Why AI Can't Fake Spontaneity: Farah Khan's Edits Film Festival Insight and the Future of AI Content

Every industry that makes things is asking the same question right now: if a machine can generate a thousand versions of anything in an afternoon, what is actually worth making? At the Edits Film Festival, filmmaker Farah Khan offered an answer that sounds simple until you try to build it into software. “Content breakthroughs come from spontaneity, relatability and realness,” she said.

That short sentence is a quiet but serious challenge to the way most AI content tools are being sold. The pitch for generative AI has been volume: more posts, more cuts, more variations, faster. Khan’s point cuts in a different direction. The things that make content break through, an unplanned moment, a feeling of “that’s me,” a sense that someone real is behind the camera, are exactly the things that are hardest to prompt, hardest to scale, and easiest to sand off when you optimise for output.

For anyone building, buying, or regulating AI tools, that gap between volume and resonance is the most important design problem of the next few years.

The Three Words That Break Most AI Workflows

Spontaneity cannot be scheduled

Spontaneity is the enemy of a pipeline. It is the laugh that wasn’t in the script, the reaction shot nobody planned, the line delivered slightly wrong that turns out to be the best line in the film. AI systems are excellent at executing a plan and terrible at noticing when the plan should be abandoned.

Most current generative tools reward precision. You describe what you want, and the model delivers something close to the description. But the best creative moments usually arrive when the description was incomplete. A tool that only follows instructions will reliably produce content that looks finished and feels empty. The next wave of useful AI will be judged less on how well it obeys and more on how well it surprises, how good it is at offering an option the creator did not ask for but immediately recognises as better.

Relatability is a data problem with a human answer

Relatability is where AI has a genuine advantage, if it is used honestly. Models are trained on enormous amounts of human expression. They can tell you which phrasing sounds like a real person and which sounds like a press release. They can spot that a script is stiff, that a caption is trying too hard, that a thumbnail reads as an advertisement rather than a moment.

But relatability is not the same as familiarity. A model trained to produce the most average version of a sentence will produce something nobody objects to and nobody remembers. The tools that win will be the ones that use that knowledge to remove the artificial parts, not to generate more of them. Think of AI as an editor that strips out jargon, not a factory that stamps out sameness.

Realness is becoming a premium product

As synthetic video, voice, and imagery improve, the value of verifiable realness rises. Not because audiences will consciously check, but because they will feel it. A room full of polished AI-generated pieces starts to blur together. The unpolished, the imperfect, the obviously human starts to stand out, not as a gimmick, but as a signal.

This has a direct commercial consequence. If everyone can produce clean, competent content at near-zero cost, competence stops being a differentiator. What is scarce becomes: a real person with a real point of view, a real moment captured once, a real story that could only have come from that specific place and time.

What This Means for the Future of AI

The near-term future of AI in media is not “AI replaces the creator.” It is a split into two very different jobs.

Job one: the volume work. Captions, cuts, translations, thumbnails, reformatting for every platform, first-draft scripts, rough edits, tagging, search, and scheduling. This work is already being automated, and it should be. It is not where breakthroughs come from, and treating it as the whole job misunderstands what audiences respond to.

Job two: the capture of real moments. Getting a camera in front of a person who is not performing, and preserving what happens. This is the part that AI cannot do for you, and the part that is about to become the most valuable skill in content.

The tools that understand this split will look different from the tools that dominate today. Instead of a blank box that generates a finished asset, expect more systems that:

• Suggest, rather than replace. Tools that surface three unexpected directions and let the human choose, instead of producing one “correct” answer.
• Preserve imperfection. Editors that flag the unpolished take as a candidate, not an error to be cleaned up.
• Prove provenance. Features that attach a verifiable record of how something was made, so audiences and platforms can tell real from synthetic.
• Learn a specific voice. Systems tuned to one creator’s or one brand’s actual way of speaking, rather than the average of everything on the internet.

Practical Implications for Businesses

Stop measuring content by how much you make

Teams that adopted AI for speed are now drowning in output that performs roughly the same as before, or worse. The useful metric is not posts per week. It is how many pieces made someone stop, share, or reply. If AI triples your output and your engagement stays flat, you have not gained efficiency, you have gained noise.

Budget for the un-automatable part

If AI handles production, the constraint moves to capture: access, trust, and time with real people. That means real interviews, real behind-the-scenes, real customer conversations, real footage of the product being used badly and fixed. Companies that shift part of their content budget from production to access will have material their competitors cannot generate, because it did not come from a model.

Use AI to protect authenticity, not to simulate it

The most underrated use of these tools is subtraction. Feed a draft to a model and ask it to find the sentences that sound like a committee wrote them. Ask it which parts a real person would never say out loud. Ask what is missing that would make this feel like it happened. This is a far better use of the technology than asking it to invent a personality.

Be explicit about what is real

Audiences forgive a lot, but they do not forgive being fooled. Clear labelling of synthetic material, and clear signals when something is genuinely captured, builds the kind of trust that survives a scandal. As detection tools improve and platform rules tighten, transparency will move from nice-to-have to requirement.

The Risks of Getting This Wrong

There is a real risk that the next few years produce a flood of content that is technically fine and culturally invisible. Homogenisation is the quiet danger of generative tools: when everyone uses similar models with similar prompts to reach similar audiences, everything starts to sound the same. The result is not just boring, it is brittle. Brands that have smoothed away every rough edge have nothing left for an audience to hold on to.

There is also a trust risk. As synthetic media improves, audiences may become cynical about everything, including genuine footage. The defence against that is not better generation; it is better proof. Creators and companies that invest early in verifiable, documented, human-made material will be the ones whose work still lands when everyone else is suspected of faking it.

What to Do This Quarter

• Audit your content for sameness. Pull your last twenty posts. Count how many could have been made by anyone. If the answer is most of them, the problem is not your tool.
• Change the brief. Ask for one unscripted moment per campaign, and protect it through the edit instead of cutting it for time.
• Rewire your AI prompts. Move from “write this” to “find what sounds fake in this.”
• Add a provenance step. Record how real footage and real quotes were captured, and keep that record with the asset.
• Measure resonance, not volume. Pick one signal that means a human responded, and report it next to your output numbers.

The Bigger Picture

The AI debate in media has been stuck for a while between hype and fear: everything will be automated, or nothing will work. The more useful framing is that AI is compressing the cost of production toward zero, which pushes all the value toward the things that cannot be compressed, a real person, a real moment, a real point of view.

Khan’s formulation is a good test for any tool, any campaign, and any team. Does this create room for something unplanned? Does it feel like it was made by someone specific for someone specific? Is it real? If a piece of content fails all three, no amount of generation speed will save it. If it passes all three, AI can help you get it in front of more people, faster, which is exactly what the technology is actually good for.