In short: AI helps most on repeatable, checkable tasks: transcripts, rough cuts, clean-up, captions, versions and scheduling. It helps least where the work is a judgement about people, rights or truth: the brief, the facts, the performance, the likeness. Put a named person in charge of approving every stage, and test one task before you widen.
AI is useful in a production workflow where the task is repeatable and someone can check the result, and it is a bad idea where the task is a judgement about people, rights or the truth. That is the whole page in one sentence. The rest of it walks through nine stages of a video or audio project and says, for each one, which tasks to hand to AI, which to keep, and what to check before the work moves on.
The page is about tasks and not tools on purpose. Tool names change every few months, and a list of them is out of date by the time you read it. The tasks do not change. The Generationally Speaking episode with George Cairns of Get AI Powers makes the same point. Its takeaways say systems matter more than tools, and that most AI projects fail because organisations do not redesign the human system around the technology.
The map
| Stage | Hand to AI | Keep with a person |
|---|---|---|
| Brief | Summarise research, find gaps | What the video must change, for whom |
| Script | First outlines, alternative openings | Facts, voice, claims |
| Pre-production | Draft shot lists, schedules, call sheets | Safety, permissions, casting |
| Shoot | Live transcript, logging | Everything in front of the lens |
| Edit | Rough cut from transcript, selects | Story, pace, final cut |
| Sound | Noise clean-up, levelling | Performance, consent for any voice |
| Finishing | Captions, resizing, draft translations | Brand, accuracy, accessibility |
| Publishing | Scheduling, caption variants | Approval of every post |
| Library | Tagging, search, answering questions | What is allowed to be said |
1. Brief
Use AI to summarise the research, list the questions your audience is likely to ask, and read your brief back to you so you can see what it leaves out. Those are all checkable.
Keep the decisions. What should a viewer think or do afterwards? Who is the audience? Which claims can the company stand behind? An AI can draft answers, but it cannot know your business, and a brief it writes for you will be fluent and generic. Our guide to briefing a studio or a video crew shows what a good one contains.
Check before moving on: a named person has signed the brief.
2. Script
AI is good at the first rough draft: an outline, three alternative openings, a shorter version, a check on reading level. Treat everything it writes as a draft from someone who has not met your customers.
Keep the facts, the voice and the claims. Every statistic, name and date needs a source you have opened. If the script will be spoken by a presenter, read it aloud before you accept it. Our guide to timing a voiceover script helps with length.
Check before moving on: every factual claim has a source, and the script sounds like a person who works there.
3. Pre-production
AI can turn a script into a first-draft shot list, a draft schedule and a formatted call sheet. It can also suggest locations to research and produce mood images for discussion, provided they are labelled as concept art and not used as a promise of the finished result.
Keep safety, permissions and casting. Nobody should rely on an AI-drafted schedule for how long a lighting change takes or whether a rig is safe, and a competent person must assess the real job. See how to plan a studio shoot for the human version of this stage.
Check before moving on: a producer has read the schedule and the risk assessment.
4. Shoot
This is the stage where AI helps least. A camera, a light, a microphone and a person in front of them are still the shoot. AI can transcribe as you go, log what was recorded and tidy notes, and that is about it. If someone suggests replacing the shoot with generated footage, when to use AI and when to shoot it sets out how to decide.
Check before moving on: every camera and every microphone was recorded, and the files are backed up.
5. Edit
This is where AI saves the most time. A transcript lets it find the good moments, strip out false starts and filler, make a rough cut and pull selects on a topic. Our episode-to-calendar workflow for The Experience podcast goes from a recorded episode to a full, checked social calendar in 45 minutes.
Keep the story, the pace and the final cut. A rough cut is a starting point. An editor still decides what the film is about and what to leave out.
Check before moving on: a person has watched the whole cut, with the sound on.
6. Sound
Noise clean-up, levelling and transcription are the easiest wins in the whole process. They are checkable by ear and they save hours.
