Chat unsticks a moment; a tutor rebuilds a model. Choose opportunistic help for local fog and deliberate instruction for weak schemas — so transcripts stop posing as learning.

When confusion hits, the modern reflex is instant: open a chat, paste the problem, receive a polished paragraph. Sometimes that is exactly right. Often it is the wrong instrument — like using a firehose on a houseplant.

Learning design distinguishes opportunistic help (return to the task quickly) from deliberate instruction (diagnose, sequence, scaffold, probe). John Sweller’s cognitive load theory and Lev Vygotsky’s zone of proximal development (ZPD) explain why the choice matters: one tool cuts extraneous friction; the other invests germane work inside a fade plan.

Sukratic Chat and Sukrat Tutor map onto those jobs. Choosing wrong wastes time and hollows out encoding. The goal is not a smarter transcript. The goal is a mind that can perform when the transcript is gone.

Two different jobs

When chat is the right call

Use opportunistic help when you were already in a guide, lecture, or mock question; the stuck spot is local; you can attempt first and climb hints; you will reconstruct after; and you can name the single stuck step.

Stay on the surface. Context-scoped help beats a new-tab odyssey — task-switching costs are real, and switching tools mid-fog often adds more extraneous load than it removes.

The point of chat is continuity: keep working memory on the original task while a narrow scaffold clears friction. Hint ladders belong here; full dumps usually do not. If you cannot name the stuck step, you may already be outside chat’s best use case.

When you need a tutor loop instead

Escalate when the same concept fails across surfaces; teach-back collapses after multiple explanations; prerequisites feel missing (every explanation assumes something you cannot produce); you are collecting chats instead of producing answers; or the topic feels like fog at the system level, not one sentence.

A tutor-style loop is probe → find gap → teach into gap → probe again. That is Vygotsky’s ZPD with fade, not an answer vending machine.

Contingent tutoring research (Chi, Roscoe, Graesser and others) keeps finding the same pattern: explanations help most when they follow the learner’s attempt, not when they replace it. Scaffolding is temporary by definition — if the same dump is still required three weeks later, you outsourced rather than taught.

A 60-second decision rule

1. Can I name the single stuck step? → chat hint
2. Can I teach yesterday’s version of this topic? → if no, tutor/guide encode
3. Have I failed this idea twice after honest attempts? → escalate
4. Is the real need timed production? → paper, not more explanation

Write the answer down. Decision rules beat vibes mid-panic. The rule is not “more help”; it is “right scope.”

If step 1 fails and step 2 fails, stop collecting opportunistic dumps — you need deliberate instruction. Tape a pocket card near your screen: local → chat; systemic → tutor; timed → paper. Decision latency is where wrong tools win.

How the products complement each other

Use Chat beside the work to keep flow. Use Tutor when the work needs restructuring. Use Guide for clarity, Notes for encoding, Cards for retrieval, Mocks for transfer. Sequencing beats tool soup.

Wrong sequence: chat dump → feel clearer → skip teach-back → fail paper → more chat.

Right sequence: diagnose → encode → retrieve → transfer, with chat only for local fog. Tools are stages in a loop, not competing brands for the same stuck feeling.

What neither tool replaces

Neither chat nor tutor replaces generative notes and teach-back, spaced retrieval, mixed questions and papers, or sleep and calendar honesty.

Help is a scaffold. Performance under novel cues is the building. A week of eloquent threads with no teach-back and no timed production is a fluency costume, not a study plan.

Robert Bjork’s desirable difficulties still apply after the help session: struggle that improves storage is not optional decoration. If help deletes struggle entirely, you may feel successful while storage strength barely moves.

Case patterns (choose in under a minute)

Pattern A: Stuck on one mock step, rest of topic teach-back works → chat hint ladder → reconstruct → continue paper.

Pattern B: Every explanation of topic T assumes moves you cannot produce → tutor/prerequisite audit → encode → then return to T.

Pattern C: You understand in chat, blank on timed section → stop collecting explanations; schedule timed production and mixed items.

Pattern D: Long thread, feeling smarter, still cannot teach-back → close chat; teach-back or switch to guide/tutor.

Write the pattern letter in your notes. Naming the pattern beats vibes. Pattern recognition is how you stop reopening chat by habit.

Avoiding tool soup

Opening five apps for one stuck moment adds extraneous load. Prefer: stay on the surface → chat if local → escalate once if systemic → reconstruct → schedule.

Tool switching is not the same as learning sequencing. One evening, three instruments, one goal: a mind that can perform without the thread. Prefer fewer tools with a fade plan over more tools with more transcripts.

The transcript test

After any help session — chat or tutor — close the thread and teach the idea for sixty seconds. If you cannot, you are not done.

The transcript test catches the most expensive illusion in AI-era studying: eloquence nearby mistaken for knowledge inside. Pair the test with a calendar note when the gap was prerequisite-shaped.

Opportunistic help ends when the step is clear; deliberate instruction ends when production works without the transcript.

Sequencing help across a single evening

Start on the surface. Local fog — chat ladder. If teach-back of the broader topic fails, stop the chat thread and switch to tutor/guide encode.

After encode, return to the original item and produce. End with a schedule note if the blocker sits earlier in the syllabus. That evening is a decision rule in motion, not a mood-driven app tour.

FAQ

Can chat diagnose?
Lightly. Deep diagnosis with prerequisite sequencing is a tutor job.

What if I only have chat?
Impose tutor discipline: attempt, hints only, teach-back, stop after two failures and switch to a clear source.

Is asking more questions always better?
More questions with no production is more fluency. Production is the test.

Should I use tutor for every stuck moment?
No — that is overkill and breaks flow. Local stuck → chat ladder. Systemic fog → tutor/guide.

How do I know I escalated too late?
If you have three long threads and still cannot teach-back the core claim, you stayed opportunistic when you needed deliberate instruction.

Related reading: Sukrat Tutor + Chat

• Diagnose Before You Study
• Ask Without Outsourcing Your Thinking
• Hints Before Answers
• Scaffolding Without Giving Answers
• Stuck Moments
• Context Switching Kills Study Focus
• Socratic Questions

Keep going across Sukrat

• Clarity Before Practice
• Checkpoints Beat Passive Watching
• Sequencing Resources in a Study Session