Positioning before AI-generated marketing

A founder asks an agent to improve the homepage. Twenty seconds later, the product “transforms workflows”, “unifies teams” and “delivers intelligent insights”. The sentences are smooth. They could belong to almost any software company. Nothing in them helps a customer decide.

The best positioning and customer-research practices offer a useful way out. Start with the customer's real competitive alternative, learn from what they already tried and listen for the progress they are trying to make. The alternative might be a direct competitor, a spreadsheet, an assistant, a manual process, or doing nothing. Only then can the company identify what is distinct, what value that creates and who cares enough to change.

This combination of competitive context, demand-side progress and evidence-safe interviewing matters in the AI era because language is no longer scarce. Specific customer truth is.

Figure 01 · The customer comparison determines the message; the agent only varies its expression

The method: make the right comparison first

Positioning is the context that makes a product understandable. A capability has no useful value in isolation; its value appears against an alternative. “Remembers company decisions” matters if the current alternative is repeatedly explaining the company to a new chat. It matters less if a customer already has a trusted operating system and only wants a writing assistant.

A useful positioning sequence becomes a chain of evidence:

Name the alternatives customers genuinely consider, including the status quo.
Identify what you can do that those alternatives cannot.
Translate those differences into outcomes customers value.
Find the customers for whom that value is urgent and material.
Choose a market frame that makes the value easy to understand.

The order matters. If a team chooses a fashionable category first, it tends to bend every customer conversation towards that story. Starting with alternatives forces the positioning to survive contact with the buying decision.

The AI-era translation: generate after the choices

An agent is excellent at expanding a defined message into channel variants. It can turn one approved customer problem, promise, proof and limitation into a landing page, sales note or interview invitation. It can also compare replies, objections and outcomes across those variants. But it should not quietly invent the positioning underneath them.

Give the agent an accepted message architecture before asking for copy:

Audience: an observable customer group, with explicit exclusions.
Problem: the recurring job or friction in the customer's words.
Alternative: what they do now and why it remains attractive.
Promise: the outcome your product can reasonably help produce.
Proof and limitation: what supports the claim and where it does not apply.
Next step: one action appropriate to this customer and channel.

Then let it generate. The constraint produces better variety because each version has a point of view. More importantly, results can revise a named hypothesis rather than disappear into a report of clicks and replies.

The same architecture should survive the sales conversation. A useful pitch does not begin with a feature tour; it helps the buyer understand why the current approach is no longer enough, what has changed, which outcome is now possible and why this product is credible. Marketing and sales can vary the format without quietly changing the company's claim.

Turn positioning into a learning loop

Positioning is not settled by a workshop. A founder may believe the alternative is an outsourced operator, while interviews reveal that customers are comparing the product with hiring nobody and tolerating the pain. That changes the proof, the urgency and the category language.

Run one bounded test at a time. Preserve the audience, alternative, message and source material used. Decide in advance which response would cause you to continue, change or stop. Keep observations such as “three prospects described the current process as tolerable” separate from the conclusion “urgency is weaker than expected”. A model can help find patterns; a person should accept the belief revision.

This is the same discipline behind a sales system that can say “we don't know”. Uncertainty is not a messaging defect. It is a reason to choose the next useful conversation.

Where the practice breaks

Teams can make positioning too tidy. Customer decisions are messy, different segments compare differently, and a company can mistake a memorable framework for proof. AI adds another trap: it can synthesise weak interviews into a confident pattern or classify every reply as support for the current thesis.

Preserve the contrary evidence. Keep the source near the conclusion. Do not infer an industry, budget or motivation because it makes the segment easier to describe. And do not keep testing copy when the deeper truth is that the product offers no meaningful advantage. Positioning cannot manufacture differentiated value that customers do not experience.

Questions for the operator

Which alternative do customers actually compare with, not which competitor worries the team?
What evidence supports the differentiated value you want to claim?
Which result would change the positioning rather than merely change the copy?

Take the last five customers who chose you, and the last five who did not. Ask what each would have done without you. Group the answers by actual alternative, not company demographic. For the strongest group, write one sentence describing what becomes possible with your product that was difficult before. Attach the customer evidence.

Now ask the agent for three messages, each for a different channel but all using that same audience, alternative, value and proof. You have moved AI from inventing your market to helping you communicate and test a market choice. Next, choose the first channel capable of testing it, then ask whether the offer can earn a real customer commitment.