ChatGPT Ads: What They Cost, Who Sees Them, and What They Can't Buy

How do ChatGPT ads work?

ChatGPT ads are paid placements that appear below a response — one slot per answer, clearly labeled sponsored — matched by "context hints" you write in plain language rather than keywords. They only reach users on the Free and Go tiers (plus logged-out sessions); Plus, Pro, Business, Enterprise, and Edu accounts never see them. Self-serve is live at ads.openai.com with a ~$25/day minimum and recommended starting CPC bids of $3–$5. Per OpenAI's own docs, ads run on a relevance-weighted second-price auction and cannot influence the answer itself.

TL;DR: Ads buy you the slot below the answer — never the answer. The data shows three ceilings: ads reach only the non-paying tiers, appear on roughly a quarter of commercial prompts, and buying them does nothing for your citations (88% of AI Mode advertisers weren't even cited for the keyword they paid for). Treat the slot as paid coverage for gaps — not a substitute for earned visibility, and never the foundation.

Seven months from "last resort" to a quarter of conversations

OpenAI spent years describing advertising as a last resort. The timeline of the last resort:

• January 16, 2026 — ads announced.
• February 9 — live test in the US, Free and Go tiers only, one labeled unit below the response.
• May 5 — self-serve Ads Manager opens, ending the pilot's ~$60 flat CPMs and $200K commitments; CPC bidding replaces the flat rate and the floor eventually lands at $25/day.
• June — Similarweb's tracking shows ads on 26% of US desktop conversations, up from 14% in May — a doubling in a month that outpaced every prior phase combined.
• August 10 — SE Ranking's 50,006-prompt study puts ads on 25.94% of commercial prompts, with one ad slot per answer.
• September — an Ads Manager plugin lets you build campaigns by chatting inside ChatGPT itself; impression-billed conversion optimization and flexible attribution windows go GA on September 16.

No paid channel has gone from pilot to a quarter of its surface this fast. Google Ads took years to become furniture on the results page; ChatGPT did it in months.

The industry split nobody priced in

While OpenAI and Google built the slot, two competitors walked the other way:

• Perplexity exited advertising entirely. After pioneering sponsored follow-up questions in 2024, it phased them out and in February 2026 told the Financial Times (via The Verge) it had no plans to return. An executive's reasoning: "a user would just start doubting everything." Perplexity serves ~780M monthly queries with no paid layer at all.
• Anthropic committed to keeping Claude ad-free — and bought Super Bowl airtime to say it, which OpenAI's CEO called "dishonest." (An anti-ads message delivered by the most expensive ads in television. Nobody said the discourse had to be consistent.)
• Google went the opposite direction: ads inside AI Mode, bidding on "AI intent categories" instead of keywords, Search and PMax campaigns auto-extending into the surface — and as of September 1, Google began auto-migrating legacy Search campaigns to AI Max. The paid layer is being moved under advertisers whether they opted in or not.

This is the first time in search history the monetization model itself diverged. On Google, ads and organic competed for the same click on the same page. In AI answers, the platforms can't even agree on whether money should touch the surface — which means "advertise on AI" isn't one decision. It's a different decision per platform, with different rules, different audiences, and different ceilings.

The two-layer model

Strip the product names away and every AI surface has the same structure:

• The answer — the response, its brand mentions, its cited sources. Earned only. OpenAI mandates ads can't influence it; Google's AI Mode cites independently of who bought the slot below; Perplexity and Claude have no paid layer to confuse it with. This is the layer our ChatGPT SEO playbook is written against.
• The slot — the labeled placement below or beside the answer. Bought. Contextual, fast, measurable, and bounded in ways that matter.

The sentence worth keeping: the answer is earned; the slot is bought. Everything confusing about AI advertising comes from pricing the slot as if it bought the answer.

Three ceilings the paid layer can't break

1. The audience ceiling: ads can't reach the users with budgets

ChatGPT ads only serve on Free and Go. Plus, Pro, Business, Enterprise, and Edu accounts — the users most likely to have software budgets, procurement authority, and a company card — are structurally unreachable by the paid layer. For B2B especially: the closer a user is to a buying decision, the more likely they're on a tier where your ad cannot appear. The earned layer is the only way in.

2. The coverage ceiling: ~26% of commercial prompts, one slot

SE Ranking measured 25.94% of commercial prompts carrying an ad — and exactly one slot per answer. The other three-quarters of commercial prompts, plus every informational and navigational prompt, are earned-only territory. Ad presence also swings by niche — SE Ranking's AI Mode study found a range from 72% of queries down to under 3%, with lead-gen verticals heavy and YMYL/informational verticals nearly empty. If your niche is in the thin end, the paid layer barely exists for you.

3. The causation ceiling: buying the slot doesn't get you into the answer

This is the finding that should kill the "just buy visibility" argument. In SE Ranking's AI Mode data, the advertiser's domain was absent from the cited sources on 88% of queries carrying an ad — only 11.53% of advertiser domains were also cited for the same keyword, and the exact URL overlap was 1.95%. Around 85% of advertisers had no organic presence at all on the keywords they paid for.

