How a Digital Mind learns from your documents, recordings and social posts

A Digital Mind learns from two places. Drive holds the documents you write, files and recordings you upload, and a FAQ. Social media syncs posts from connected Instagram, TikTok, YouTube, X and LinkedIn accounts. Each item shows Learning… and then Learned, after which the Mind can answer from it. Start with your FAQ and your current offer.

What does a Digital Mind learn from?

Knowledge has two areas, and they behave differently.

Facts live here; instructions don't. A rule such as "ask one question at a time" or one that sends refund requests to a named person belongs in the Playbook; how to write one is covered separately. For the bigger picture, see what a Digital Mind is.

Inside Drive, keep one topic per document. A document saves as you type and is learned straight away. The FAQ sits at the top: write a question and its answer, and a new row appears for the next. Documents and FAQ answers don't spend training credits; files, recordings and social posts can spend AI credits, with the cost shown before you confirm.

Which files can you upload, and how big can they be?

Drive accepts these as of October 2026:

The Mind reads the words in an image, such as a menu or a price list, and what the image shows. Plan-dependent caps on stored files and total recording length also apply.

For a single fact, write a document or a FAQ answer instead of uploading a file.

What do Learning… and Learned mean?

In Drive, an item shows Learning… while the Mind reads it and Learned when it has finished. Wait for Learned before you test it in the chat preview. If an item shows Failed, select Try again.

Posts from Social media show four statuses.

Selected posts are learned at the next daily sync, so expect a delay. The Mind keeps the first posts it learned; once your plan's capacity is full, newer posts wait for room, which moving older posts to Trash can free up.

How does retrieval ground an answer?

Grounding means the system looks up relevant passages from your material first, then writes the answer with those passages in front of it, instead of relying only on what the language model absorbed in training. The research name is retrieval-augmented generation (RAG). Lewis et al. (2020, NeurIPS) (opens in a new tab)^[1] paired a pre-trained language model with "a dense vector index of Wikipedia" searched by a neural retriever, and on language generation tasks reported "more specific, diverse and factual language" than a parametric-only baseline.

Swap Wikipedia for your Drive and posts and you have the idea. That simplifies a 2020 research system, and Plandora's documentation does not publish how its own retrieval is built, so read it as how the technique works, not as a specification. What the product does say: Search Mind knowledge stays on in the Playbook, the app describes replies as grounded in the content saved or synced to your Mind, and replies may show source citations that visitors can open.

Two consequences follow. A grounded answer is only as good as its passage, so a stale document can produce a confident wrong answer. And grounding narrows the room for error without removing it: a survey of hallucination in natural language generation (opens in a new tab)^[2] describes deep-learning generation as "prone to hallucinate unintended text", and the Lewis abstract listed providing provenance and updating knowledge as open research problems in 2020. Keep a person involved in financial, legal, medical and other consequential decisions, and see why AI replies go wrong.

Why should expired content stop being used?

Retrieval typically ranks passages by how well they fit the question, and a price list does not announce that it is last year's. If both lists sit in Drive, the Mind can draw on either.

Set dates… on a document or file limits its use to between two dates; Plandora's glossary calls this the validity period. Use it for anything with a natural end, such as a seasonal offer, an event schedule or a fixed-term policy. For a fact that changes often, keep one current document or FAQ answer and trash the rest.

What should you add first?

The order is a judgement, not a Plandora rule: repeated questions first, slow-to-prepare material later.

1. Your FAQ: the five questions you are asked most. Plandora's guidance is to start there and test them in the chat preview. Short, exact answers leave little to interpret, and repeated questions are where a gap shows first.
2. One document on what you offer. Who it is for, what is included, what is not, and how it runs from first conversation to delivery. If you sell advice or services, the solutions page for experts describes the scope and fit questions a Mind handles early; answer those.
3. The policies visitors ask about. Refunds, cancellations, rescheduling, delivery, eligibility. Add validity dates where they change, and put anything that needs a person in the Playbook's Boundaries.
4. Your own explanatory material. Guides, slide decks, handouts, a recorded talk. Your method and reasoning sit here, which a FAQ cannot hold. Files and recordings can spend AI credits, so add the ones that answer real questions, not the whole archive.
5. Social media, once the above is in place. Posts show how you talk, but they are informal, written for a moment and may disagree with today's offer. Added after the curated material, they supply voice instead of setting the facts.

What should you leave out?

Content you do not have the right to use. Other people's articles, a course whose licence does not cover this use, a competitor's brochure. As the US Copyright Office explains (opens in a new tab)^[3] (page accessed 22 October 2026), copyright under US law exists automatically once an original work is fixed, and the owner holds the exclusive right to reproduce it. Uploading someone else's work can involve making copies of it, which is why permission matters. You can still use another person's work with a licence, with permission, under an exception such as fair use, or where it is in the public domain; the US exceptions sit in sections 107 to 122 of the Copyright Act, and other countries differ. Plandora's FAQ puts the duty on the customer: you must hold the rights for everything you upload. Whether a specific upload qualifies is a question for a qualified lawyer in your jurisdiction.

Private client data. Client files, intake forms, session notes, chat exports. Plandora's FAQ advises against uploading confidential, regulated or sensitive personal information unless your agreement and workspace setup expressly support it. Data protection law points the same way: the UK ICO's data minimisation guidance (opens in a new tab)^[4] quotes Article 5(1)(c) of the UK GDPR, which requires personal data to be "adequate, relevant and limited to what is necessary" for the purposes of processing. That page, accessed on 22 October 2026, notes it is under review after the Data (Use and Access) Act. Other jurisdictions have their own rules. Get qualified review before you upload anything personal.

Outdated prices and superseded terms. Date them or trash them, as above.

Anything you would not show a visitor. Replies draw on whatever the Mind has learned and may show source citations that visitors can open, so keep internal notes and unfinished drafts out of Drive.

How can you tell the Mind lacks knowledge, and what does it do?

1. Test before you share. Ask realistic questions in the chat preview, including ones you expect it to fail. Preview replies use AI credits.
2. Read early conversations. In Audience › Inbox, fix missing facts in Knowledge and unclear instructions in the Playbook.
3. Open Insights. The Conversations report shows top questions, answer coverage and knowledge in use. For a No answer or Partial answer, Add an answer opens the FAQ with the question filled in. "Answered" is a system assessment, not proof the visitor was satisfied, and answer analysis needs a paid plan.

What the Mind does when knowledge is missing depends on what you have told it. Plandora's documentation does not promise one fixed reply, so do not assume it will say it does not know. The public FAQ describes a Mind as designed to recognise when it should not answer and to hand off, and says owners can shape fallback messages. Set the rule yourself: in Boundaries, name the topics that need a person and who takes them; a handoff flags the conversation in Inbox, where you can Take over. Search web is opt-in, for when current public information helps, and extra usage charges may apply.

What happens to what you upload?

• Ownership. Per Plandora's FAQ, customers keep ownership of their original content and give Plandora limited permissions to host, process, transmit and display it to run the service.
• Model training. Knowledge you add is processed to operate your Mind and produce grounded responses. The public terms and privacy policy do not make a blanket promise covering every underlying AI provider, model or enterprise configuration; for a contractual commitment on training, isolation, retention or residency, ask Plandora's sales team before you upload.
• Removal. You control which sources are active. Removing one changes what the Mind can use, but separate conversation, audience, connector, backup or legal-retention records may follow different retention rules.

The data and privacy answers on Plandora's FAQs go into more detail.