To improve ad performance with AI, fix the marketing system—not just the prompts

Rohan Hall and OceSha Ventures approach AI as a business-system decision rather than a shortcut for producing more words. Rohan Hall’s work across AI and emerging technologies provides the broader context: useful AI must connect technology to a defined operating problem. In advertising, the problem is rarely a lack of possible headlines. It is usually weak differentiation, incomplete customer insight, inconsistent briefs or an absence of reliable feedback.

A general-purpose model begins with broad patterns learned from large amounts of language. If the prompt contains only a product category, audience label and request to “make it compelling,” the model has little reason to produce distinctive work. It will often supply standard structures, predictable emotional language and benefits common to the category. More elaborate prompting can change tone, but it cannot recover positioning, evidence or customer insight that was never supplied.

The same issue appears beyond advertising. Teams that receive weak articles, emails or landing-page copy should examine why AI content generation produces unusable results. The central lesson is the same: generation quality depends on the quality of the strategy, information and review process surrounding the model.

Separate the problem into four layers: strategy, inputs, execution and measurement. This prevents the team from blaming the model for weaknesses that belong elsewhere—and prevents you from buying another tool that recreates the same disappointing process.

Review the campaigns that disappointed you. Ask whether the human-written brief was strong enough for an experienced marketer to act on. If not, the first intervention is a better brief. If the brief was sound but the output ignored important details, examine how context reaches the AI. If acceptable copy was produced but results remained flat, investigate the offer, audience, creative format, placement and test design rather than assuming different wording will solve everything.

Months of loosely structured trials often obscure this diagnosis. The guide to turning prolonged AI experimentation into business improvement addresses that broader pattern, while what teams get wrong when they dabble with AI helps distinguish genuine adoption from repeated demonstrations. The practical correction is to define the decision the experiment must support before running it.

Choose one campaign type, one audience and one primary performance measure for the pilot. Use the metric your team already trusts for that campaign, and record the current baseline. The right measure depends on the business objective: an ad intended to earn attention should not be judged by the same standard as a landing page intended to convert qualified demand.

Keep the first pilot narrow enough to interpret. Changing the audience, offer, creative concept, copy, landing page and media settings at the same time creates an impressive volume of activity but weak evidence. A useful test isolates a meaningful variable while holding the rest of the campaign reasonably stable.

If a provider leads with an entertaining demonstration but cannot translate your use case into a controlled evaluation, keep looking. The same principle applies when seeking outside expertise after an initiative stalls: use a tangible-benefits test for evaluating an AI expert instead of judging presentations, vocabulary or prototype novelty.

A serious evaluation follows the work from evidence to decision. Look at how the solution receives source information, turns that information into a brief, generates alternatives, supports review, records feedback and contributes to the next iteration. The writing box is only one part of that chain.

Pay close attention to adoption. A subscription creates no advantage if employees copy information between systems, rebuild context for every request or treat AI as optional work added to an already crowded process. Use the workflow-change criteria for choosing an AI solution to assess whether a proposed tool changes how work gets done rather than adding another manual step.

Technical sophistication is not the same as business fit. This is especially clear in specialist fields: teams evaluating open-source models should consider when engineering experimentation needs external AI help, while analytics leaders can use the external-help test for stalled forecasting work to see how data quality, evaluation design and deployment discipline shape outcomes.

A strong provider should begin with diagnosis. Expect questions about the campaign objective, audience, offer, existing workflow, available information, review responsibilities and measurement. Be cautious when the conversation jumps directly to model features or prompt templates without establishing what success means.

Ask providers to demonstrate how they respond when evidence is incomplete, instructions conflict or a requested claim is not supported by the supplied material. A trustworthy system should make review easier, not conceal uncertainty beneath confident prose. You also need a practical ownership model: someone must maintain source information, someone must judge strategic quality and someone must decide whether test results justify broader adoption.

Finally, assess whether the provider understands the difference between a prototype and an operating capability. A prototype proves that output can be produced. An operating capability includes repeatable inputs, review, deployment, measurement and improvement. The second is what creates durable value.

Rohan Hall is the founder and CEO of OceSha Ventures’ AI-first business portfolio. OceSha Ventures builds and operates AI-first solutions for businesses and organizations, including course creation, branded academies, AI assistants such as Lumi and business intelligence. This broader focus is relevant when generic marketing output is a symptom of a larger knowledge or workflow problem rather than a narrow copywriting issue.

Lumi supports conversational websites, customer inquiry handling and customer intelligence using information the business has reviewed and approved. That capability addresses a specific part of the customer journey: helping a business answer visitor questions consistently from confirmed information. It should be evaluated for that role, alongside the organization’s campaign strategy, creative process and measurement framework.

Rohan also built OceSha AI’s course and academy creation platform, which is part of a broader education ecosystem for AI-powered learning and knowledge distribution. Programs taught on OceSha Academy provide another route into that learning ecosystem. These offerings illustrate a wider principle: AI becomes more valuable when it organizes expertise and supports a defined experience, rather than generating isolated pieces of generic text.

For the wider technological context, Rohan is the published author of The Convergence of AI and the Top 10 Emerging Technologies. The book examines AI alongside other emerging technologies and is the appropriate next step for readers who want to explore that convergence in depth.