Selects is not a search box, and it is not a magic "make it good" button. It is an assistant editor that reads your footage. The editors who get usable rough cuts on the first try are the ones who brief it like an assistant editor: state the goal, name the material, give the rules, define the shape. Below is the anatomy of a prompt that works, a library of prompts by task, and the ones that reliably waste your time.
What chat-based editing is in Selects
Chat-based editing means you describe the edit you want in plain language and Selects performs it on real footage, instead of you performing it clip by clip on a timeline.
The important part in Selects is what the chat has access to. Selects analyzes both the transcript and the video itself before you type anything. It transcribes, detects speakers, syncs multi-cam, identifies silences and filler words, and organizes clips by scene and topic. So when you prompt, you are not describing an effect to apply. You are giving editorial direction to something that has already watched the footage.
That distinction changes how you should write. A prompt in a generic AI video tool is a command. A prompt in Selects is a brief.
Selects also stops where craft begins. It builds the stringout and the storyline, then hands off to Premiere, Final Cut Pro, or DaVinci Resolve with organized bins, color-coded topics, markers, and transcripts. Effects, grade, sound design, titles, and music are yours. Prompting for them is the single most common way people waste a run.
For how to open the chat panel and start a draft, see Prompt-based AI editing: generate a draft using AI chat.
The one mistake almost everyone makes
The one mistake almost everyone makes is prompting like it is a search box.
"funny parts"
"best moments"
"cut it down"
These fail not because the AI is weak but because they are unanswerable. Funny to whom? Cut down to what length, for what platform, keeping which thread? An assistant editor handed that note would come back with questions. Selects, being agreeable, comes back with an edit, and it will be someone else's edit.
The fix is not longer prompts. It is prompts with decisions in them.
The four-line brief
Every prompt that works in Selects has four things in it. Not always four sentences, but always four decisions.
Line | What it answers | Example |
Goal | What is this edit for? | "A 12-minute YouTube episode from a 90-minute interview" |
Material | Which footage, which section, which speaker? | "Cam A and Cam B only, skip the pre-roll before the slate" |
Rules | What to keep, what to cut, what never to cut | "Cut all setup chatter. Keep every question in full. Never trim mid-sentence." |
Shape | Order, pacing, structure | "Open with the strongest claim, then chronological, close with the advice segment" |
Put together:
"Build a 12-minute YouTube episode from this interview. Use Cam A and Cam B, skip everything before the slate. Cut setup chatter and repeated takes, keep every question in full, never trim mid-sentence. Open with his strongest claim about pricing, then run chronological, and end on the advice segment."
That is a brief. It is also the difference between a rough cut you refine and a rough cut you rebuild.
Tip: if a human assistant editor could not execute your prompt without asking a follow-up question, Selects cannot either.
Prompt library by task
This prompt library groups the most useful prompts by what you are trying to do, with a weak version and strong versions of each.
Files, folders, and structure
Prompting for files, folders, and structure is the most underused part of chat in Selects, and the highest leverage. The organization pass determines what every later prompt can reference. Get names and grouping right first, and the rest of the session gets easier.
Weak prompt: "organize my footage"
Strong prompts:
"Group everything by shooting day, then by scene. Name each bin with the date and a three-word topic summary. Put all B-roll and cutaways in a separate bin, and flag any clip with no usable audio."
"Reorganize the stringout by topic instead of chronology. Merge the two segments where we discuss hiring into one bin called Hiring, and keep the order within each topic chronological."
"Split this into three bins: on-camera interview, walk-and-talk, and product demo. Anything ambiguous goes in a Review bin, do not guess."
"Rename clips to speaker plus topic plus take number. If a take is a restart, mark it as such rather than renaming it."
Why these work: they specify the grouping key (day, topic, setup), the naming convention, and what to do with the ambiguous middle. That last one matters more than people expect. Telling the tool where to put uncertainty stops it from quietly making choices for you.
Finding a clip
Selects can find a clip by description because it has read the transcript and watched the video. Use both channels. Describe what was said, or what was visible, or both together.
Weak prompt: "find the good part"
Strong prompts:
"Find where she explains the pricing change and starts with 'the reason we did this'."
"Find every moment someone laughs after a question, and give me two seconds before the laugh."
"Find the section where he is holding the product up to camera and talking about the material."
"Show me all the times the whiteboard is visible and the speaker is pointing at it."
"Find the third take of the intro, the one where he does not stumble on the company name."
Why these work: they combine a spoken anchor (an actual phrase, a topic) with a visual or behavioral anchor (holding something, pointing, laughing). Quoting even a fragment of remembered dialogue is the single strongest retrieval signal you can give.
For the search panel version of the same thing, see Search Your Footage: Text and Scene Search.
