Why we should always ask Ai what it needs to know

Analysing some of the feedback I get here, it looks as if stepping out of the marketing arena slightly on occasions is quite valued – in particular when it comes to using Ai tools. So I will continue to do this occasionally; let me know what you think.

Using Ai well often means resisting the urge to ask for the final answer straight away. In many cases, better results come from splitting the task into two stages: first using Ai to clarify the brief, then using it to produce the actual output. It is a simple habit, but it can make the difference between something that merely sounds convincing and something that is genuinely useful. When we hear stories of Ai hallucinating, I believe it’s often because it hasn’t been given structure and direction. Here are some examples.

• Structure, then draft. Instead of asking Ai to write an article in one go, we might first ask it to provide several possible angles about the issue under consideration. Then we can choose one and give it some proper guidance when getting it to create the outline (another step often omitted).
• Define criteria, then analyse competitors. Before reviewing competitors, we could first ask: “What are the most important dimensions for comparing the websites of B2B businesses selling blue widgets?” …and only then set the Ai off on researching competitors using those dimensions. That stops the analysis looking at a load of irrelevant angles.
• List assumptions, then recommend strategy. Suppose we’re intrigued to find out the current thinking on whether we should we invest more in webinars or search ads. Don’t just ask. Start off with “What assumptions would you make before recommending one over the other?” then comment on those assumptions before asking it recommend an approach. Context is everything, but it can still be forgotten.
• Define an angle, then write subject lines. To get suggestions for subject lines, we might ask: “What makes an effective subject line for a technical B2B audience in a conservative sector?” Then we can choose the styles and approaches we like, before asking for subject lines emphasising those.
• Likely causes, then diagnostics. Suppose lead quality has dropped. Instead of just asking for ideas why this might have happened, and getting a heap of possibilities with no sense of priority, ask “What are the most likely causes of falling lead quality in a B2B paid search campaign?” “Put those into a diagnostic order, starting with the fastest checks.”
• Objections, then persuasive argument. If we are writing a landing page, a preparatory step could be: “What objections might a cautious buyer have before enquiring?” and then after choosing the most important: “Write the landing page copy so those objections are addressed naturally.” That gives us copy built around actual resistance instead of just asking for landing page copy based around a general list of benefits.

These are just a few random examples. The best way of summarising all this is that Ai tools need comprehensive guidance, or they’ll make assumptions that we don’t even know they’re making. It’s rare that we give them that guidance. However, they’re also good at defining what guidance they need – so why not ask them first?