FARMCreate useful AI-assisted TikTok content
Use practical AI prompts to produce TikTok hooks, storyboards and slideshow variants, review product accuracy and deliver approved files to PhoneFarm.
Use AI to shorten the work between a tested content idea and an approved production packet. Give it the product facts, ask for a specific output, choose the useful version, then film or edit the evidence. Asking for “a viral video” skips the decisions that make the output usable.
This hypothetical workflow uses Example Speaking, an app with recording, replay and retry. The deliverable is three opening variants, one demonstration body and a six-card version. PhoneFarm receives the finished video files; the AI work happens in your chosen production tools.
Choose the job you want AI to do
| Production job | Give the model | Ask it to return |
|---|---|---|
| Find angles | Anonymized real questions, audience and product facts | Grouped customer problems and possible demonstrations |
| Write hook variants | One problem, one body and a fixed offer | Distinct openings that lead into that same body |
| Build a storyboard | Approved script and available footage | Shot-by-shot recording list |
| Adapt a format | Approved script and the target format | Short card text or overlays with visual directions |
| Localize | Source copy, intended audience, product availability | Natural wording plus terms a local reviewer should check |
| Review an edit | Transcript and selected frames | Specific unclear claims, repeated lines and missing shots |
Choose one job per pass. Combining research, claims, script, visuals and performance judgment in one request makes errors harder to find.
Give the model facts before asking for ideas
Prepare a small source packet: the feature list, a current app screen recording, the audience problem and the action you want the viewer to take. Remove personal information from example accounts and customer questions.
Here is a complete prompt you can adapt:
You are writing for job seekers whose written interview answers sound awkward aloud. Our app records an answer, replays it and lets the user record another attempt. Available footage: written notes, a presenter trying one answer, the real record/replay controls and a shorter second attempt. Keep this body fixed: first attempt → replay → shorten one sentence → retry. Write six spoken openings across three approaches: recognizable frustration, challenge a habit, and show a contrast. For each, give the exact words and opening shot. Use conversational language. Do not add features, statistics, testimonials or outcomes outside these facts. End with “Try one answer before rehearsing the whole interview.” Return a table, not an essay.
A usable selection might be:
| Approach | Chosen line | Shot |
|---|---|---|
| Recognizable frustration | “Your interview answer sounds better in your head.” | Notes beside the presenter |
| Challenge a habit | “Stop rehearsing the whole interview. Fix one answer first.” | Replace a long question list with one question |
| Contrast | “Read it. Say it. Different answer?” | Written sentence beside the spoken attempt |
Keep the selected copy, then discard weaker options. A folder of fifty unused hooks is not a production batch.
Turn the chosen idea into a production packet
Use a second prompt after selecting the opening:
Build a 20–25 second storyboard for the selected hook. Use the approved body and actual app controls. Return columns for approximate time, exact spoken line, shot and on-screen text. Show the useful action before the product invitation. Flag any missing footage. Also write a six-card adaptation: one readable sentence per card and one visual instruction. Preserve the product facts and ending.
Review the sequence against the actual footage. A good AI draft might write “Replay the answer and cut the sentence that drags.” That is filmable. If it adds “Our AI automatically rewrites your answer,” remove it: this example's supplied feature list does not contain that function.
Send the editor the approved table, source clips, text style, caption and destination. Use the complete script recipes as a reference for how much detail the packet needs.
Keep creative variations meaningful
For the first test, keep the body, product action and ending fixed while varying the opening. AI is useful here because it can suggest several ways into the same demonstration.
After choosing an opening worth repeating, expand the substance: a different interview question, a clearer example or a real audience objection. Keep those changes in a separate batch. That lets the team distinguish an opening test from a topic test.
A useful adaptation prompt is:
Keep the “attempt → replay → improve → retry” structure. Adapt it from “Tell me about yourself” to “Why do you want this role?” Give an original example answer, a specific sentence to shorten and the required footage. Keep every product feature unchanged. Mark the answer as a demonstration, not a customer testimonial.
Use generated visuals for the right purpose
Use generated illustrations to explain a concept or support a designed card. Use actual product footage for feature demonstrations. A generated screen can make an attractive picture while inventing buttons your customer will never find.
For a card illustration, a useful prompt is: “Create a simple editorial illustration of a long written answer beside a short spoken sentence. Leave clear space for text added in the editor. No app interface, logos, testimonial or realistic identifiable person.” Add the approved wording in your editor so spelling and typography remain under your control.
Judge a generated scene by whether it helps the story. Check unnatural motion, changing objects, unreadable text and visual repetition. Changing a few pixels in someone else's content neither improves the idea nor gives you ownership of it.
Run an AI-specific quality pass
Before export, answer five questions:
- Does every feature shown exist in the supplied product?
- Can the viewer follow the action without a missing shot?
- Are any claims, quotes or results invented?
- Is the first frame readable and the spoken copy natural aloud?
- Does the final invitation lead to the promised task or page?
Then use the asset checklist for sound, framing and the finished file. Repeatedly correcting the same error is a reason to improve the source packet or prompt, not review it manually forever.
Understand SynthID, C2PA and visible watermarks
These describe different things: SynthID embeds a detectable watermark in supported generated media; C2PA Content Credentials record signed provenance information; visible watermarks and platform labels communicate information to viewers. The generation tool and export path determine which signals are present.
Keep provenance separate from the reach question
A provenance signal alone does not establish a distribution penalty. Evaluate the creative and audience response instead of assuming every weak result came from AI.
Check the finished media's disclosure needs
TikTok requires labeling for realistic AI-generated images, audio and video, and may add automatic labels. Review the actual media and follow its AI-content guidance. Writing a script with AI and generating a realistic person speaking are different production choices. Commercial-content disclosure and music rights also need the appropriate checks.
Move the approved result into PhoneFarm
Export named versions such as SPK-01_hook-b_v1.mp4. Upload and preview them in Videos, keep the caption and intended account with each file, and follow the available publishing workflow.
After verifying publication, compare the planned creative variable using the testing guide. Keep the prompt and approved packet beside the result so the next batch can repeat the useful process.