Ideating and planning the edit
Do your ideating and planning before you ask for a cut. Chat is genuinely useful as an editorial thinking partner because it has read all 90 minutes and you have not, not carefully, not yet.
Weak prompt: "what should I do with this"
Strong prompts:
"Read the whole transcript and give me three possible storylines for a 15-minute episode. For each, list the segments in order and tell me what gets left out."
"What is the strongest 60 seconds in this recording and why? Give me your top three candidates with timecodes."
"This is a two-hour lecture. Propose a chapter structure with titles and timecodes, and flag any section that repeats an earlier point."
"Where does this conversation drag? List the three longest stretches with no new information."
"We promised the client a hero video and three social cuts. Propose what each one should be, drawn from different material, no overlap between them."
Why these work: they ask for options and reasoning, not a decision. You stay the editor. Use this pass to argue with the tool, then commit, then cut.
Trimming, cutting, and restructuring
Trimming, cutting, and restructuring is where precision in the prompt pays off most, because vague trimming instructions produce cuts that are technically correct and rhythmically dead.
Weak prompt: "remove silence"
Strong prompts for silence and filler words:
"Remove silences longer than 0.8 seconds, but leave pauses that follow a question. Those are thinking beats and I want them."
"Cut filler words: um, uh, like when used as filler, you know. Leave 'so' and 'right' at the start of sentences, they carry rhythm."
"Remove false starts and repeated sentences, keeping the last complete attempt."
"Tighten breathing room to about 300ms between sentences, but do not cut into laughs or overlapping speech."
Strong prompts for restructuring:
"Move the segment about the funding round to right after the intro. Keep the rest in order."
"Reorder the topics so the most concrete story comes first and the philosophical part comes last."
"This runs 18 minutes and I need 12. Cut from the middle sections, not the open or close, and tell me what you removed."
"Remove every moment where he repeats a point he already made, keep the clearest version of each."
Why these work: they give a threshold and an exception. The exception is the part that makes a cut feel human. "Remove silence" gets you a machine-gun edit. "Remove silence except after questions" gets you an edit that breathes.
Multi-cam camera switching
Multi-cam camera switching runs by default, so most of the time you do not need to prompt for it. When your project has multiple camera angles, Selects cuts to whoever is speaking, avoids jump cuts, and mixes in wide shots at natural moments.
Prompt for camera switching when you want something different from that default. This is the capability that separates chat-based prep from transcript-only tools, because Selects can see who is speaking, who is reacting, and what is on screen, so your camera direction can be editorial rather than mechanical.
Weak prompts: "switch cameras" and "cut between cameras every few seconds". The first is already the default, so it adds nothing. The second replaces a natural rhythm with a mechanical one.
Strong prompts:
"Hold on the listener for a beat when they react visibly."
"Stay wide during the setup and go to singles once the conversation gets going."
"Use the wide whenever both are laughing or talking over each other. Never cut to a camera where the speaker is out of frame or looking away."
"Cut to the overhead camera every time she picks up an object on the table, and hold until she puts it down."
"Minimum four seconds on any angle. No cut within one second of a sentence ending, cut on the breath instead."
"During the Q and A, cut to the audience camera only when a question is being asked, then back to the stage."
"Cam B is out of focus after the 40-minute mark. Do not use it from there on."
Why these work: they encode the rules a human editor holds in their head but never writes down. Minimum shot duration, cut on the breath not on the word, hold on reactions, never cut to a bad frame. Give Selects those rules explicitly and the switching becomes yours rather than generic.
To check speaker-to-camera mapping and audio tracks before you start, see Multi-cam Camera Switching Settings: Configure Speakers, Cameras, and Audio.
What not to prompt for
Selects is a prep and assembly tool, so there are things not to prompt for. It builds the structure and hands off. It does not, and should not, do the finish.
Do not prompt for:
Transitions, whooshes, zoom punches
Color grading, LUTs, looks
Titles, lower thirds, captions styling
Music selection, sound design, mixing
Motion graphics or effects of any kind
This is a deliberate boundary, not a missing feature. The prep work (sync, organization, selection, structure) is where the hours go and where automation actually compounds. The finish is where your taste lives and where a client can tell the difference between you and someone else. Handing the first part to a tool so you have time for the second is the entire point.
If you find yourself typing "make it look cinematic", you are asking the wrong stage of the pipeline.
Prompts by content type
Different formats fail in different places, so here is where to aim your prompts for each content type.
Talk show and podcast
For a talk show or podcast, the hard part is conversational rhythm and camera logic across a long, low-event recording.
"Build a 45-minute episode from this two-hour recording. Three cameras: host single, guest single, two-shot. Hold on the listener when they react, use the two-shot for crosstalk and laughs. Remove silences over one second and false starts. Keep all tangents that get a laugh even if they are off topic. Open with the strongest 30 seconds from anywhere in the conversation as a cold open, then go chronological."
