synter's campaigns sits in a $30 to $45 band across the four largest markets. The US peak day delivered 34,347 impressions and 116 clicks for $1,410, an implied CPM near $41 and a CPC of $12.16. The UK ran cheaper. One strong UK day produced 20,277 impressions and 99 clicks for $666, a CPC of $6.73 and a CPM around $33.

CPC lands between $5 and $13 once a campaign has real volume. Canada ran $6.54 to $11.37 on its better days. Australia ran $5.34 to $13.65. CTR across all markets falls between 0.3% and 0.75%, with roughly 0.5% as the typical figure. That CTR sits below paid search because the placement is contextual and in-conversation, closer to a high-intent native band than a typed-query result.

Thin ad groups distort the math. A low-volume Canada day spent $86 for 3 clicks against 2,802 impressions, a $28.62 CPC that means nothing except that the ad group lacked delivery. New Zealand, running a $1,000 budget across three ad groups, showed the same problem. Consolidate ad groups early so each one accumulates enough impressions to find a stable price.

How Synter's data compares to published benchmarks

The secondary numbers track Synter's observed ranges. OpenAI's pilot used a flat $60 CPM behind a $200,000 monthly minimum. The self-serve launch opened a $25 to $60 CPM range that varies by category (flyweel.co). On CPC, OpenAI recommends a $3 to $5 starting bid, and bids under $3 routinely fail to clear delivery. B2B campaigns run hotter, with some agencies reporting $8 to $15 CPCs because the buyer pool is narrower (justglobal.com).

The barrier to entry collapsed fast. The pilot minimum was $200,000 per month, dropped to $50,000 in April 2026, then fell to zero when self-serve opened on May 5, 2026 (flyweel.co). A meaningful diagnostic test now runs $5,000 to $15,000 over two to four weeks, enough volume to see whether your hints and landing pages clear the auction.

Read these figures as directional, not as platform averages. They come from one advertiser, Synter, in the paid-media and B2B SaaS category, and CPMs swing day to day as early fill stays volatile. Your CPCs will shift with vertical, geo, and creative. Budget in CPM terms, expect mid-single-digit to low-double-digit CPCs at volume, and watch starved ad groups inflate the number.

Context hints are natural-language phrases you write at the ad group level to describe the situations your buyer is in, not the products you sell. OpenAI's ad system treats them as semantic guidance to the auction, never as bid terms. The platform is explicit. "Context hints are not keywords. They do not guarantee placement in specific conversations and do not use exact-match logic." (justglobal.com)

The auction reads four inputs at once. Your context hints, your landing page content and URL, your headline and description copy, and your relevance-weighted second-price bid all feed the same decision. Get any one of them right and the system matches your ad to the right conversation.

That structure changes how you write the inputs. A Google keyword bids on a typed query and offers exact, phrase, and broad match types. A ChatGPT context hint reads the full conversation a person is working through and matches probabilistically with no match types at all.

You also get no audience segments and no user-level data of any kind, only aggregated performance metrics (justglobal.com). The conversation is the targeting signal.

Write hints as problems, not products

The format trips up most advertisers on the first try. "Project management software" fails because it names a category the user would never type into a chat. "Manager trying to reduce missed deadlines and improve team visibility" works because it describes the moment a buyer hits the problem your product solves.

Write from the user's perspective in plain conversational language, never marketing language. Make each hint specific enough to exclude irrelevant matches without narrowing reach to nothing. A practical trick. Ask ChatGPT how it describes your product category, then mirror that phrasing in your hints.

Treat the landing page as a targeting input

The selection system reads your landing page URL and content as part of the match. A generic homepage that covers ten use cases sends a muddy signal and weakens your auction eligibility for any single conversation.

Build one dedicated page per ad group, each one thematically tied to its hints. If your ad group targets sales forecasting conversations, point it at a page about forecasting, not your product overview. The closer the page reads to the conversation, the more often the auction picks your ad.

Google reads the query a user types into a search box. ChatGPT reads the full conversation a user is working through with the assistant. That single difference cascades through every other part of how the two platforms target, control, and measure ads.

Google gives you exact, phrase, and broad match types, negative keywords, and a documented auction with Quality Score. You bid on specific strings and you control which strings you exclude. ChatGPT offers none of that. Context hints act as semantic guidance, not bid terms, and OpenAI does not publish its auction model or expose a keyword-style negative list (justglobal.com).

User data is where the two platforms diverge most. Google layers demographic and behavioral signals on top of the query. ChatGPT hands advertisers conversation context only, with no demographics, no browsing history, and no cookie data. Advertisers receive aggregated views and clicks and nothing else (openai.com).

You should run both, at different points in the funnel. Google captures the user who already knows what to search for and types it. ChatGPT captures the user still framing the problem, working through "how do I fix this" before they ever phrase it as a query. Pair them and you reach the same buyer at two distinct moments, one where intent is explicit and one where it is still forming.

Build your campaigns around where the buyer is in their decision rather than around your product lineup. A buyer asking "how do I reduce missed deadlines" sits at a different distance from purchase than one searching "vendor pricing." Map each stage to a matching intent signal, CTA, and bidding goal, and keep those goals separate.

Never put awareness and conversion objectives in the same ad group (Just Global). The two attract different conversations and force the auction to optimize against itself.

