In Part 1, we covered how ChatGPT Ads is a recommendation layer, not a search channel. And why the startups that nail down conversational ads quickly will have a long-run advantage.

This piece covers the four things you need to be doing with your early ChatGPT campaigns: wiring attribution before the first dollar goes out, writing Context Hints around problem states instead of keywords, making creative more like a recommendation instead of a pitch, and committing to a 30-day validation test.

The underlying mechanics

Before the tactics, let's look at why your old ad playbooks won't transfer. ChatGPT Ads differ from Google and Meta in a few key ways.

First, targeting is conversation-based, not keyword- or audience-based. Google matches the keywords someone types. Meta matches audiences and behaviors. ChatGPT matches the actual conversation someone is having in the moment. So instead of picking keywords or audiences, you describe the kind of conversation (and the person having it) that you want your ad to show up in.

Second, the context the user is in is unlike anywhere else you advertise. On Instagram or TikTok, you're interrupting someone who came to be entertained. On Google, you're catching someone mid-search. On ChatGPT, they're deep in a focused, high-trust conversation, working through a real decision and often spelling out their exact situation as they go. That headspace changes what they're receptive to and how your ad has to fit in.

Third, it's a recommendation layer, not an interruption or traditional search engine. Your ad appears right after ChatGPT has given the user an answer they trust, framed as a natural next step rather than something that barges in out of nowhere. Copy built to interrupt likely works against you here.

Last but not least, you're optimizing toward clicks and impressions, not conversions. On Meta, you hand the algorithm a conversion event and it goes and finds buyers for you. ChatGPT Ads can't do that. At least not yet. The system isn't optimizing toward the right buyer for you. That's your job.

Which is exactly what the rest of this piece is about. How to approach the channel and set yourself up for success from day one.

First things first: the pixel

The measurement environment on ChatGPT Ads is quite limited compared to Google and Meta. So launching without rock-solid tracking means your first month generates almost no usable data.

OpenAI's OAIQ pixel is a JavaScript snippet placed site-wide. Anyone who has set up Meta's pixel will find the event structure familiar. Install it, QA it, and confirm conversion events are firing before you add payment info to the Ads Manager.

One important caveat: OpenAI doesn't support pixel deployment through Google Tag Manager yet, but you can still deploy through GTM with a community template (like Stape), or just drop the snippet directly into your site code.

Alongside the pixel, wire up the Conversions API. The CAPI is the server-side complement. It sends conversion events from your backend to OpenAI without relying on the browser. The pixel handles client-side tracking; the CAPI handles server-side. If you've set up Meta's CAPI post-iOS 14, this is the same idea. Both launched on May 5 alongside the self-serve Ads Manager. Running both from day one gives you the most complete measurement picture available on the platform right now.

The Conversions objective is rolling out in early June 2026. Accounts that had conversion events configured by June 1 get early access starting June 5. If you're reading this and haven't set up the pixel yet, you'll be a few weeks behind on conversion-optimized campaigns. Until your account is granted access, you're working with Reach (CPM) and Clicks (CPC) objectives only.

Reporting has a multi-hour lag, so same-day data isn't reliable. Attribution windows aren't configurable per-campaign yet; OpenAI sets the tracking window on their side. Daily optimization is not possible. Commit upfront to not panicking if the dashboard looks flat on day 5.

Outside of the platform, you should be tracking branded search volume trends in GA4 in the two weeks following launch. Measure assisted conversions. Track the lift in direct traffic. A significant portion of conversions will happen outside the immediate click window. If you're only measuring same-session conversions, you're dramatically undervaluing the channel. In your own analytics, set conversion windows to track 7-14 days past OpenAI's 30-day cutoff. ChatGPT users often see your ad, sit on the decision, and return through branded search or direct traffic in week 5 or 6. The platform stops crediting those touches at day 30; your internal attribution shouldn't.

Context Hints are not keywords

We're seeing lots of startups get this wrong. (Those Google Ads habits die hard.)

A ChatGPT Ads "Context Hint" is a plain-language description of up to 280 characters, written at the ad group level, that tells OpenAI's semantic matching system which conversations your ad belongs to. The system reads intent, not words. It does not match keyword fragments. Writing Context Hints like a keyword list is the most common targeting mistake on the platform.

The difference in practice:

Weak hint: "CRM sales software best tools comparison"

Strong hint: "A founder evaluating CRM options for a small remote sales team struggling with pipeline visibility and inconsistent follow-up"

The second works because it describes a situation, a person, and a moment of uncertainty. ChatGPT's targeting system aligns the hint with inbound conversations. When you feed it keywords, it tries to match conversations containing those words, which rarely surfaces the high-intent conversations you actually want.