Additional passes worth running:
"Find every moment that would work as a standalone clip for social. I want ones that make sense without context, 30 to 60 seconds each."
"The middle 20 minutes drag. Show me what I can cut without breaking the thread."
Vlog
For a vlog, the hard part is that the good material is scattered across a lot of nothing, and the structure has to be built rather than found.
"Organize this by location and time of day first. Then propose a 10-minute structure: arrival, the main activity, and a reflective close. Cut anything where I am walking without talking unless the shot is genuinely beautiful, and tell me which ones you kept and why."
"Find every piece of usable B-roll and group it by subject so I can place it myself."
"Cut my pieces to camera down to the takes where I do not restart, keep the last good attempt each time."
Events and conferences
For events and conferences, the hard part is volume, multi-cam with an audience, and knowing which parts of a long day matter.
"This is a full day of stage recordings. Split by session using the slate and the speaker change. For each session build a stringout with the intro trimmed off and any dead time before the speaker starts. Stage camera as primary, cut to the audience camera only during questions and applause. Flag any session where the stage audio drops."
"Find every moment the audience laughed or applauded, with 10 seconds of lead-in. That is my highlight reel material."
Lectures and educational content
For lectures and educational content, the hard part is structure and density. The content is good, the pacing is not.
"Chapter this two-hour lecture by topic with timecodes and titles. Remove long silences while he writes on the board, but keep the board visible in a shorter form. Cut to the slide camera whenever he references a slide, back to the wide when he steps away. Remove the administrative section at the start about the syllabus."
"Find every place he repeats a definition and keep only the clearest one."
"Propose a 15-minute version that keeps the three core concepts and drops the examples."
Interviews and documentary
For interviews and documentary, the hard part is protecting meaning while cutting hard.
"Build a stringout organized by question topic, not chronology. Never trim inside a sentence and never cut a qualifier off the end of a statement. Keep the pauses before her answers, they are doing work. Flag any answer where she contradicts something she said earlier."
Common prompt mistakes
These common prompt mistakes come up again and again. Here is why each one fails and what to write instead.
Prompt | Why it fails | Write this instead |
"Make it engaging" | No definition, no constraint | "Cut anything with no new information. Target 12 minutes." |
"Cut the boring parts" | Boring is not in the transcript | "Cut sections that repeat an earlier point or run over 90 seconds on one idea." |
"Remove all silence" | Produces an airless edit | "Remove silences over 0.8s, keep pauses after questions." |
"Switch cameras dynamically" | Camera switching already runs by default | "Use the wide more often, and hold on the listener when they react." |
"Make it look professional" | Finishing request in a prep tool | Handle in your NLE. Prompt for structure instead. |
"Do everything" | Nothing to review, nothing to correct | Run it in passes. See below. |
Work in passes, not in one prompt
Working in passes beats one giant prompt. The best sessions look like a conversation, not a command. The order that works:
Organize. Get bins, names, and grouping right.
Explore. Ask what is in there, where the strong material is, what the options are.
Structure. Commit to a storyline and ask for the stringout in that order.
Tighten. Silence, fillers, false starts, length target.
Camera. Only if you want something different from the default switching.
Handoff. Export to Premiere, Final Cut Pro, or DaVinci Resolve and finish it yourself.
Doing it in that order means each prompt operates on something already correct. Doing it in one giant prompt means when the result is wrong you cannot tell which instruction caused it.
And correct it in place. "Too tight, add half a second of handle to every cut" is a normal thing to say, and it is faster than reaching for the trim tool.
FAQs
Do I need to write long prompts?
You do not need long prompts, you need decided ones. "12 minutes, cut setup chatter, keep questions in full" is short and complete. "Please carefully create an engaging and dynamic edit" is long and empty.
Can Selects find a clip if I only remember what it looked like?
Yes, Selects can find a clip from a visual memory alone. It analyzes the video as well as the transcript, so visual descriptions work: what someone was holding, where they were looking, what was on screen behind them.
Do I need to prompt for camera switching?
You do not need to prompt for camera switching. Selects cuts to whoever is speaking, avoids jump cuts, and mixes in wide shots at natural moments. Prompt only when you want something different, like featuring one angle more or skipping a camera that went out of focus.
Can Selects add transitions or color?
No, Selects cannot add transitions or color, by design. Selects handles prep and assembly and hands off a structured timeline to your NLE. Finishing stays with you.
What formats does this work best for?
Prompt-based editing works best for long-form and multi-cam projects: interviews, podcasts, talk shows, vlogs, events, lectures, scripted video. Anything where the prep work is a large share of the total time.
What if the first cut is wrong?
If the first cut is wrong, say what is wrong and re-prompt that one thing. It is a conversation, not a render. Correcting one pass is much faster than restating the whole brief.