Which ad platform fits what you sell
The most expensive channel decisions I see are rarely bad campaigns. They’re good campaigns on the wrong platform.
There’s a single question underneath the choice, and it isn’t about budget, industry or company size:
Does your customer already know they want this?
If they do, they are searching, and your job is to be found at that moment. If they don’t, nobody is searching, and your job is to interrupt someone convincingly enough that they start wanting it. Those are opposite motions, and the platforms are built for one or the other.
Everything below follows from that.
The short version
| What you sell | Start here | Why | Add later |
|---|---|---|---|
| Professional & B2B services (consulting, legal, accounting) | Google Ads Search | The need is specific, urgent and searched for. Intent is everything; volume is low but each lead is worth a lot. | LinkedIn for targeted demand generation once Search is saturated |
| B2B SaaS | Google Ads Search + LinkedIn | Search catches the category-aware; LinkedIn reaches the right job title before they know the category exists. Long cycles need both. | Meta retargeting as a cheap reminder layer |
| E-commerce with known demand (established categories) | Google Shopping / PMax | People search for the product by name. Being in the results with a good price and image is most of the battle. | Meta for prospecting once Shopping is capped by demand |
| New or unfamiliar consumer products | Meta | Nobody searches for a thing they’ve never heard of. You need visual demonstration to an audience defined by interest, not by query. | Google Search for brand terms as soon as people look you up |
| High-consideration consumer purchases (furniture, travel, treatments) | Meta + Google Search | Meta creates the want, Search catches it days later. Running only one means paying to create demand someone else collects. | YouTube for demonstration-heavy cases |
| Local services (clinics, trades, restaurants) | Google Ads + Google Business Profile | Local intent is searched, and the map pack often outperforms the ads. Cheap, immediate, geographically bounded. | Meta for local awareness in dense areas |
| Marketplaces & travel | Google Ads + feed-based campaigns | Competing against large portals means winning on specificity — route, class, date — not on generic terms. | Meta retargeting on abandoned searches |
The trap in the middle column
Look at that table again and you’ll notice most rows start with Search. That’s the natural instinct — searched demand converts, so buy it — and it has a failure mode worth naming.
Demand capture can only convert interest that already exists. Keep buying capture without creating any, and you end up competing for a fixed pool of buyers at a steadily rising price. Search Engine Land put it plainly in a piece on the demand capture trap, noting Google’s own research that brands running YouTube alongside search saw an 8% lift in conversion volume and a 4% reduction in cost per acquisition.
So the “add later” column isn’t optional garnish. It’s what stops your Search costs climbing every year.
Where the AI assistants fit now
A growing share of buying journeys no longer starts in a search box. People ask ChatGPT, Gemini or Perplexity for a recommendation and act on the answer.
The paid side is genuinely early. OpenAI began its first live test of ads in ChatGPT in early 2026 — placed in a clearly labelled section beneath the chat rather than inside answers, shown to logged-in free and lower-tier subscribers, with matching based on conversation topics and history (Search Engine Land). Industry expectations for 2026 point the same way: ads appearing across the AI platforms and those platforms operating as network partners (Search Engine Land). Formats aren’t comparable across providers yet and inventory is limited — which is an argument for learning the channel now rather than when it’s crowded.
The organic side matters more today. When someone asks an assistant “who does X in Y”, a small set of businesses get named. Whether yours is among them isn’t random — it depends on how your business is described across the sources these systems draw on, how clearly your offering is stated, and whether that information is consistent everywhere it appears.
The practical starting point is unglamorous: go and ask. Put the ten questions your customers would ask into ChatGPT, Gemini and Perplexity and read what comes back. If competitors are named and you aren’t, you’ve learned something worth acting on. If nobody in your category is named, that’s an opening.
The three mistakes that cost the most
Running Meta for a searched product. If people already look for what you sell, buying interruption means paying to create demand that exists. Take the demand first; it’s cheaper.
Running Search for an unknown product. Nobody searches for a category they can’t name. You’ll find a few hundred low-volume queries, conclude the market is small, and be wrong.
Judging platforms on last-click. Meta creates want; Search collects it. Last-click gives Search the credit and Meta the blame, so companies cut the platform generating the demand and then wonder why Search volume falls two months later. If you run both, you need measurement that can see this — otherwise your reporting will systematically recommend the wrong decision.
Where to actually start
If you sell something people search for: Google Ads, brand and generic separated, tracking right before you scale anything.
If you sell something people don’t know exists: Meta, real creative testing, and the expectation that the first budget buys learning rather than sales.
If you sell to businesses: usually both Google and LinkedIn, with the honest expectation that the cycle is long and the first three months tell you very little.
And in all three cases, spend an hour asking the AI assistants about your category. It costs nothing and it’s the cheapest competitive research currently available.
Not sure which row describes your business? A short call usually settles it — and if the answer is “the platform you’re already on”, I’ll say that too.