Direct answer
Not first. An AI marketing tool doesn't decide who you're for, why they'd pick you, what you sell or what you say — it produces more of whatever answers already exist, faster. If those answers were never agreed, the tool multiplies the vagueness. Gartner's 2026 survey found marketing leaders putting 15.3% of budgets into AI while only 30% reported mature readiness. Settle the four decisions first; then a tool has something worth scaling.
The pitch for AI marketing tools is almost always about speed. Ten articles a week instead of two. A personalised email for every segment. Ads in forty variations by lunchtime. The pitch is accurate. The tools do all of that. What the pitch leaves out is the question speed can't answer: forty variations of what?
An owner looking at that pitch is usually being asked to decide on a tool before anyone has decided on the thing the tool is meant to carry. That is the order most businesses buy marketing in, and AI makes it more expensive, not less, because it removes the one brake the old way had. When writing a homepage took three weeks, somebody eventually had to agree what it should say. When it takes three seconds, nobody has to.
What the surveys actually show
The money is already moving. Gartner's 2026 CMO Spend Survey, published in May, found marketing leaders allocating an average of 15.3% of their marketing budgets to AI. The same survey found that only 30% report mature or fully developed AI readiness, and 70% say their internal marketing processes are not yet mature enough to implement and scale it. That sample is 401 marketing leaders in North America, the UK and Europe, the vast majority at companies with more than $1 billion in revenue — organisations with marketing departments, data teams and budgets an owner-led business will never have. If the gap between spend and readiness is that wide at that scale, it is not narrower in a company of 120 people.
The CMO Survey, run by Duke University with 308 marketing leaders at US companies in January 2026, tells the same story from the usage side. Its 2026 report says AI use in marketing has more than tripled since 2022; its own series puts AI at 13.1% of marketing activities in 2024 and 24.2% in 2026, with generative AI alone at 22.4%. Over the same two years, it reports, the performance of marketing technology has not improved: no marketing-technology activity scores above 5 on its 7-point scale. When the same marketers rate how well AI-produced strategy fits their target markets, the average is 4.4 out of 7.
Put those two surveys side by side and the pattern is plain. Adoption is racing. Results are not. The tools work; what they are being pointed at is the problem.
AI doesn't fill the gap where a decision should be. It fills it with the average.
Why the tool multiplies the missing decision
Every piece of marketing rests on four decisions: who this is for, why they'd pick you over the alternative, what exactly you sell them, and what you say to them. Call it the upstream audit. A person writing by hand has to make those decisions, even badly, to finish a sentence. A generative tool does not. Give it an undecided brief and it completes the sentence anyway, with the most statistically likely words — which is to say, the words everyone else in your category already uses.
That is not a flaw in the tool. It is what the tool is for. A language model trained on millions of homepages writes the average homepage extremely well. If your business has decided what makes it different, the tool can carry that difference into forty formats at once. If it hasn't, the tool carries the absence into forty formats at once, and now there is forty times more of it to undo.
From the book: before a first meeting with an accountancy practice, I put screenshots of nearly thirty accountancy homepages from Belgium and the Netherlands on a wall, with theirs among them, and asked whether they could pick out their own — or say why a visitor should choose them. They couldn't. The typical opening line in that group read something like: "Since 2015, we offer bookkeeping with expertise and personal attention." That sentence is exactly what an AI tool produces when nobody has told it anything specific. Those thirty firms reached it by hand. A tool would have got them there faster, and in more languages. For the record, I may have been too direct in that meeting; they chose another partner.
What to decide before you buy anything
None of this means an owner should stay away from AI. It means the order matters. Settle the four decisions in writing, in language a new salesperson could use on a Monday. Then ask what a tool would be for — and the question gets a real answer: carrying this message to this buyer, in these places, at a volume a small team could never manage by hand. That is a tool worth paying for. Buying the tool first, and hoping the decisions will emerge from the output, is how a business ends up with more content and the same homepage everyone else has.
The same arithmetic applies to budget. More spend multiplies whatever message it is given, which is the argument in why spending more on marketing won't fix it; an AI tool is simply a faster way of doing the same multiplication. If you are not sure the audience decision has ever been made in your business, the clearest signs you never decided who your ideal customer is gives you the checklist. The five-question Quick Diagnostic runs the short version of the test in ten minutes, and the longer argument — decide first, then spend — is Reports, Not Revenue.
| Tool first | Decisions first | |
|---|---|---|
| What the tool is given | A category, a product list, "write something engaging" | A decided buyer, a reason to choose you, a defined offer, a message |
| What it produces | The average of the category, in volume | Your position, carried into more places |
| What the reports show | Output: posts, variations, impressions | Conversations with the buyer you chose |
| What it costs to undo | Every asset it made, rewritten | Nothing; the decisions were made first |
- Before any AI purchase, write the four decisions — audience, positioning, offer, message — in one sentence each. If you can't, that is the first job, not the tool.
- Ask the vendor to show what the tool produces from your current homepage copy. If the output could belong to any competitor, so could your input.
- Name the single task the tool will carry, and for which buyer. "Everything, faster" is not a task.
- Measure it on conversations with the buyer you decided on, not on the volume of output.
- Revisit the tool once the decisions change — not the decisions once the tool arrives.
