The ChatGPT Ads Playbook: Formats, Specs and Setup in 2026

OpenAI's advertising platform is no longer a pilot. ChatGPT ads launched in the US on February 9, 2026, expanded to 40+ countries by August, went into public beta across Europe in the same month, and self-serve access is rolling out through September. With 700–900 million weekly ChatGPT users, this is the new attention surface — and it still carries early-adopter pricing. This is the complete operator's playbook: formats, specs, budgets, targeting and measurement, in plain language.

First, how is this different from Google or Meta ads?

On Google, you bid on keywords; on Meta, you target people by behavior and interests. On ChatGPT, there are no keywords and no interest targeting built the classic way. Instead, OpenAI reads the conversation itself — if your business context hints match what the user is asking about, your ad becomes eligible. One ad team, no bidding on 'organic+paid doublespeak' — your job is relevance, not reach engineering.

• Ads appear contextually, inside the conversation the user is having
• Relevance decides eligibility; no demographic or interest lists in the EU for now
• The format behaves like an answer, so good copy reads like a recommendation

The three ad formats

OpenAI has productized three placements. The sponsored card appears below the ChatGPT answer as a context-matched unit. The 'Ask ChatGPT about this ad' follow-up lets a user keep the conversation with the brand inside the product. And product feed ads serve commerce-style units for merchants. None of them use keywords in the Google sense — targeting runs on context hints instead.

• Sponsored card below the answer, labeled to stay distinct from the response
• Interactable 'continue the conversation with the ad' follow-ups
• Product feed units for retailers with uploaded catalogs
• No keyword matching — context hints define eligibility

Context hints: your replacement for keywords

Instead of typing 'best salon Bengaluru' as a keyword, you write context hints — short phrases that describe the situations in which your ad should be eligible. Think of them as descriptions of your ideal customer's question: 'salon for bridal makeover', 'home cleaning service twice a month'. You can add up to 2,000 characters of hints per ad group, and OpenAI's system matches them intelligently against the conversation context.

• Write hints as the customer's question, not ad keywords
• Cover the variants your customers actually use in conversation
• Keep hints specific to one intent per ad group for cleaner data
• Use full 2,000 characters if your business has many use cases

Creative specs (the rules that matter)

The format rewards brevity. Titles can reach 50 characters but 16–24 performs best; descriptions go up to 100 with a 32–48 sweet spot. Images must be square, at least 256×256 and no larger than 1200px, and OpenAI rejects assets with baked-in text — copy belongs in the fields, not the image. A simple rule: if your headline reads like the most helpful answer under a question, it's right.

• Title: max 50 chars; target 16–24 for readability
• Description: max 100 chars; target 32–48
• Square image, 256×256 minimum, 1200px maximum
• No text rendered inside the image asset

Bidding, budgets and targeting

You can bid on CPC or CPM, and pricing is relevance-weighted second-price — the more relevant your ad, the less you pay per outcome. Early CPCs start around $3–5. Minimum daily budgets are $25 in the US and €15 in the EU, with a CPA objective in testing. Targeting uses context hints (up to 2,000 per ad group), country-level geo with US state/DMA/ZIP detail, platform selection, and custom audiences of 25,000+ for inclusion. EU guardrails: no demographic or retargeting targeting in Europe for now, and ads only surface on Free and Go tiers — Plus, Pro, Business and Enterprise stay ad-free.

• CPC and CPM bidding, relevance-weighted second-price auction
• Starting CPCs roughly $3–5 in competitive niches
• Min budget $25/day (US) or €15/day (EU)
• Context hints cap at 2,000 per ad group; CPA objective in testing
• EU/EEA: no demographics or retargeting; self-serve expanding Sept 2026

Measurement and the trust question

Tracking uses the familiar web stack: an OpenAI pixel plus Conversions API, a delivered-conversion parameter and UTM support — though attribution is short by design, so expect a last-click bias. On the trust side, OpenAI is explicit that ads never influence model answers; placements are separated from responses and labeled as sponsored. That separation is what makes the format work long-term.

• Pixel + Conversions API for server-side tracking
• UTMs supported; attribution window is short
• Ads never shape model answers — sponsored units stay separate
• Users on Free and Go tiers see ads; paid subscribers don't

How to structure your first campaigns

Map context hints to the purchase-intent patterns your ICP already asks AI ('best X in Bengaluru', 'X vs Y price'), build one ad group per intent cluster, and install the pixel and CAPI before launch so learning doesn't start from zero. Run the CPC baseline for two weeks before layering CPM for presence, and keep titles factual — the audience is actively mid-question and reads your ad as an answer.

• Audit recent questions your sales team hears — those are your first hints
• One ad group per intent cluster, hints tailored to that cluster
• Install pixel + CAPI before launch day so learnings accumulate
• 2-week CPC baseline, then add CPM for presence

Bottom line

ChatGPT ads combine a massive engaged audience with near-zero advertiser saturation in India, and the answer-like format rewards useful, specific copy. ADZBE is one of the early agencies running OpenAI placements. Book a free audit and we'll map whether ChatGPT ads fit your funnel before the land-rush pricing fades.