Wrote a prompt
Wrote a prompt this week to generate ad copy for the new ChatGPT placement, and building it meant throwing out almost everything Google Ads taught me. The slot looks like a search ad. It behaves like nothing of the sort.
🧠 Prompt of the Week
You are a direct-response copywriter who specializes in ChatGPT ad placements — the labeled ad box that appears at the bottom of a ChatGPT answer. The reader is mid-conversation and high-intent, not scrolling a feed.
PRODUCT: [2–3 sentences: what it does, who it's for, the one outcome it delivers]
CONVERSATION CONTEXT: [the question/topic the user is asking ChatGPT when your ad would appear — e.g. "how do I cut SaaS churn in the first 30 days"]
Write 5 ad variations. Each must:
- Read as a natural continuation of a helpful answer, not an interruption
- Open with a specific, true observation about the reader's situation (no "#1", "best", or "leading")
- Name the concrete next step or outcome, not a generic CTA
- Stay under 25 words
- Avoid: "free trial," "start today," "revolutionize," exclamation marks, emojis
Output each as:
AD: [copy]
WHY IT FITS THE SLOT: [one line — the conversational intent it matches]
Then rank all 5 from most to least likely to earn a click from a high-intent researcher, and justify your #1 in one sentence.The engineering decision that makes this work: the ranking step forces the model to apply judgment instead of dumping five interchangeable lines, and the "WHY IT FITS THE SLOT" field ties every line back to a real intent state - which is the entire game in this placement. Banning the feed-words ("free trial," "start today") stops it defaulting to Google-brain on autopilot. Paste it and you'll get five conversation-native variations plus a reasoned pick to test first.
📌 The Breakdown
Everyone's reading the May 5 news as "ChatGPT ads opened up — go run them." That's the surface. The second-order story is the part that actually matters for SaaS.
The same update that killed the minimum-spend barrier also shipped a measurement pixel and a Conversions API. Until then, the loudest complaint about the platform was that you couldn't see what happened after the click — which meant high-consideration B2B SaaS, with its long sales cycles, had no business reason to spend. That objection is now mostly gone. (Deeper CRM closed-loop attribution is still rolling out across 2026, but you can finally track conversions, not just impressions.)
Stack that on top of how the targeting works. It's contextual, not keyword-based — your ad surfaces while someone is mid-conversation about the exact problem you solve. So you've got a high-intent reader, in an active problem-solving state, and now the ability to measure the outcome. For bottom-funnel offers, that's a genuinely new acquisition surface, not a novelty.
Here's where most teams will faceplant: they'll launch a brand-awareness campaign with feed-style copy, get nothing, and conclude "ChatGPT ads don't work for SaaS." Wrong copy, wrong funnel stage, wrong conclusion.
Your 30-minute move this week: don't launch anything yet. Take your single highest-intent use case — the problem people are literally typing into ChatGPT right before they'd need you — and write five conversation-native ads for it with the prompt above. Bank them. When your vertical opens up, you launch with copy built for the room while everyone else is still pasting in Google headlines.
⚡ Quick Hits
CPC bidding landed alongside CPM. For SaaS, CPC is the only model that makes sense early — CPM rewards reach, and reach was never your problem. Qualified clicks are.
Ads only show to Free and ChatGPT Go ($8) users — Plus, Team, and Enterprise are ad-free. Underhyped implication: your enterprise buyers literally can't see these. This is a PLG / self-serve / SMB channel, not an ABM one. Match your offer to who's actually on the free tier.
Targeting is contextual — conversation topic plus chat history, not keywords. Stop thinking in search terms. Start thinking: "what problem is someone mid-solving when I want to appear?" Write to the moment, not the keyword.
AI ad spend is projected to climb from ~$35B to ~$142B by 2030. The copy conventions for this medium get set in the next 18 months. Whoever cracks conversation-native copy now writes the playbook everyone else copies later.
BYE! SEE-YA!
And,
Keep prompting, Sarvesh
If this issue made you rethink even one line of your ad copy, forward it to the founder on your team who's about to throw budget at ChatGPT ads.
