ChatGPT advertising opened to all US businesses in early June, in the same window Google Marketing Live pushed its own AI ad formats. The question landed in every client call the following week: do we move budget.
Our answer is usually no, and then sometimes yes. The reasoning matters more than the answer.
New inventory is always cheap, and that is the trap
Every new ad surface goes through the same cycle. Early access, low competition, unusually good cost per click, a wave of case studies from people who got in first. Then demand catches up, the auction normalises, and the numbers look like everything else.
The early window is real. Some of the best returns we have generated for clients came from being early on a channel. But being early only pays if you can tell whether it worked, and that is where new inventory usually fails.
Worth remembering. A cheap click on a channel you cannot attribute is not a cheap lead. It is an unmeasured cost that looks like a win in the platform dashboard.
The three conditions
Before we move any budget onto a new surface, we check three things. If any one fails, we wait.
One: can you measure it end to end. Not clicks. Not platform-reported conversions. Can you follow a person from that ad to a pipeline entry in your CRM. If the answer involves the phrase “we will look at assisted conversions”, you are not ready. This is the condition that fails most often, and it is why we start with tracking rather than campaigns.
Two: does the intent match your funnel stage. Someone asking a conversational assistant a question is in a different mental state than someone typing a commercial keyword into a search box. That is not worse, it is different. Discovery intent needs an offer built for discovery, not the demo request form you use for high-intent search traffic.
Three: can you fund it without taking from what works. If the test money has to come out of a campaign currently producing leads, the test is not a test. It is a bet on a channel you have no data for, paid with the one you do.
PAID MEDIA
Test new channels without breaking what works
We run structured channel tests with a fixed ceiling, a single conversion action and a clear kill date, so a failed experiment costs you a number and not a quarter.
How to structure the test when you do go
Assume the three conditions pass. Here is the shape we use, because an unstructured test on new inventory produces a number nobody can interpret.
One conversion action, chosen in advance. Not three. Pick the single event that means something commercially, usually a qualified form submission or a booked call, and judge the channel on that alone. Teams that track four events end up arguing about which one counts.
A ceiling and a kill date, written down before launch. An amount you can lose without flinching, and a date on which you stop regardless of what the numbers look like. Without the date, a mediocre test survives on hope for two quarters.
Landing pages built for the intent, not reused from search. This is the part most teams skip and it is the one that decides the outcome. Someone arriving from a conversational assistant has usually asked a question, not typed a commercial keyword. Sending them to a page that opens with a demo form wastes the click.
A holdout you can compare against. Run the test while your existing channels continue unchanged, so that any movement in total pipeline can be attributed rather than assumed.
At the end, one question decides it: did qualified pipeline increase by more than the spend. Not clicks, not cost per click, not what the platform reports as conversions.
What your analytics already tells you, for free
Before spending anything, look at what assistant traffic is already doing on your site. Most accounts have some, and almost nobody looks at it.
Three things worth checking this week. How much referral traffic arrives from assistant domains, and whether it is growing month over month. What those sessions do once they land, compared with organic search sessions. And whether any of them convert at all today.
If that traffic already converts without you spending a euro on it, the paid version deserves a serious test. If it arrives and bounces, buying more of it will not fix the landing experience that caused the bounce.
This costs nothing and takes an hour. It is also the only evidence you will have that is specific to your business rather than to someone else’s case study.
What we are actually doing
For most accounts, we are not moving budget yet. We are doing three cheaper things.
We are watching whether assistant traffic already reaches the site, and what it does when it arrives. That data exists today in analytics and costs nothing to look at.
We are checking that landing pages answer a question rather than pitch a product, because conversational referral traffic behaves like early-stage research traffic.
And we are keeping a small test line ready, so that when we decide to move, it comes from a budget that was always meant to be spent on learning.
Being early to a channel pays only if you can prove it worked. Otherwise you just bought cheap clicks and a good story. MyDigipal
For accounts where the audience genuinely lives in conversational tools, and where measurement is already solid, the calculation changes and we move sooner. That is a small minority today. It will not stay small.
Who should move first
Three profiles where we would push rather than wait.
Products people research by asking questions rather than by comparing specifications. Anything where the buyer’s first move is “how do I solve this” rather than “who sells this”. Conversational placement matches that phrasing far better than a keyword auction does.
Categories where search competition has become punishing. If your cost per click has doubled in two years and the auction is crowded with aggregators, a newer surface with fewer bidders is worth a serious look, provided you can measure it.
Teams with an existing content library answering real questions. That library is the asset that makes conversational referral traffic convert, because the landing experience already matches the intent. Without it you are buying clicks into a brochure.
If none of those describe you, waiting a quarter costs almost nothing. The inventory will still be there, it will be better documented, and someone else will have paid to learn what works.
The honest position is that this inventory will matter, probably sooner than most advertisers expect, and that almost nobody is currently equipped to measure it. Fix the measurement first. The budget decision gets easy after that.
If you want to know what a structured channel test would cost on your account, the calculator gives you a range in two minutes.
Sources: Digital marketing news roundup, 1-15 June 2026 - Top marketing AI stories, week of 1 June 2026 - AI and marketing news roundup, June 2026
