📋 Available on: Growth (up to 3 fields), Enterprise (up to 10+), Agency, and Custom plans. Not available on Startup or Free plans.
AI Filters (formerly Magic Fields) automatically read, watch, and understand your user-generated content, then tag each post with the classifications your workspace cares about — so you can filter, report, and surface content without tagging it by hand. This article explains how that happens behind the scenes and what to expect day to day. For how to set them up and use them, see Using AI Filters to Organize Your Content.
How Classification Works
Each AI Filter is an AI model with a specific job — for example, "which product is featured?", "is this positive or negative?", or "is this safe for paid ads?". When a post is analyzed, the model looks at the whole post, not just the caption:
Visuals — what's on screen, including products, logos, and scenes (photos and video frames).
Audio — what's said in a video, via its transcript.
Text — the caption and hashtags.
The model returns a value (or values) and Archive saves it on the post as a filterable attribute, exactly like a tag. Each AI Filter runs independently, so a single post can be classified by several filters at once.
When AI Filters Run
Automatically on new content. As new UGC arrives in your workspace, AI Filters run on it without any action from you.
Retroactively, on request. Existing content isn't reprocessed automatically. To apply a new or updated filter to content already in your workspace, ask your Customer Success Manager — it's a deliberate, one-time backfill.
After transcription, when needed. Filters that rely on what's said in a video wait until that video's transcript is ready before they run, so the audio is included in the analysis.
How Long It Takes
Processing is automatic and happens in the background. For a single new post, a filter typically completes within a few minutes of the post being captured. Two things can make it take longer:
Transcript-dependent filters only run once the video's transcript finishes, so they trail the post's arrival.
Large retroactive backfills (applying a filter to thousands of existing posts) are processed in batches and can take longer to fully complete.
Standard and Pro Filters
AI Filters come in two quality tiers, depending on how hard the classification is:
Standard — for most classifications. 1 credit per post, per field.
Pro — a more capable model for harder or more nuanced tasks. 2 credits per post, per field.
Your Customer Success Manager sets the tier when the filter is built, based on what it needs to do.
How Accurate It Is
AI Filters are a strong, scalable estimate — not a guaranteed ground truth. Treat them as a directional signal, and keep these in mind:
Give feedback. Each result in the post's detail view has thumbs-up / thumbs-down buttons. Flagging correct and incorrect results helps improve accuracy over time.
Locked option lists. A filter can be locked to a fixed set of values (e.g. your official product catalog) so it only ever returns approved options. Ask your CSM to lock or unlock a field.
Better input, better output. Posts with clear visuals, captions, and (for video) a transcript classify more reliably than thin or ambiguous ones.
What to Expect
A small share of posts may not get a value. Occasionally a post can't be analyzed — for example, the media can't be processed or a video's transcript isn't available. Those posts are simply left blank for that filter rather than guessed.
Reprocessing costs credits again. Re-running a filter on a post (for example, after changing the field) counts as a new run and consumes credits again.
New filters don't change old content on their own. A newly added filter only runs on content going forward unless you request a retroactive backfill.
⚠️ If your workspace runs out of credits, AI Filters pause for new content and those posts stay unclassified until credits are replenished. After topping up, ask your CSM to re-run the affected content — it doesn't reprocess automatically.
Credits
Each filter run draws from your workspace's monthly credit allowance — 1 credit for Standard, 2 for Pro, per post per field. You can track usage under Settings → Credit Usage. For full details, see Understanding Your Archive Credits & Billing.
Questions about setting up or fine-tuning an AI Filter? Reach out to your Customer Success Manager, or contact Archive Support via Intercom chat or at [email protected].