A useful formula for writing hints: Persona + Intent + Scope. Who is the person? What decision are they making? What context narrows the conversation toward your category? Test each hint by reading it aloud. If it doesn't sound like something you'd say to a colleague briefing a copywriter on a buyer scenario, rewrite it.

On bids: the platform recommends $3-5 CPC as a starting point. OpenAI's recommended floor ($3) is technically valid, but from what we're seeing at Saturation, it routinely fails to clear delivery thresholds. Many founders start at $3, see nothing for a week, and conclude the channel doesn't work. Start at $5 or above until you have delivery data, then adjust.

Another thing we're seeing ChatGPT Ad newbies miss is matching your Context Hints to the conversation stage. A user asking "what is a CRM?" is in a fundamentally different mental state than one asking "which CRM integrates with Slack and costs under $30 per seat?" Your hint needs to match the depth and specificity of the conversation it's targeting. Build separate ad groups for each stage:

Early-stage (educational): The user is orienting themselves in a category. Hints should describe the broad problem. "A small business owner trying to understand whether they need a dedicated CRM or can keep using spreadsheets." Creative should lead with educational value and low-friction CTAs.

Mid-stage (comparison): The user is actively evaluating options. Hints should describe the specific decision context. "A SaaS founder comparing CRM options for a team of 10-15 people with both inbound and outbound sales motions." Creative should include specific differentiators and social proof.

Late-stage (decision): The user is close to choosing. Hints should describe the moment of commitment. "A VP of Sales evaluating whether to migrate from Salesforce to a lighter CRM without losing pipeline data." Creative should reduce risk and uncertainty: free trials, migration support, customer references.

OpenAI currently provides no audience size estimates, reach forecasting, or targeting diagnostics. You cannot verify how broad or narrow your matching actually is. Don't build one perfect hint and assume it's working. Run three to four variations simultaneously and compare impression delivery patterns over 30 days. That's what's working for our clients at Saturation.

Creative that fits a conversation

OpenAI claims that sponsored content never influences what ChatGPT actually says. The ad appears below the AI's response in a clearly labeled box. The user has just read an answer they trust. Your ad is the first thing they see after that.

Traditional PPC copy is built to interrupt. In ChatGPT, interruption loses. The creative that wins reads like a natural next step inside an ongoing conversation, not a pitch that arrived uninvited.

The specs are tight: short headline, description capped at 48 characters, square image minimum 256x256px. There is no room for a full pitch. The goal is to show the user you understand their situation.

See the difference:

Google-style: "Best CRM for Startups. Try Free."

ChatGPT-style: "Still updating pipeline manually?"

The second works because it names the problem the user is already thinking about. Forty-eight characters is not enough space to explain what you do. It is enough space to show the user you understand their situation.

Strong ChatGPT creative does four things: names the problem directly, matches conversational register, reduces uncertainty rather than amplifying urgency, and positions the product as a useful recommendation rather than a demand for attention.

Landing pages matter here more than most advertisers expect. But don't overthink it: the user arriving from a ChatGPT ad has already been primed by the conversation they were having. They're further along than a cold Google click but less committed than a branded search visitor.

Your page needs to answer three questions fast: what is this, who is it for, and why does it fit my situation. Specificity beats polish. A page that says "built for 10-50 person sales teams running outbound" will outperform a generic "the #1 CRM platform" page every time, because the user just came from a conversation where they described that exact situation.

How campaign structure differs by vertical

The same product in different conversation contexts needs different Context Hints and different creative framing.

B2B SaaS buyers use ChatGPT heavily for vendor research and operational problem-solving. Focus your Context Hints on their problem, not the overall software category. "Trying to manage international contractors across five countries?" is better than "Global HR platform for modern teams" because it matches the actual conversation the buyer is having. The more your hint describes a real problem in the language of someone living it, the more naturally your ad fits inside a research conversation.

If you're selling a B2B product, you want to be targeting moments of comparison. If your product has a "vs" page or a comparison table, your buyers are having ChatGPT conversations about you right now. If the purchase is impulse-driven or visually motivated, other channels will outperform. One important constraint: ads only reach Free and Go tier users, and ChatGPT skews young. 58% of US adults under 30 have used it, vs 25% of those 50-64 and just 10% of those 65+ (Pew, 2025). For consumer products targeting digital-native buyers, the fit is strong. For products with older buyer demographics, you may find the ad-eligible audience doesn't map to your ICP today.

