The Complete Guide to OpenAI Ads.

1. What OpenAI Ads is (and isn't).

OpenAI Ads is OpenAI's self-serve advertising platform. Advertisers buy placements that appear as labeled "Sponsored" cards directly beneath the organic answers ChatGPT returns to a user's prompt. Both cost-per-click (CPC) and cost-per-impression (CPM) bidding are supported. The platform went GA in May 2026; the previous $50,000 minimum spend was removed at GA, opening the surface to any advertiser.

It is not: a sidebar widget, a Google Display ad, a programmatic banner, or a sponsored chatbot answer. The organic answer comes from the model. The sponsored card sits beneath it.

The honest mental model: imagine a Google Search ad that runs after the answer instead of before it.

2. How the auction actually works.

When a user submits a prompt, OpenAI's classifier maps it to a topic and intent — a "prompt family." Eligible advertisers compete for the limited number of sponsored slots beneath that answer. Standard second-price auction logic applies, with intent quality (landing experience, click-through rate history, relevance) factored into rank — the familiar Quality Score idea, applied to a conversational unit.

Three things to know:

• The slot count varies. Most answers get one slot. High-commercial-intent answers get up to three.
• There is no "search term." The full prompt is the unit of intent. We map prompts to families and bid families.
• Relevance compounds. Strong CTR + landing experience reduces effective CPC over time. Same shape as Google.

3. CPC vs CPM: when to use which.

Use CPC when the goal is a transactional conversion you can attribute to the click — purchase, sign-up, demo, application. CPC is the dominant choice in the platform today.

Use CPM when the goal is brand impression in a category — top-of-funnel for a launch, awareness for a high-consideration B2B. CPM is becoming a real choice for brand-led campaigns, particularly with the inclusion of brand-safety adjacency reporting.

In our accounts, the typical split is 80% CPC, 20% CPM. We test the inverse on brand-led engagements.

4. Creative for the conversational unit.

The ad unit is a card with a headline, an optional second line, structured detail rows, and a destination URL. The user reads it in the flow of an answer, not on a SERP grid. So the creative rules are different.

• Headlines must answer. "Best ergonomic chair for tall people" works. "Get a great chair" does not.
• Structured detail rows. Price, key features, location — readable in 1–2 seconds.
• No marketing pad. "Industry-leading" and "world-class" reduce click-through measurably. Cut them.
• Brand voice fit. The ad sits in the AI's voice. Don't fight the register — write in it.

We A/B from day one. Three headline variants minimum per ad group, rotated weekly.

5. Prompt families, not keywords.

Stop thinking about keywords. Start thinking about prompt families — clusters of full-sentence requests that share intent and category.

Example for a running-shoe brand: a single prompt family might cover "best running shoes for flat feet", "recommend a stability running shoe for overpronation", "shoes for runners with plantar fasciitis" — all the same buying moment. Build the family, write to it, bid it.

Negative prompts work like negative keywords but at the family level. Curate weekly.

6. Measurement and attribution.

The platform reports clicks, impressions, CTR, CPC, and conversions if you've configured conversion tracking. Default conversion tracking uses platform pixel + UTM. Server-side conversion tracking is supported and recommended — particularly for direct-response B2B and long sales cycles.

Most accounts under-attribute. We recommend:

• Server-side events (CAPI-equivalent) wherever your stack supports it.
• UTM strategy that distinguishes prompt-family at the campaign level.
• Cross-channel measurement model — multi-touch or MMM — for accounts where the AI-ad click is part of a longer journey.

7. What it actually costs.

Typical CPC ranges as of mid-2026:

• DTC ecommerce: $3–$8 per click
• B2B SaaS / finserv: $8–$25 per click
• Local services: $1–$5 per click
• Premium B2B (legal, enterprise infra): $25–$60 per click

CPMs run roughly half the equivalent Google Display CPM, ~1.5x typical Meta CPM. Auction density is still light in most categories — that will change.

8. How it compares to Google + Meta.

Vs Google Search: Higher intent on consideration/recommendation queries. Lower CPC in most categories today. Different attribution shape — the click lands in deeper-consideration mode.

Vs Meta: Different funnel entirely. Meta is interruptive discovery; OpenAI Ads is intent-led. Don't pit them — run both.

Vs Perplexity: Same buyer in different research modes. Run both for any consideration-heavy category.

9. A 30-day launch playbook.

1. Days 1–3: Map prompt families for your category. Validate them by typing into ChatGPT manually.
2. Days 4–7: Stand up the account. Three campaigns, 10 ad groups. Three headline variants each.
3. Days 8–14: Launch with a daily budget cap. Daily bid + negative-prompt curation. Watch CTR.
4. Days 15–21: Read signal. Kill ad groups with CTR below 2%. Test new headline variants on the winners.
5. Days 22–30: Scale spend on the campaigns hitting your CAC ceiling. First written report. Decision: continue, scale, or wind down.

10. Brand safety, policy, compliance.

OpenAI Ads enforces category-level placement controls and brand-safety filters. Advertisers get adjacency reports. Health, finance, and regulated categories require compliant landing pages and substantiated claims.

Restricted categories at GA: alcohol (jurisdiction-bound), pharmaceuticals (prescription-only restricted), gambling (jurisdiction-bound), weapons, surveillance, cryptocurrency (most jurisdictions). The list will evolve — we track changes monthly.

11. What's next: Perplexity, Gemini, AIO.

Perplexity is in US beta with sponsored related-questions and brand-answer slots. Gemini pulls Google Ads inventory into conversational answers. Google AIO has no paid slot yet but is the most cited surface on the web. We expect at least one more LLM ad surface to ship before end of 2026 — possibly Anthropic Claude.

The multi-LLM playbook is the right way to plan for this. See our multi-LLM service.

12. Frequently asked questions.