They told you to get on ChatGPT ads early.

July 17, 2026

The Speed Read

An analyst tracking OpenAI’s ad business says it’s on pace to hit roughly 10% of its own year-one forecast, while eMarketer pegs the ceiling for the entire chatbot ad market at $5.41B against OpenAI’s $100B by 2030 projection. The “be first” pitch is running on someone else’s math.
Open-web publisher ad supply fell 32-41% year-over-year in Q2 across U.S. and U.K. markets, per Ozone data covering 20 billion impressions. Rising CPMs are papering over the hole.
Buyers are quietly rolling back Meta’s AI creative tools after outputs failed basic brand quality checks. Meta keeps pushing adoption because Advantage+ needs training data.
New peer-reviewed research finds ChatGPT access is tied to a 9% drop in a user’s traditional Google search activity. Q4 search forecasts built in January are already wrong.

The ChatGPT Ads “Be First” Pitch Is Running on Bad Math

OpenAI told investors it would generate $100 billion in ad revenue by 2030.

An analyst at Adweek now says the company is on pace to hit roughly 10% of its own year-one forecast. Separately, eMarketer pegs the ceiling for the entire chatbot advertising market at $5.41 billion. Not 10% behind OpenAI’s number. Eighteen times smaller than it.

That matters for every operator sitting through a pitch this quarter. The “get in early” urgency, the FOMO decks, the rate cards: all of it is priced against OpenAI’s projection. None of it is priced against the market’s.

The structural problem isn’t that ChatGPT ads don’t work. It’s that conversational AI hasn’t built purchase-intent behavior yet. People use ChatGPT to think, research, and draft. Shopping intent hasn’t followed. A Google query signals a decision in progress. A ChatGPT conversation signals a question being formed. Those are different signals, and auction pricing hasn’t caught up to the gap.

Platform hype cycles almost never fail because the technology doesn’t work. They fail because the buying behavior isn’t there yet. ChatGPT is finding that out in public, at scale, with investor expectations attached.

The practical move: keep testing AI search channels at low budget while inventory is cheap and uncrowded. That window may still be real. Stop there. Don’t scale until someone can show you measured attributable revenue tied to a conversion your team controlled and tracked. Not impressions. Not CTR. Revenue.

First-mover advantage is a story. First-with-ROAS is the only advantage that survives a forecast reset.

If you’re navigating budget allocation heading into Q4 and want a second opinion on where AI channels actually fit, that’s what our media audits are for.

The Open Web Is Shrinking. Rising CPMs Are Hiding It.

The number that should be in every media plan right now: publisher ad supply fell 32 to 41 percent year-over-year in Q2, across the U.S. and U.K., per Ozone data covering 20 billion impressions.

Not traffic. Supply. The inventory pool your programmatic budget buys from contracted by up to 40% in six months, while AI zero-click search did what years of cookie deprecation threats never quite managed.

The reason most brands haven’t felt it: eCPMs are rising to fill the gap. You’re paying more per impression on a shrinking pool of sites, and the blended CPM figure looks like market stability. It isn’t. It’s a pressure valve.

Two things break as this continues. Frequency capping goes first. Your ads cycle through the same users more often because there are fewer sites to spread across. Incrementality erodes next. Bidding harder for the same shrinking inventory against the same buyers compresses your reach advantage and inflates cost per outcome.

Programmatic display was positioned for years as the diversified, open-web play. The open web is 40% smaller than it was a year ago. The play is different now.

Retail media, CTV, and creator inventory aren’t alternatives to eventually test. They’re where reach is going as the open web thins out. Brands building those relationships now are doing it before Q4 demand makes access expensive.

One specific thing to do this week: pull your programmatic frequency reports from Q1 and Q2 and compare average frequency per unique user. If it’s climbing without a corresponding drop in CPA, you’re not getting more exposure. You’re hitting the same people harder in a smaller room.

Quick Takes

Buyers are quietly rolling back Meta’s AI creative tools. AdExchanger reports agencies have started pulling Meta’s AI-generated creative from client accounts after outputs failed brand quality checks: broken text on-image, distorted product visuals, faces that go uncanny mid-video.

Meta keeps pushing adoption because Advantage+ needs training data to improve. Three moves if you’re running Meta this week: audit every AI-generated variant active in the last 30 days and kill anything a human wouldn’t have approved; turn off automatic Advantage+ creative enhancements on your top-spending campaigns and keep them on for test budgets only; put a manual creative review step back in your workflow before AI variants go live. The algorithm learns what you let through.

ChatGPT access is tied to a 9% drop in Google search. A before-and-after study on actual user behavior (not a survey) puts the first hard number on a curve most teams are still guessing at.

If ChatGPT penetration in your customer base grows 30% by Q4, you’re looking at a mid-single-digit demand hit on Google before your holiday campaigns even launch. Build a downside case into any search-heavy Q4 forecast. Put non-Google discovery channels on committed budget lines in Q1 planning, not “test and learn” allocations.

Google and Meta wrote the calculators grading their own ad performance. Google Meridian and Meta Robyn have become the de facto open-source standard for marketing mix modeling, right as MMM replaced last-click attribution for most performance teams.

The problem: MMM isn’t a truth machine. It’s a set of assumptions about baselines and saturation curves, and whoever writes the default assumptions shapes the answer. Ask your agency two questions this week: which MMM are you running, and who validates the model’s priors? When Meridian says Google over-performed, can you independently verify that? If the answer is “we trust the model,” that’s not measurement. That’s convenience.

Hundreds of creators. Zero new headcount.

Every Shopify brand wants more creators posting about their product. Almost none have the team to make it happen at scale.

partnerUP is that team.

You guide the brief. From there the AI does what a roomful of coordinators would: reaching hundreds of creators, ranking them by fit, handling contracts and negotiation, paying on delivery, and tracking what each one drives back to your store.

You approve the talent and watch the content roll in. No agency retainer. No new headcount.

Your first two creators are on us. See what they bring back.

Start with two free

The Last Word

This week landed hard data hits on two of performance marketing’s biggest operating assumptions: that AI ad platforms will grow into their projections, and that programmatic display is your diversified reach play.

Neither one is finished. Both need repricing in your Q4 plan.

The teams that navigate the next 18 months well won’t have the most aggressive AI roadmaps. They’ll be the ones whose channel decisions are built on what the numbers say, not what the pitches projected.