The ChatGPT Ad That Sounds Like Real Advice

The AI ad revolution finally arrived. Five months in, it's moved faster than almost anyone predicted ,from a $200,000 enterprise pilot to a self-serve platform with no minimum spend at all. Here's what actually changed, and what it means for your next move.

Picture this. You're planning a kitchen renovation. You open ChatGPT, type out your situation ,your budget, your timeline, your hatred of open shelving ,and ask for advice. The AI gives you a thorough, helpful answer. Then, just below it, clearly labelled, sits a sponsored placement from a home improvement brand that paid to be there.

Not woven into the answer. Not disguised as advice. Separate, visible, and marked.

That distinction ,between the answer and the ad ,is still the defining design decision of conversational advertising in 2026. But the platform underneath it has moved through more change in five months than most ad platforms see in five years. This update walks through what's actually different since the February launch: pricing, access, geography, ad formats, targeting controls, and the early performance data now coming in from real campaigns.

What actually launched

On February 9, 2026, OpenAI officially began testing ads in ChatGPT for free and Go tier users in the United States. The format was not a covert sponsored response embedded inside the AI's answer. Ads appeared at the bottom of ChatGPT's responses in a clearly labelled "Sponsored" section, with a visual border separating them from organic answers. OpenAI calls this "answer independence," and has positioned it as a non-negotiable cornerstone of its advertising approach.

The AI's answers are not for sale. The space below them is. That part hasn't changed. Almost everything else has.

The access story: from $200,000 gate to zero

The pilot launched as an invite-only, high-minimum product. Here's the actual sequence, now that it's played out in full:

• February 9, 2026 ,Pilot launches in the US, restricted to a small roster of large brands and agency partners (Dentsu, Omnicom, Publicis, WPP among them), with a reported minimum commitment around $200,000 and a flat $60 CPM.
• April 2026 ,The minimum drops to $50,000 as OpenAI onboards a second wave of advertisers through a semi-managed process.
• May 5–6, 2026 ,OpenAI removes the minimum entirely and opens a self-serve Ads Manager at ads.openai.com to any US business. No agency required.

That's a $200,000 floor collapsing to $0 in under three months ,a far more aggressive access curve than Google Ads or Facebook Ads followed in their early years.

Alongside the self-serve launch, the platform picked up features performance marketers actually need:

• CPC bidding, running alongside the original CPM model, with OpenAI recommending a starting range of roughly $3–$5 per click.
• A conversion pixel and Conversions API, letting advertisers tie post-click actions (signups, purchases, demos) back to specific campaigns ,the "data blackout" that used to define AI-chat advertising is now partly closed, though the content of a user's actual prompt still isn't shared with anyone.
• CPM pricing that's already come down from its launch level ,the flat $60 pilot rate has given way to a real-world clearing range of roughly $25–$60 depending on category, as inventory has grown.

Geographic expansion

The pilot was US-only at launch. On May 7, 2026, OpenAI announced plans to expand into the UK, Mexico, Brazil, Japan, and South Korea "in the coming weeks." Japan and South Korea went live as targetable markets in early July, alongside a broader ads-manager update. Canada, Australia, and New Zealand have also opened, largely through managed or agency arrangements rather than full self-serve access.

Reach is still small next to Google or Meta, and the audience remains weighted toward Free and Go tier users by design ,but the map has gone from one country to eight in five months.

New formats and interface changes

The static "favicon plus text" ad unit that launched in February is still the core format, but the surrounding tooling has expanded quickly:

• Product feed ads launched May 12, 2026, letting retailers surface individual SKUs rather than a single brand-level ad.
• Custom audiences entered gated rollout May 14 and became more broadly available in July: advertisers can now upload lists of 25,000+ users (email or phone match) to include or suppress from a campaign, and set bid multipliers by audience at the ad-group level. It's the first real first-party-data targeting lever the platform has offered ,a meaningful step toward the kind of targeting that Google and Meta advertisers already take for granted.
• A refreshed static ad card format rolled out across web and mobile in late June ,more compact, with larger visual elements, replacing the plain text-and-favicon unit from launch.
• Suggested ad drafts, added in early July, use a business's existing website metadata to pre-fill an ad ,image, headline, description ,for the advertiser to review and approve. It's explicitly not AI-generated copy; it's a prefill convenience, not a creative tool.
• AI-generated ad variations, spotted the same week under an "Add new ad" option, go a step further ,ChatGPT drafts new ad variations for an existing campaign, which the advertiser then reviews, edits, and approves before they go live. Early advertiser reaction has been mixed: some see real time-savings, others have publicly criticized the output quality.
• A new Overview tab in Ads Manager gives advertisers account-health monitoring, recommended optimization tasks, and a larger trend chart for key metrics ,the kind of dashboard-level view that was missing at launch.
• Video and carousel formats are expected later in 2026 but haven't shipped yet.Where the guardrails moved, and where they didn't

The original policy excluded ad placement near politics, health, and mental health topics, with a standing ban on dating, alcohol, drugs, and gambling as ad categories. An April 2026 update softened part of that: the blanket exclusion around medical, legal, and financial advice contexts was lifted, while the sensitive-topic guardrails for politics and mental health stayed in place.

That said, OpenAI's advertiser-facing policy documentation is inconsistent on a few categories ,legal services and healthcare appear in both the "limited" and "disallowed" lists depending on which help-center page you read, and financial services, dating, and health remain formally excluded as advertiser verticals even where the content restriction eased. In practice, legal-services ads have been observed serving as of July 2026 despite the disallowed listing, which tells you the policy is still being actively worked out rather than finalized. The under-18 exclusion (self-reported or predicted) has held steady since launch.

What the early tests are actually showing

This is the part that didn't exist in April ,real performance data from real campaigns, rather than projections.

• Revenue moved fast. ChatGPT's ad platform reportedly crossed $100 million in annualized revenue within six weeks of the February launch ,before self-serve access even existed. OpenAI has stated a $2.5 billion ad revenue target for 2026, with longer-range projections (reported, not confirmed by OpenAI) reaching as high as $100 billion by 2030.
• Click-through rate is the honest constraint. Early third-party data has put ChatGPT ad CTR at roughly 0.91% ,well below Google Search's benchmark CTR in the mid-single digits. That's the number to weigh against the platform's high-intent framing: the placement is well-positioned, but users aren't yet clicking at search-ad rates.
• Conversion quality looks different from click volume. Criteo has reported that ChatGPT-referred traffic converts at roughly 1.5x the rate of other referral channels in its own retail client data ,an interesting counter-signal to the low CTR, though it's a single ad-tech vendor's book of business, not an independently verified industry number.
• Delivery is still throttled relative to eligible inventory. Reporting earlier this year suggested fewer than 20% of eligible free-tier users were seeing ads on a given day, even though roughly 85% of Free and Go users are technically eligible ,meaning current revenue reflects a small fraction of the platform's eventual ad capacity, not its ceiling.
• Auction dynamics are now public. ChatGPT ads run on a relevance-weighted, second-price auction, similar in spirit to Google's model ,a strong, tightly targeted ad on a modest bid can out-perform a weak ad backed by a bigger budget.

The competitive picture, revisited

The core map from April mostly holds, with one notable update: Google has explicitly ruled out ads inside Gemini for now, with its advertising leadership stating publicly that Gemini currently carries no ads and has no plans to introduce them. Google's public energy instead has gone into agentic search, its Search I/O update made a newer Gemini model the default engine behind AI Mode, framing capability rather than an ad-free promise as its differentiator.

Anthropic, Microsoft, and Perplexity still haven't made public statements about advertising inside Claude, Copilot, or Perplexity's own assistant, respectively. That silence, five months into ChatGPT's ad rollout, is itself a data point ,either those companies see no near-term need to follow OpenAI's lead, or they're building quietly. Nothing here should be read as a forecast of what any of those companies will do.