Write creative that states the problem

Lead headlines with the problem or the outcome, not your brand name. "Transform Your Business" tells a reader nothing, so cut it. Write the description as one sentence of value plus a clear action, and drop superlatives like "#1 platform" or "industry-leading." OpenAI's ad system reads your copy as a matching signal, so vague language costs you both relevance and clicks.

Use product UI screenshots, dashboards, and workflow visuals instead of stock lifestyle photography. A picture of your actual results matches a buyer working through a real problem. A photo of someone smiling at a laptop matches nothing.

Know which categories can run

SaaS and cloud software, marketing and revenue technology, HR tech, analytics, security, and business productivity tools all qualify. Financial services, legal services, healthcare, and data brokerage stay restricted. Confirm your category before you spend setup time, because OpenAI also blocks sensitive conversation topics like health, mental health, and politics from carrying ads at all.

Close the measurement gap yourself

Click-based attribution loses data in low-cookie environments, so add a "How did you hear about us?" question to your post-conversion flow. That single field recovers attribution that UTM tracking misses on AI traffic. CRM connections to Salesforce, HubSpot, and Microsoft Dynamics are planned for later in 2026 for closed-loop reporting, so until then, track lead quality downstream rather than counting form fills.

Run a meaningful test now while the auction stays thin. Just Global flags the early-mover advantage directly. Competition will rise as more advertisers register, which means today's CPCs reflect a market that hasn't filled up yet. With 29% of B2B buyers already using AI tools as their primary discovery channel, the audience is real even though the auction is young.

Yes, and the spend is real. Synter's June 11, 2026 account report documented 129,663 impressions and $4,428 in spend across five markets from a single live advertiser account. Those numbers come from working campaigns with measured CPMs and CTRs, not a sandbox demo.

OpenAI's own pilot data points in the same direction. The company reports that ads have not measurably dented consumer trust in ChatGPT, that users dismiss ads at low rates, and that relevance keeps improving as the model learns which placements fit a conversation. Fewer than 20% of eligible users see an ad on any given day, which keeps the experience light enough that most people don't notice the ads.

The platform-level signals are louder. OpenAI Ads crossed $100 million in annualized revenue within six weeks of opening self-serve access, with more than 600 advertisers onboard. OpenAI projects $2.5 billion in ad revenue by the end of 2026 and $100 billion by 2030. Demand-side behavior backs the bet, with 29% of B2B buyers now using AI tools as a primary discovery channel.

What the data still doesn't show

Be skeptical of anyone quoting ChatGPT Ads ROAS right now. No named advertiser has published verified return-on-ad-spend figures. The public sources carry no vertical-specific CPC breakdowns beyond Synter's own first-party numbers. The flat-rate pilot pricing of $60 CPM tells you what early test budgets looked like, not what a SaaS or retail campaign actually returned.

Synter's benchmarks fill part of that gap, with one important caveat. The CTR range of roughly 0.3% to 0.75% and CPCs of $5 to $13 at volume reflect one advertiser in the paid-media and B2B SaaS category. Your CPMs and clearing prices will move with vertical, geo, and creative, and early fill stays volatile day to day.

The honest read is that ChatGPT Ads has moved past curiosity into a market with real money and named partners like Target, Ford, and Adobe. The performance literature is thin because the platform is young, so treat first-party data as your best guide and run your own diagnostic before you trust any published average.

OpenAI Ads Manager runs in its own window, with no connection to the rest of your stack. Search Ads 360, Skai, and the major mix-modeling vendors have not shipped named OpenAI Ads integrations. You pull impressions and spend from one tab, then paste them into a spreadsheet next to your Google and Meta numbers to see the full picture. That breaks the moment you run five geo-split campaigns at once.

Synter closes that gap with a named function, pull_openai_ads_performance, that reads impressions, clicks, spend, CTR, and CPC directly from your OpenAI Ads account. The same agents that read the data also build the campaigns. Synter structured its own live account as five country campaigns (US, UK, CA, AU, NZ) with topic-stacked ad groups beneath each, and reports every metric in the same view as Google, Meta, LinkedIn, Reddit, TikTok, X, Amazon, and Spotify.

The track record sits behind it. Synter connects to 20+ ad platforms through direct APIs, has managed more than $26 million in spend across 600+ advertisers, and documents CPA improvements above 70% (Synter blog). OpenAI Ads is one platform inside that set rather than a bolted-on export. You watch CPC drift on a thin New Zealand ad group and a heavy US campaign on the same dashboard, then act without switching tools.

Agents that execute, not just advise

Most cross-channel tools stop at recommendations and wait for a human to click approve. Synter calls its model Autonomous AI Execution. The agents make the changes inside the ad account directly, consolidating a low-volume ad group that spiked to a $28 CPC or shifting budget toward the UK campaign that returned a $5.41 CPC. Approval-first platforms like Roadway hand you a list of suggestions and leave the execution to you. Synter does the work in the account.

The same agents run from the command line. Synter's MCP server connects to Claude Desktop, so you can issue plain-language instructions across X Ads, Spotify Ads, Amazon Ads, and the rest. Ask it to compare CPM across your five OpenAI Ads markets, or to lift the daily cap on the campaign with the best CTR, and it executes against the live account. Running OpenAI Ads next to every other platform stops being a reporting chore and becomes one conversation.