Leela with an AI guide: how the full history of moves is used

An AI guide is useful not because it “knows the answer of the cell,” but because it can hold more conversational elements than a static description. It relates the initial intention, the current move, and what the player has already written. The more precise those inputs are, the less likely the result is to be a universal paragraph that could fit any situation.

Which context can shape the response

In a normal game flow, the AI works only with what is available in the session: the intention, the cell numbers and themes, the order of moves, user responses, and interpretive instructions. It does not automatically access participants’ thoughts, private messages, bank records, or events the user has not described.

This is why it helps to distinguish AI interpretation from a template response. A template repeats the same cell meaning. Contextual analysis tries to connect that meaning with the question and earlier notes, but the connection remains a hypothesis.

Why a sequence matters more than one cell

A single symbol can be interpreted in many ways. A sequence reveals a different kind of information: whether themes of control, waiting, choice, boundaries, or responsibility recur; whether the player’s language changes; and whether later answers contradict the first assumption.

AI can point out a repetition, but it should not turn it into a diagnosis. “You linked three different cells with fear of making a mistake” is grounded in the session text. “You have a disorder, therefore you are afraid” goes beyond the available evidence.

How to write answers that produce a specific discussion

Include four elements:

• fact: what happened and can be checked;
• reaction: what you thought or felt;
• unknown: what information is missing;
• alternative: another explanation that may fit.

“This cell is about my job” contains almost no context. A better response is: “My manager postponed the role conversation twice. I read that as a refusal, but there has been no direct answer. Another possibility is that the decision has not been made.” Now the AI can ask about evidence and criteria rather than inventing the story.

A useful dialogue that keeps the decision with the player

Player: “I connect this cell with the project being stuck.”
AI: “Which observable fact shows that it is stuck, and which part is your inference?”
Player: “The deadline moved, but the budget was not cancelled. I assumed the team had lost interest.”
AI: “What other reasons could explain the delay, and which question can you ask the team?”

The value is not a statement that the project will fail or succeed. It is a clearer list of missing data. The page on how AI analyses Leela responses explores this process in more detail.

What AI cannot establish

A model does not make an assumption true by writing it fluently. It may:

• mix an observation with an interpretation;
• reinforce a false premise supplied by the user;
• miss irony or cultural context;
• sound more confident than the evidence allows;
• reproduce inaccurate information;
• fail to recognise the severity of a crisis.

Keep the right to reject a response, ask for alternatives, and compare the output with real data. The full boundary list is available in limits of the Leela AI guide.

Privacy without unnecessary detail

A useful answer rarely requires passport numbers, surnames, precise addresses, or full account details. Replace people with roles, remove details that do not change the decision, and do not paste confidential documents. Information about another person does not become your data merely because it appears in your question.

In a group session, each participant should decide which answers become shared. Private answers in Leela allow the move to remain visible while the personal content stays closed.

When a human is the better option

A human facilitator can notice silence, non-verbal distress, group pressure, and the moment when the discussion should stop. They can also adapt the pace beyond a text interface. Human expertise must still be evaluated: a facilitator should not diagnose or demand obedience to an interpretation.

AI is convenient for calm, self-directed, structured work. The choice is examined in Leela with or without a facilitator.

A better way to request a final summary

Ask the AI to separate the result into four blocks:

1. facts provided by the player;
2. recurring themes in the responses;
3. several possible hypotheses;
4. one low-risk action and a review criterion.

Then edit the summary in your own words. Do not preserve a claim you disagree with merely because it sounds persuasive. The Leela AI final summary guide explains how to review the result.

Open AI-guided Leela

Start in Leela by Reflecta. Write your own reaction to a cell first, then use AI for questions, alternatives, and note organisation. That order keeps authorship of the decision with the player.