ChatGPT ads didn't fail. You measured the wrong number.

Welcome back to the front row. This week, the number everyone is quoting about ChatGPT ads is the wrong one.

Every marketing feed told you the same thing this month: ChatGPT ads flopped. A click-through rate seven times below Google. Budgets sitting unspent. A reporting tool that didn't even work. All true. And all beside the point — because while everyone argued about the click, the thing that actually pays was moving the other way.

The number that buried a channel

Three weeks into OpenAI's ad pilot, the verdict was already written.

The headline number: a 0.91% click-through rate, against a 6.4% benchmark for the same category on Google Search. Roughly seven times fewer clicks.

For most marketers, that settled it. ChatGPT ads don't work. Move on.

But here's where it gets uncomfortable.

Write the channel off on that one number and you do two things at once. You hand an unpriced, high-intent surface to whoever is paying attention. And you guarantee you re-learn it later — at auction prices, against people who started in 2026.

Why the click is the wrong instrument

A search user is hunting for a destination. They click to leave. A ChatGPT user is mid-task — they stay in the conversation. Same word, "click," completely different behaviour. Measuring one against the other's benchmark was always going to look like failure.

Now read the numbers the headlines skipped.

Criteo, the first ad-tech partner wired into the platform, reports that users arriving from LLMs convert at roughly 1.5 times the rate of other referral channels. Not click. Convert.

Similarweb, tracking referral traffic across the web, clocked ChatGPT referrals jumping more than 150% week over week — not flat, not fading, accelerating.

And the excuse that made all of this un-provable just expired. Through the early pilot, OpenAI's Ad Manager couldn't even show advertisers their own results — weekly CSV exports, no conversions. As of May, conversion tracking is live in the beta, with cost-per-action bidding arriving in early June.

So the channel that "can't be measured" can now be measured. Right as the people who waited for proof start paying attention.

The pattern under the pilot

This isn't really about ChatGPT.

Every new AI surface arrives the same way: judged first on the metrics of the thing it replaces, mispriced second, and quietly compounded by whoever ignores the verdict.

Search looked like a toy next to print. Social looked like a distraction next to search. The scoreboard always lags the surface.

Where this leaves your account

Look at how you are judging AI placements right now.

If your dashboard ranks them on CTR next to your branded search campaigns, you have already decided they lose. The comparison is rigged — by you.

You are not measuring a channel. You are measuring a channel against a yardstick built for a different one.

What to do this week

Three moves. None of them need a budget.

Read every AI surface you already touch on assisted and post-click conversions, not CTR. Judge it on what happens after the click.

Switch on conversion tracking in ChatGPT's Ads Manager beta now that it exists — even at a token spend — so you have a baseline before cost-per-action bidding goes live and the cheap-learning window closes.

Map your category honestly. High-consideration, research-heavy buying — B2B software, finance, travel, education — fits a conversational placement. Impulse retail mostly does not. Know which one you sell before you spend a dirham.

The pilot didn't prove ChatGPT ads work. It proved the industry still grades new channels on old report cards.

The advertisers who win the next surface won't be the ones who clicked fastest. They'll be the ones who stopped trusting the click.

Prompt of the week

Before you spend on any AI ad surface, find out whether your category even belongs there.

You are a paid media strategist. My business is [what you sell], and my buyers typically [how they research before buying]. On a scale of 1–10, score how well a conversational AI ad placement (like ChatGPT ads) fits my category, based on buyer intent and research depth — and explain the score in two lines. Then give me the five highest-intent questions my buyers ask an AI assistant before they're ready to purchase, and a one-line ad angle for each.

What this gives you: a fast read on whether the channel fits you — plus five intent-matched angles you can test the day it opens in your market.

This week

Google starts deleting your old reporting data this month. Anything older than 37 months gets wiped unless you export it first — pull your historical benchmarks now or lose them.

Google warns against buying your way into AI answers. Alongside the May core update, Google said it will treat paid or manipulated AI mentions as a violation — visibility in AI answers is earned, not bought.

Microsoft bundles AI bidding and attribution across platforms. New multi-platform tooling aimed squarely at the cross-channel buyers Google and Meta are fighting over — worth a look if you run more than one platform.

Tool of the week

Pine AI — 19pine.ai

What it does: an autonomous agent that picks up the phone for you. It calls companies, negotiates bills, cancels subscriptions, and chases refunds, then reports back what it saved. Pine claims a 93% success rate on tough negotiations and around 20% off the average telecom bill.

How to use it right now: hand it one annoying recurring task — a too-high internet bill, a subscription you keep forgetting to cancel — and let it run the call. You only pay when it actually saves you money. (It's US-first today, so coverage outside North America is still catching up.)

Why it matters this week: this is the same shift as the main story, pointed at your customer. When an AI agent is the one calling your support line, working your retention offer, or cancelling on a buyer's behalf, your churn flow and your customer service are no longer built for humans alone. The agent reads your retention script very differently than a frustrated person does. Start asking who is really on the other end of the call.

My take

I've watched our industry do this three times now.

A new surface shows up. It gets measured against the old one. It "fails." Everyone moves on, relieved they don't have to learn something new.

Then it quietly eats a third of the budget, and we all pretend we saw it coming.

What gets me about the ChatGPT ads pile-on isn't that people are wrong about the click. They're right. 0.91% is bad — if a click is what you're buying. But nobody running these campaigns at scale is buying clicks. They're buying position inside the answer, at the exact second a buyer is deciding. That's not a banner. That's the most expensive real estate in marketing, priced like a clearance bin because the report card hasn't caught up.

I'm not telling you to pour budget into ChatGPT. The pilot is still rough, and most of you can't even run it yet.

I'm telling you to stop outsourcing your judgment to a benchmark.

The click stopped being the point a while ago. Most of us just haven't updated the scoreboard.

That's Issue #7 of NVM. If this gave you one thing to act on this week, share it with someone who needs it. Reply to this if you have a question — I read every one. See you next week. — Mostafa

NVM · Front row seats to AI and digital marketing

• Subscribe → nvm.marketing/newsletter
• Join the community → nvm.marketing/community
• Write for NVM → nvm.marketing/blog