The Call

In 2023 an OpenAI engineer wrote a line that is now sitting in a court filing: “no matter how prominently we show the links, users won’t click.”

The filing comes from publishers suing OpenAI and Microsoft, and it cites Microsoft’s own data comparing Bing Chat to traditional search. Times sites saw click-through rates 87 to 93% lower in chat. Ziff Davis properties landed between 51 and 94% lower.

OpenAI and Microsoft dispute the framing and argue fair use. No court has ruled. But the numbers came from inside the companies building the answer layer, and they are describing the behavior of their own users.

Here’s what that does to your business. As more of the research that precedes a purchase happens inside an AI answer, the clicks you can trace become a smaller slice of the demand you paid to create. Your reports get worse. Your demand might not.

Q4 started today. In January someone will hand you a report on it.

If that report is built on last-click, there is a good chance it shows a decline the business didn’t actually have, and you will spend the first quarter of next year cutting budget or questioning a team based on a measurement artifact.

The reason this is a decision for this week rather than next quarter is that every alternative measure needs a before. Branded search trend, direct traffic baseline, a geo holdout. None of them can be reconstructed in January. You either start the clock before spend scales, or you spend Q1 arguing about numbers nobody can settle.

So the call is what you’ll judge Q4 on, decided now, while the baseline is still available to take.

For most brands at your size that’s about two hours of work. Pull the four-week average on branded search and direct traffic before your Q4 budgets step up, write the numbers down somewhere that isn’t a dashboard, and pick one channel to hold out.

Two hours now, against a quarter of arguing later.

If you want an outside read on whether your measurement will survive this quarter, that’s what our audits cover.

Ask Your Team

“What share of last quarter’s revenue started with a click we can trace?”

Green flag: a number, plus an honest account of what isn’t traceable.

Red flag: “all of it.” Nobody’s attribution is that clean. That answer means someone is reading platform-reported conversions as fact.

“If last-click gets worse in Q4 but revenue holds, how would we tell which one is wrong?”

Green flag: they name a method. Geo holdout, branded search baseline, matched market test.

Red flag: they’d go look at more platform dashboards. The dashboards are the thing in question.

What Good Looks Like

Three baselines, all of which have to be taken before the quarter runs.

None of these can be built retroactively. That’s the entire reason this is a first-week-of-October job rather than a January one.

If you only do one, do the branded search baseline. It’s fifteen minutes and it’s the single best check on whether your paid spend is creating demand that your attribution never sees.

Ignore This Week

Loud right now. Safe to skip at your size.

McDonald’s launched a media network with a $1B revenue target. Retail media is becoming something every brand with foot traffic sells. Not a Q4 decision for you.

Not ignorable: Apple’s iOS 27 blocks The Trade Desk from serving ads in Safari. If any part of your open-web programmatic runs through TTD, your Safari reach is about to drop without anyone sending you a note about it. Ask your agency two things this week: what share of our programmatic impressions were Safari, and what replaces them.

What percentage of your revenue still starts with a trackable click? Hit reply, I’m curious how wide the range is.