The formats are moving
Ad products inside AI assistants are still being designed, renamed and occasionally withdrawn. What we buy this quarter may not be what exists next year, and any agency claiming a stable playbook here is guessing.
When your customer asks an assistant for a recommendation, that answer is becoming an ad surface. This is about being there before the auction forms.
AI advertising is paid placement inside AI assistants — sponsored positioning in the answers people get when they ask ChatGPT and its equivalents for a recommendation. It is bought inventory, it works while you pay for it, and it stops when you stop.
It is not the same thing as GEO, which earns your way into those answers organically. Running one without the other is the common and expensive mistake. We wrote about the difference here.
There is a well-worn pattern to advertising inventory. A new surface appears. Almost nobody advertises on it because it is unproven and the audience seems small. The businesses who go early pay very little for a great position. Then the surface becomes obvious, everyone arrives at once, and the price triples in eighteen months.
Paid search did exactly this. So did Facebook. So did every channel that is now expensive. The businesses that are cheap to find on Google today are, almost without exception, the ones who started when it was cheap.
AI assistants are at the start of that curve. The audience is no longer small — a very large number of people now ask an assistant before they ask a search engine — but the advertiser competition has not caught up. That gap is the opportunity, and it is temporary by definition.
This is the newest thing on this website and we would rather be straight about its edges than sell it as settled.
Ad products inside AI assistants are still being designed, renamed and occasionally withdrawn. What we buy this quarter may not be what exists next year, and any agency claiming a stable playbook here is guessing.
Attribution inside a conversation is genuinely harder than attribution on a click. We measure what can honestly be measured and flag what cannot, rather than producing a confident-looking number built on assumptions.
Nobody should move their whole budget here. It is an early position in a channel we expect to matter enormously — which is a different and more modest claim than "this replaces your search spend".
This distinction is the single most useful thing on this page. Most of the confusion in the market — and a good deal of the bad advice — comes from treating these as one product.
| AI advertising (paid) | GEO (organic) | |
|---|---|---|
| What you are buying | Placement inside the answer | Nothing — you are earning the mention |
| How fast it works | Immediately, once approved | Weeks to months |
| What happens when you stop | It stops that day | It keeps working, and decays slowly |
| What it costs over time | Rises as competitors arrive | Falls per mention as the asset compounds |
| What it is good for | Speed, testing, and claiming an early position | Durability, credibility, and the answers nobody can buy |
| Should you run both? | Yes — they cover each other's weakness | Yes — and this one first if you must choose |
Advertising inside AI assistants is emerging rather than mature, and what is available varies by platform and changes quickly. We will tell you exactly what is buyable at the point you engage rather than describe a menu that may not exist by the time you read this. What is reliably available today is the organic side — GEO — which is why we sequence it first.
A search ad competes for a click on a page of ten results. An AI placement competes to be part of a single synthesised answer where there may be no list at all. Fewer positions, much higher stakes per position, and a completely different creative problem — you are influencing a recommendation rather than writing a headline.
Do GEO. It is slower but it is durable, it costs nothing per impression once earned, and it is available now. Paid AI placement is the accelerant, not the asset.
No, and we are deliberate about this. That range is the target we design an engagement toward — a model built from what each channel typically contributes, not an average of measured client outcomes. We do not have published before-and-after CAC data yet, and until we do we will not present it as a result. When a client's own reporting produces real numbers, we will say so and show them.
The businesses that are cheap to find on Google today all started when it was cheap. This is that moment again.