I hear the same story from experts constantly: they sign up for an AI coaching platform based on a sleek demo video, then try to actually upload their content and hit a wall. A YouTube channel the platform can't touch. A Notion vault with years of frameworks it won't read. A podcast archive it can only accept one file at a time, by hand.

By the time most experts discover the gap between the demo and the real ingestion workflow, they've already paid for the first month. I've watched this happen over and over, and it's almost never because the expert didn't have enough content. It's because the platform couldn't actually read the content they already had.

Why the "what can I upload" question matters more than most comparison articles admit

Every AI coaching platform promises to "learn from your content." But the engineering reality underneath that promise varies wildly. Some platforms transcribe your YouTube channel automatically and keep it synced. Others require you to download videos, transcribe them yourself, and upload text files. Some crawl your entire website. Others let you paste one URL at a time.

This isn't about features for the sake of features. It's about whether you can actually use the expertise you already have. If your knowledge base lives in Notion and the platform has no Notion connector, you're looking at weeks of manual copy-paste work. If your best teaching happens on video and the platform only accepts text, you're starting from scratch instead of from your archive.

The gap between marketing language and technical documentation makes this worse. Some platforms advertise "upload videos" when their own help docs clarify they only accept YouTube links. Some advertise "unlimited documents" in the sales copy while the support pages quietly list a hard file cap.

Here's what each platform actually accepts, pulled directly from each one's own current documentation:

Platform Document Formats Video Ingestion Audio/Podcast Integrations Website Scraping Notable Limits
Delphi PDF, DOC, DOCX, TXT, other common formats Direct upload (5GB max), YouTube videos + channels as updating feeds Direct upload, podcast episodes + series as feeds Google Drive, Notion, Granola, Obsidian Single URL or connected site feed (public pages only) No public total file limit documented
Coachvox PDF, DOCX, TXT only (plus CSV for Q&A pairs) None - videos must be transcribed to text first None - audio must be transcribed to text first None documented Single public page only, no crawling 25MB/file, 100K characters/file, free trial: 30 sources (5 URLs), paid: 100 uploads default
Personify PDF, DOCX shown Direct MP4 upload, YouTube channel URLs Audio/video transcription (Free: 15 min/month, Pro: 240 min/month) None documented URL scraping supported, crawl depth unclear Free: 10 sources/100K characters. Pro's own support docs list 500 sources/10M characters, despite "unlimited" marketing language
Rocky.ai PDF, pasted text (DOCX unclear) YouTube/Vimeo/MP4 streaming links only - no direct upload, transcription behavior unclear Not documented None documented None documented Video must be linked, not uploaded directly
CustomGPT.ai PDF, DOCX, PPTX, XLSX, CSV, Markdown, 1,400+ formats advertised YouTube video/playlist/channel with auto-sync (no direct MP4 upload documented) Direct audio file upload Google Drive, Dropbox, Box, SharePoint, Notion, Confluence, GitHub, more Full crawl or sitemap 100MB/file, 50 files/request, 1GB/batch, 5,000-20,000 documents per agent depending on plan
Pickaxe PDF, DOC/DOCX, TXT, Markdown, PPTX, XLSX, XML, HTML, YAML, CSV (with OCR for scanned PDFs) Direct MP4 upload, YouTube video/playlist/channel with auto-sync Direct MP3/WAV upload, RSS feed support Notion, Google, OneNote, OneDrive/SharePoint Single page or full-site crawl with optional daily auto-sync Free tier: 3 files, individual file limit around 99-100MB, Gold-tier quantity limits not clearly published
BuddyPro PDF, DOCX, TXT, native Google Docs, images (text-in-image transcription only) Direct MP4 upload (audio track transcribed, roughly 2-hour max recommended per video), YouTube channels/playlists, Vimeo folders Direct MP3 upload Google Drive folder structure (no native Notion connector) Specific public page URLs added to a designated doc (no automatic full-site crawl) No per-file size ceiling documented; file and document names must be unique within their folder trees

The content gaps the marketing pages don't mention

Two patterns show up across multiple platforms, and they matter more than most experts realize until after they've already committed.

First, marketing language doesn't always match technical documentation. Rocky.ai's site talks about uploading videos, but the actual technical docs clarify that direct video upload isn't supported. You can only provide YouTube, Vimeo, or streaming MP4 links, and the documentation doesn't clearly confirm whether that linked content is transcribed for AI knowledge or simply attached as media. Personify's Pro tier is marketed as offering "unlimited" documents, but its own knowledge-limits support page states a cap of 500 sources and 10 million characters. This isn't about catching any platform in a lie. It's about the gap between sales copy written for clarity and technical limits that only show up once you read the help docs.

Second, when a platform says it "learns from video," that almost always means it transcribes the audio track, not the picture. The visual content - charts, diagrams, on-screen demonstrations, anything you draw or point to - is invisible to the AI unless you describe it out loud. This applies to BuddyPro, Personify, Delphi, and CustomGPT.ai alike. None of them document true visual understanding or scene analysis. They process video by extracting the spoken words from it.

