AI Makes Content Cheaper. Where Should Your Marketing Budget Go? A Conversation with Yan

AI can make content cheaper. It cannot make a bad idea worth watching.

Lower production costs can create the illusion that the marketing problem has been solved. Yet more assets do not prove that the message deserves a customer’s attention.

A field experiment involving 758 consultants found that GPT-4 helped participants complete some writing tasks faster, while performance fell on a different task beyond the model’s effective capabilities. Production efficiency and sound judgement remain separate questions (Dell’Acqua et al., 2026).

In this edited founder Q&A, Yan explains where she would reinvest the savings: understanding the customer, testing genuinely different ideas, distributing the strongest work and removing the obstacles that appear after somebody becomes interested.

If production costs fall, should the marketing budget fall too?

Yan: Not automatically. I would first ask what the saving allows the business to learn or improve.

If you can now make thirty videos instead of four, that may be useful. But the number of assets is not the decision that matters. You still need to know which ideas are worth producing, who needs to see them and what should happen next.

Making the wrong message cheaper does not make it more persuasive. It can simply help you repeat the wrong message more efficiently.

I would move more of the budget into finding out what deserves the customer’s attention and what helps them choose you. The right balance depends on where the next obstacle sits.

Where should a brand start?

Yan: Start with the customer, not the production tool.

What are they trying to decide? What is stopping them from buying? What difference would actually matter to them?

Those questions help you decide what the content needs to do. One customer may need to understand whether a product fits a small home. Another may care about how it survives everyday family life. Those are two different reasons to buy, even if the same product sits at the centre of both stories.

I would rather test genuinely different ideas than create ten minor edits of one idea. Changing the opening line may improve execution without testing the underlying proposition.

What is the difference between more variations and better testing?

Yan: A useful test changes something meaningful.

Imagine a furniture brand. One piece of content shows how a table works in a small flat. Another shows how it copes with daily family use. Each piece tests a different customer concern.

By contrast, ten versions of the same “beautifully designed table” message may give the team more output without giving it much more understanding. The brand might learn which edit attracts a click, but still not know which reason to buy has real weight.

Before producing variations, write down what each one is testing. A different customer problem, buying occasion or proof point gives the test a clearer strategic purpose.

How should a team decide what is working?

Yan: The evidence should match the goal.

Views can tell you that something was seen. They do not automatically tell you that somebody understood the offer, trusted it or intended to buy.

If the goal is purchase intent, look for behaviour that brings the person closer to a purchase. If the goal is trust, look for signs that people understand and believe the proposition. If the goal is learning, decide what question the test is meant to answer before it runs.

Then increase the spend behind work that produces the relevant signal. This does not require pretending that one metric explains everything. It requires being clear about what the content was supposed to change.

Does cheaper production make distribution less important?

Yan: No. Making thirty videos does not mean thirty videos will reach the right people.

Production creates the asset; distribution gives the right audience a chance to encounter it. I would protect enough budget to put distinct ideas in front of relevant people, then decide whether the response merits broader distribution.

What should brands be careful not to automate away?

Yan: I would protect the work that makes the brand recognisable and believable.

That might be a strong creative idea, a founder explaining a difficult decision or a customer showing how the product genuinely fits into their life. AI can help adapt that material into more formats. But the story, the experience and the evidence still need to be real.

The risk is removing the judgement, specificity and proof that gave somebody a reason to care.

What happens after the content creates interest?

Yan: This is where I think many budget discussions become too narrow.

Can the interested person get their question answered? Can they understand the offer? Can they buy easily? Can they reach someone who can actually help them?

If that part is broken, more content may simply send more people into the same dead end. The best next use of the saving may therefore sit outside content production. It may be clearer product information, a stronger enquiry process, better localisation or a simpler path to purchase.

When entering another market, more content cannot repair a proposition customers do not understand or a route to purchase that does not work for them.

A simple reinvestment framework

Yan’s recommendation is to examine the saving across six possible jobs:

There is no universal percentage for each area. The practical question is: where can the saving remove the next obstacle to growth?

What another founder can take from this

AI creates production capacity. Treat the saving as a chance to improve the decisions around content: produce what is worth testing, distribute it well enough to learn and fix what happens after attention.

That is a more useful ambition than simply making more content for less.

FAQ

Should AI reduce our overall marketing budget?

It may reduce particular production costs, but that does not determine the right total budget. First identify whether research, distribution, measurement, localisation or the customer journey still limits growth.

What should we do with the money saved on content production?

Reinvest it where it can remove the next obstacle: better customer insight, stronger ideas, appropriate human review, useful distribution, clearer measurement or an easier path to purchase.

Is producing more content a good AI strategy?

Only when the extra output has a purpose. More content is useful when it tests distinct ideas, reaches relevant people or adapts proven thinking for a suitable context.

Which marketing metrics should we use?

Choose evidence that matches the content’s job. Awareness, understanding, trust, enquiry and purchase require different signals. Define the decision before selecting the metric.

Does this advice change for brands entering China?

The principle remains the same, but the work may require additional customer research, localisation, platform judgement and route-to-purchase checks. Do not assume cheaper production fixes a weak market proposition.