The rest of the original comparison still tracks:

• Claude remains the platform of choice for nuanced, long-context, high-consideration research ,the exact moment conversational advertising is supposed to shine, without an ads product attached to it as of this writing.
• Perplexity still holds a meaningful head start on real conversational-ad performance data through its sponsored follow-up questions model.
• Gemini still leans on Google-ecosystem targeting continuity across Search, Docs, Gmail, and Google Ads ,now paired with an explicit no-ads stance inside the assistant itself.
• Microsoft Copilot remains the most instructive live preview of enterprise conversational advertising, still being stress-tested in production.

What smart brands should be doing now (updated playbook)

The original four-step sequence still applies, but step 4 changes completely now that the access barrier is gone.

Step 1 ,Build your organic AI presence first. Unchanged, and if anything more urgent: citation in credible sources, structured data, review volume, answer-friendly content, and comparison/FAQ formatting still drive whether ChatGPT mentions your brand for free.

Step 2 ,Run Perplexity campaigns now. Still the most accessible low-minimum entry point for building institutional knowledge about conversational ad creative and targeting.

Step 3 ,Stay active on Copilot. Still a useful, underused classroom for conversational ad mechanics.

Step 4 ,This is no longer a "get qualified" conversation. With the $200,000 (then $50,000) minimum gone as of May 5, 2026, any US business can open an account at ads.openai.com and test a real campaign for the price of a small Google Ads test. The strategic question has shifted from "can we afford to get in" to "does our audience and funnel stage justify testing now, given a 0.91% CTR and a platform still working out its guardrails." For considered, high-intent B2B or big-ticket consumer categories, testing now is cheap and the data is genuinely new. For low-margin, high-volume categories chasing clicks, the current CTR numbers argue for patience.

Writing creative that works in this format

This part of the original guidance holds up well against the new data. The user has just received genuinely useful, unbiased information. Your ad arrives in the moment immediately after ,a moment of goodwill, and with the new custom-audience and CPC tools, a moment you can now actually target and measure.

The structure that works: acknowledge the problem the user just asked about → position your brand as the natural next step → one specific proof point → one clear action.

What to avoid: hype language, vague positioning ("world-class," "industry-leading"), long copy, and generic creative recycled from a display campaign. With the new refreshed ad-card format giving more visual real estate, there's slightly more room for a supporting image ,but the copy discipline hasn't changed.

Where this is heading

The agentic-commerce question from the original piece ,who does an ad actually target when an AI is making the purchase decision on someone's behalf ,still doesn't have a clean answer. But the measurement infrastructure OpenAI has now shipped (conversion pixel, Conversions API, CPC bidding, custom audiences) suggests the company is building toward answering it with real attribution data rather than waiting for the theoretical problem to resolve itself.

Cost-per-engaged-conversation and conversation-value-optimization bidding, mentioned as coming in the original piece, haven't formally shipped yet as standalone bidding objectives ,CPC and CPM remain the two live options as of this writing, with cost-per-action reported to be in development without a confirmed date.

Conclusion

Five months after launch, the shape of the opportunity has changed more than the shape of the ad itself. The format is still transparent, the answer-independence principle is still intact, and the audience is still the free and Go tier. But the barrier to entry has gone from a six-figure commitment reserved for holding companies to an account anyone can open this afternoon. The pricing has softened, the targeting has gotten more precise, and the first real performance numbers ,a modest CTR, a more encouraging conversion signal from at least one ad-tech vendor, and $100 million in annualized revenue inside six weeks ,are now on the table instead of projected.

The strategic read hasn't changed as much as the mechanics have: build organic AI presence now regardless of budget, use Perplexity and Copilot to build institutional knowledge cheaply, and treat ChatGPT ads as a live, still-evolving channel worth testing at whatever budget makes sense for your funnel stage ,not a $200,000 bet you have to make all at once.

Not sure where your brand fits into this shift? Talk to Search Ally Agency about building an AI-search strategy, from organic visibility to testing your first ChatGPT ad campaign.