Marketer Reacts to ChatGPT 5.2, Disney+Sora, & OpenAI's Final Move of 2025

By Dan Sanchez · Episode published

AI benchmarks can signal real progress, but marketers get better results by testing the work they actually need done, preserving context, and keeping human review in the loop.

Get the next practical AI marketing episode wherever you listen.

Apple PodcastsSpotifyYouTube

Better benchmarks do not remove the need for better judgment

When a new AI model arrives, the first conversation usually centers on benchmarks. Did it score higher? Did it beat the previous release? Did it move up the leaderboard?

Those numbers can be useful, especially when they measure work that looks closer to real knowledge work instead of another abstract test. But they are still only one part of the decision. A model can make startling progress on a difficult task and still miss something simple that a person notices immediately.

That unevenness is the most useful thing for marketers to understand. AI capability is expanding fast, but it is not expanding evenly.

Test the work you actually need done

I tested the model with an old I Spy page full of tiny objects. A previous version could not identify a medium-difficulty object. The newer model found it, though it took time. On a much harder object, it needed a clue before it could locate the answer.

That is a better lesson than a generic claim that vision improved. The model had made real progress, yet it still benefited from human direction. In marketing, that same pattern shows up everywhere. A tool may summarize a transcript, generate a campaign visual, or structure a draft well, then make an error that changes the meaning or makes the result unusable.

The practical response is simple: test AI against a real task from your work. Use material you know well enough to judge. Check where it succeeds, where it gets vague, and what instructions help it recover. That tells you more than a score alone can.

Context and specialization both matter

The source comparison also showed why one model will not win every task. Gemini handled long transcripts well in the workflow discussed. Claude was especially strong for turning transcript material into book chapters. ChatGPT remained the preferred day-to-day tool because its personalization made it easier to work through ideas without re-explaining the full background.

That is not a reason to build an oversized tool stack. It is a reason to notice where a tool earns its place. A model that carries forward the nuance of your project may be more valuable than one that wins a decontextualized comparison. A model with a larger context window may be the right choice when you need to work through a large source file. The job determines the tool.

For a marketing team, this means preserving the context that helps any model perform: the audience, offer, source material, approved claims, brand constraints, and the decisions already made. Then ask the model to do a bounded task. The better the brief and review process, the more useful the output becomes.

Treat rights as part of the creative brief

The discussion around Disney and Sora raised another practical issue. New creative capabilities can make a marketer want to borrow a familiar character, style, or cultural reference. But a platform partnership does not automatically answer every question about how that material can be used in marketing outside the platform.

That is where restraint matters. A creative possibility is not the same thing as permission. Before incorporating a protected character or brand into client work, confirm the terms and the rights that apply to the actual use. The best idea is not useful if it creates an avoidable legal problem.

The AI race is moving quickly, and that will continue. The marketers who benefit most will not be the ones who repeat every release claim. They will be the ones who test the work, keep a human review loop, use context carefully, and know when a creative tool needs a real-world boundary.