The ordering makes it worse: the ad sits below the answer. If the answer recommends a competitor by name, your sponsored card renders underneath their endorsement. A paid slot under someone else's earned citation is a strange place to spend money.

When the slot is the right buy

Honest version, because the ceilings don't make ads useless — they bound what ads can do:

• Speed. Earned visibility compounds over months; a campaign is live in days. Launches, seasonal windows, and "we need to exist in this conversation next week" are real use cases.
• Gap coverage. Prompts where you have zero earned presence but demand clearly exists — the slot is a patch while the earned layer builds. It's a patch, not a fix; 85% of advertisers use it as a permanent one.
• Catalog commerce. Product-feed ads and Shopify integration make ChatGPT a real retail surface for stores with clean product data — closer to Shopping than to Search.
• Cheap signal. $25/day buys you delivery data on which conversations OpenAI's system thinks you're relevant to — useful intelligence even if the clicks are thin.

One caveat from the field: SE Ranking ran real campaigns — ~97,000 impressions, 1,263 clicks (~1.3% CTR), and "very few sign-ups." Early evidence says the channel delivers clicks better than conversions right now. Judge it on cost per conversion against your other channels, not on novelty.

How to run the decision

1. Measure the earned layer first. Run your 10–20 buyer prompts through ChatGPT, AI Mode, and Perplexity and log where you appear in the answers — the tracking walkthrough covers the method. You cannot evaluate paid coverage without an organic baseline.
2. Sort prompts into three buckets. Already cited → defend, don't spend. Cited but volatile → earn stability. Absent with clear demand → this is where the slot earns its budget.
3. If you buy, buy narrow. Context hints are situation descriptions, not keywords — write them like the prompt you wish you were cited on. Keep landing pages crawlable: OAI-AdsBot must be able to read your landing page or the ad gets rejected — the same crawl access the earned layer requires. The plumbing is shared even though the layers aren't.
4. Watch the mismatch rate. About 1 in 7 ads appears on an off-topic prompt — over 50% in niches like News & Politics. Contextual targeting is phrase match's younger, more confident sibling.
5. Re-read the ceiling monthly. The platform shipped meaningful changes on September 3 and September 16 alone. Anything in this post with a date on it has a shelf life — verify against OpenAI's ads docs before spending.

Where this falls short

• "Ads don't influence answers" is policy, not physics. OpenAI says the systems are separate and Perplexity's exit suggests the trust cost is real — but the incentive to blur the line will only grow with the revenue. If the wall ever moves, the earned layer's value goes up, not down.
• The CTR is structural. The ad arrives after the answer satisfied the query — Topify's analysis puts observed CTR near 0.9% and calls it a design constraint, not a creative problem.
• Sponsored Agents change the math if they ship. OpenAI is testing conversational ad units — a sponsored agent users can talk to after seeing your ad. If that lands, the slot stops being a banner and starts being a funnel. Watch it.
• None of this is stable. Penetration doubled in a month; auction mechanics, tiers, and formats are quarterly-moving targets. The two-layer model is stable; every number attached to it isn't.

Frequently Asked Questions

Can you pay to get cited in ChatGPT's answers?

No. OpenAI mandates a hard separation: ads appear below the response and cannot influence, rank, or alter the answer. The empirical check agrees — SE Ranking found only 11.53% of AI Mode advertisers' domains were also cited for the same keyword. The answer is earned; the slot below it is bought.

How much do ChatGPT ads cost?

Self-serve has no contract or minimum commitment — a ~$25/day campaign minimum, with CPC as the common bid model and $3–$5 recommended starting bids per OpenAI's documentation. The pilot era of $60 flat CPMs and $200K commitments ended when Ads Manager opened in May 2026.

Who actually sees ads in ChatGPT?

Users on Free and Go plans, including logged-out sessions, in supported markets (US, UK, Canada, Australia, Japan, South Korea, New Zealand). Plus, Pro, Business, Enterprise, and Edu accounts see no ads — and neither do Perplexity or Claude users, since both platforms run without a paid layer.

Should we run ChatGPT ads or invest in AI visibility?

Sequence it, don't choose blindly: baseline your earned citations first, then buy the slot only where earned presence is absent and demand is proven. If you're already cited, the marginal ad is coverage for Free-tier users only — the decision framework in our GEO vs SEO breakdown applies one layer up.

Do ChatGPT ads work for agencies and B2B?

The structure is hostile to B2B — the buyers with budgets sit on ad-free tiers — but narrow use cases survive: launches, geographic pushes, and catalog-style offers where the Free audience is the actual market. For most B2B visibility goals the earned layer is the primary channel; the slot is a supplement with a ceiling.

---

The answer is earned; the slot is bought. Managed builds the earned layer — the part that can't be outbid. The platform runs it on your infrastructure.