Voice is different. Synthetic voices and voice clones are a consent question before they are a technical one. Do not clone a voice without the speaker's agreement in writing, and label synthetic speech so listeners know. We used a consented cloned voice for part of our audio course, and it is labelled as synthetic. The voice and likeness consent checklist lists the points to cover, and how we clone a voice with consent shows the process.
Check before moving on: anything synthetic is labelled, and anyone whose voice or face is used has agreed in writing.
7. Finishing and versions
Captions, resizing for vertical and horizontal formats, draft translations and descriptions all suit AI. They are repeatable and a reader can check them.
Keep the brand and the accessibility sign-off. Captions need a read-through for names and technical terms. A translation needs a speaker of the language to look at it before it goes out.
Check before moving on: one person has read every caption and every translated line.
8. Publishing
Scheduling, caption variants for each platform and a monthly summary of what performed are routine, and routine is what AI does well. We run a weekly content engine with a person approving each post, and that approval step is the point of it.
Check before moving on: nothing goes live without a name against it.
9. Library and reuse
Once you have a library of finished work, AI can tag it, make it searchable and answer questions from it. Our own site has an "Ask the studio" voice and text agent. Decide what it may and may not say before you switch it on, because the library may contain things that were right for one audience and wrong for another.
Check before moving on: someone owns what the library is allowed to say.
Where AI goes wrong
- No named approver. If nobody owns the check, nobody does it.
- Starting with the tool. Pick a task, then a tool for it.
- Trusting fluent text. A confident paragraph is not a checked one.
- Using a face or voice without consent. See who is responsible for an AI-made ad.
- Removing the person. The Cairns episode puts it that automation should improve human work, not just remove it. If the time you save is not spent on something better, you have saved nothing.
- Skipping the explanation. The same takeaways say that teaching the why beats teaching the button. If your team does not know why a step exists, they cannot tell when the tool has got it wrong.
This is general information, not legal advice. For rights, consent and liability questions, ask a solicitor.
How to start
- Pick one repeating task. Choose something your team does every week, such as turning a recording into captions and clips.
- Write the steps down. Mark each one as repeatable or as a judgement.
- Try AI on the repeatable steps only. Keep a person on the judgement steps.
- Name the approver. One person, one checklist.
- Time it before and after. Include the time spent checking.
- Keep a log of mistakes. What it got wrong tells you where the guardrails go.
- Widen only when the log is quiet.
If you would rather have someone map this for your own business, The AI Studio starts with a conversation, and its AI Strategy Intensive produces an opportunity map and a 90-day roadmap. For a wider view of what is real for small businesses today, see AI video for small businesses.
Sources
- Generationally Speaking, From Hype to Workflow: Making AI Actually Useful (episode 30, George Cairns of Get AI Powers), read 5 November 2026
Related guides
- AI video for small businesses: what's real, what's risky
- When to use AI and when to shoot it for real
- Who is responsible for an AI-made ad?
- Voice and likeness: a consent and contract checklist
- How we clone a voice with consent
- All guides and tools
The picture at the top of this page is an illustration, not a photograph of a client or a real job.
Questions we get asked
Where can AI be used in video production?
AI suits repeatable tasks such as transcribing, finding the best moments, rough edits from a transcript, noise clean-up, captions, resizing and scheduling. It is a poor fit for the brief, fact-checking, casting, performance and anything that depends on someone's consent. Every output still needs a person to check it.
Will AI replace a video production team?
Not as a whole. The shoot, the story and the sign-off still depend on people, and what changes is how much time they spend on repeatable tasks. The better question is which tasks in your process are repeatable and checkable, and what your team does with the time saved.
What should I ask a production company about its use of AI?
Ask which stages use AI, who checks the output, whether anything in the finished work is synthetic and will be labelled, what happens to your material and scripts, and who is responsible if something is wrong. A good answer names a person and a process, not a product.
granary.digital · 01243 884993 · studio@granary.digital · The Old Grainstore, Adsdean, Chichester, West Sussex, PO18 9DN · Prices exclude VAT unless marked inc VAT