Ecommerce and DTC brands are a fit for ChatGPT Ads when the product requires research: supplements, skincare with ingredient considerations, high-end electronics, specialty gear. ChatGPT conversations about "best running shoes for flat feet under $150" or "cleanest protein powder that doesn't taste terrible" are exactly the moments where a well-placed ad earns the click. Weak fit: fashion, accessories, and visually-driven categories where the purchase happens through images, not conversation.

Professional services like agencies, consultants, legal, and accounting are also a natural fit for the new ChatGPT Ads platform. People use LLMs constantly for vendor research: "I need a fractional CFO for a Series A SaaS company. What should I look for?" Your Context Hints should describe the client's situation, not your company's credentials. "A founder deciding whether to hire a full-time CFO or bring on a fractional CFO for their first institutional fundraise" is way better than "fractional CFO services for startups."

ChatGPT Ads also work great for products like ed-tech and professional development. People are looking for recommendations for learning resources, certifications, and skill development inside conversational environments. Decisions about courses and certifications are naturally deliberate. Context Hints around career transitions, skill gaps, and professional development goals tend to generate strong impression delivery in this category.

These are just a few industry examples, but we're seeing ChatGPT Ads work across the startup world. Just make sure your targeting is tight.

Custom audiences are arriving

On May 14, OpenAI began rolling out custom audience uploads. We're still not sure how or when they'll be fully rolled out, but you'll be able to bring first-party customer lists into the Ads Manager to suppress existing customers from seeing your ads or to build targeting around known prospects. This is a significant upgrade. Until then, there was no way to layer your own data on top of ChatGPT's contextual matching.

Two immediate applications. First, suppression: if you're running a CPC campaign, every click from an existing customer is wasted spend. Upload your customer list and exclude them. Second, prospect targeting: if you have a list of leads who've engaged but haven't converted, you can target them specifically inside ChatGPT conversations where they're researching your category. The combination of your first-party data and ChatGPT's contextual matching is something no other AI ad platform offers right now.

CRM integrations (Salesforce, HubSpot) are on OpenAI's stated roadmap for later in 2026, which would enable closed-loop attribution from ChatGPT impression to closed deal. That's not available yet, but custom audience uploads are the first step in that direction and they're going live imminently.

What a minimum viable first test looks like

At Saturation we're seeing that most founders underfund their first test on ChatGPT Ads and conclude that it's not right for them. But that's a misread.

Before launch, four things need to be in place:

• OAIQ pixel installed and conversion events confirmed firing
• At least three to four Context Hint variations running simultaneously
• At least two creative variants per ad group
• Bids starting at $5 CPC or above until delivery data exists

On budget: a practical floor for testing is roughly $50-$100 per day, which generates enough delivery volume to learn from inside 30 days. At $5 CPC, that's 10-20 clicks per day, or 300-600 clicks over a test cycle. Below that volume, you won't have enough data to distinguish signal from noise across your Context Hint variations. A minimum viable first test costs roughly $1,500-$3,000 over 30 days. If that's outside your budget right now, wait until it isn't rather than running a $500 test and concluding the channel doesn't work.

During the 30-day window, watch three signals. Impression delivery patterns across Context Hint variants: this tells you which problem states your category actually appears in, and it's the most valuable thing the first test can teach you. CTR as a signal of creative fit: whether the creative earns a click from someone mid-conversation is a meaningful read on whether it reads as relevant or intrusive. And branded search lift in GA4 in the two weeks after launch: this is where a meaningful portion of ChatGPT's conversion influence shows up, even when referrer data doesn't pass through.

What success looks like after 30 days: you know which problem states trigger your category. You have a baseline CTR for creative fit. You have a read on branded search movement. You have enough to make two or three concrete decisions for the next cycle. You do not have ROAS. That is not what the first test is for.

The goal right now is getting a fast early start

ChatGPT Ads is new enough that testing properly, with pixel and CAPI wired up, Context Hints written as buyer scenarios, creative that fits a conversation, and a structured 30-day test with a real budget, is itself a competitive advantage. Most advertisers we're seeing aren't doing all of these things. The ones who do will have more data, better creative instincts, and established presence by the time the auction fills up.

In Part 3, we'll cover how ChatGPT Ads works alongside organic AI visibility, and why the companies running both as a single workstream are seeing compounding advantages the paid-only advertisers aren't.

If you want help setting up your first ChatGPT Ads test the way we'd run it, book a free consult at saturation.co →

Thanks y'all!