If your teaching style relies heavily on visual explanation, that's a real limitation across the entire category, not a weakness of any single platform. The AI won't "see" your whiteboard session or your slide deck unless you narrate what's on the screen.

Coachvox takes a different approach entirely. It's built as a text-to-training-pairs system, not a multimedia ingestion platform. No native video ingestion, no audio ingestion, no YouTube connector, no podcast feed. Everything has to be transcribed to text first, then uploaded as PDF, DOCX, or TXT. The documentation is unusually clear and consistent about this limit. If you want to use video or audio content on Coachvox, you handle the transcription yourself before you upload anything.

Where BuddyPro sits on this specific dimension

On BuddyPro, the knowledge ingestion workflow is built around a connected Google Drive folder structure - Files, Texts, and URLs subfolders - rather than a general-purpose connector to every possible cloud tool. That makes it genuinely strong for audio and video-heavy expert content: direct MP4 upload, whole YouTube channels and playlists, Vimeo folder transcription, direct MP3 files, and native Google Docs all work without needing a third-party transcription step first.

It's also a real limitation if your knowledge base lives primarily in Notion, or if you want a platform to automatically crawl your entire website. Website ingestion on BuddyPro works by adding specific public page URLs to a designated document, not by pointing the system at your domain and letting it discover everything on its own.

I'll say this plainly rather than spin it: that's a tradeoff, not a hidden strength. BuddyPro is built for experts who want a premium, relationship-first AI digital twin they can sell as a standalone product for $1,000 to $2,000 a year, and the Google Drive-centric ingestion workflow fits that model well, since most established experts already have their core content as video files, audio recordings, PDFs, and Google Docs. But if most of what you know lives in Notion pages or across a sprawling website, the setup takes more manual work on BuddyPro than it would on a platform like CustomGPT.ai or Pickaxe, which were built with broader connector support from the start.

What this means for your actual decision

Before you pick a platform, audit what you actually have. Not what you plan to create. What already exists. Open your Google Drive, your Notion workspace, your YouTube channel, your podcast host. List the formats. Get a real sense of how much is there.

If most of your expertise sits in Notion across hundreds of documents, a platform with no Notion connector means weeks of manual export and upload. If you have a large YouTube archive and the platform requires downloading and transcribing each video by hand, that's a project that realistically never gets finished. If your teaching relies on visual diagrams, the AI will miss half of what you're explaining no matter which platform you pick, since none of them currently understand video beyond the audio track.

Audio and video archives tend to be an expert's biggest hidden asset. Most coaches and consultants already have a real backlog of recorded content sitting around: client calls they saved, webinars they ran, old course videos, podcast interviews. A platform that can ingest that content directly and transcribe it automatically lets you go from a pile of old recordings to a trained AI twin in days. A platform that can't leaves you stuck transcribing everything by hand before you've even started.

CustomGPT.ai and Pickaxe currently have the broadest documented ingestion flexibility: the widest range of file formats, the most cloud integrations, and full-site crawling. The tradeoff is that neither is purpose-built for the relationship-first AI coaching model - they're general-purpose AI agent builders that happen to work for this use case too. Delphi and BuddyPro are more opinionated, built specifically around the idea of an AI version of an expert, and their ingestion workflows reflect that: strong support for the formats experts actually generate - video, audio, documents - with less flexibility for edge cases outside that. Coachvox is the most text-focused of the group, which is the wrong fit if you want to lean on existing multimedia, but the right fit if you'd rather keep full control over the exact training data as Q&A pairs.

The real lesson here: the "what can I upload" question isn't a technical footnote you check after you've already picked a platform. It's the difference between launching in days and quietly abandoning the project three months later because the manual transcription work never ends. Read the documentation, not just the demo. Make sure the platform you choose actually ingests the content you already have, whether you build it yourself on a platform like BuddyPro or anywhere else on this list. And once it's live, remember that keeping that knowledge current is a separate, ongoing question worth planning for from day one.

Related Articles

Sources (checked October 16, 2026): Delphi help center (help.delphi.co), Coachvox support documentation (support.coachvox.ai), Personify knowledge-limits support page (personify.fyi/support/knowledge), Rocky.ai creator help center (help.rocky.ai), CustomGPT.ai developer documentation (docs.customgpt.ai), Pickaxe knowledge base documentation (pickaxe.co/learn/knowledge-base), and BuddyPro's own training documentation (docs.buddypro.ai/training/upload-know-how).

If you want to talk more about training an AI coaching clone on your existing content, feel free to catch me on LinkedIn or wherever I'm at in the world at the moment you're reading this, which is usually San Francisco, Prague or Bali.

David Riha · AI Digital Twin Builder · October 16, 2026

Share this article