Top Ways Tutors Can Use AI to Enhance Their Sessions (From Someone Who Runs a Tutoring Company)

How can tutors use AI to improve their tutoring sessions?

The honest version: AI doesn’t replace the judgement that makes you a good tutor — it removes the work that stops you from using it. Used well, it does three things in a session: it tells you exactly where a student is stuck before you sit down, it marks and explains the routine work so your time goes to the hard parts, and it turns what happened in the session into evidence a parent can actually see. The trap is the opposite: pasting a maths question into a general chatbot and trusting the answer. General models hallucinate on maths and don’t know your curriculum. So the rule that runs through everything below is simple — let AI do the volume, you keep the judgement.

I run a tutoring company. Here’s how AI earns its place in our sessions, in the order it actually matters — including the exact workflow I run, the tools I use, and roughly what it costs me in time.

1. Walk into every session knowing where the student dropped marks

The single biggest upgrade isn’t in the session — it’s the ten minutes before it. Instead of “so… how did the homework go?”, you already know: this student lost three marks on the same probability step twice this week, and nailed everything else.

Here’s the exact workflow I run. I text the student and ask them to send me whatever they’ve had marked — recent tests, results, and a photo of every question they got stuck on — and I drop it into one document. Then I hand that to an AI assistant (I use Claude Code because I’m on a paid plan; NotebookLM does the job too) and, dictating through Wispr Flow so it’s fast, I tell it: this student is in Year X, here’s their test and the questions they’re struggling with — pull out their areas of weakness, write two practice questions for each one they got wrong, list the weak areas by topic, and put it all into Obsidian as three notes: one listing the weaknesses, one with the matching practice questions, and one with the solutions.

Ten minutes, and I walk into the session already knowing exactly where the marks are leaking — with targeted practice already built. That’s the whole difference between a tutor who prepares and one who opens with “so… how did the homework go?”

Here’s what that ten-minute pass actually produces — a real Year 10 student of mine (name changed to Daniel), built from a Number & Algebra test he scored 35% on. Three notes come out of it:

① Weaknesses — the diagnosis

• Literal / abstract algebra (the biggest gap). Every question with letters for coefficients was left blank or wrong — yet he can do the numeric twin of each. The block is purely symbolic.
• Algebraic fractions + index laws — expanding brackets, dividing fractions (× reciprocal), and $x^m \div x^n = x^{m-n}$.
• Multi-step questions left blank — “don’t know where to start” on unfamiliar shapes.
• Sign & arithmetic slips under pressure — the method was right, the execution wasn’t.
• Graphing inequalities / regions of the plane — not attempted at all.

② Practice — two questions per weak skill, the numeric twin first, then the same move with letters

• (i) Solve $\dfrac{x}{8}+\dfrac{x}{12}=1$. (ii) Now solve $\dfrac{x}{4a}+\dfrac{x}{6b}=1$ for $x$.
• Expand and simplify $15-3(a^2b-2ab^2)+4(ab)^2$ — watch the bracket signs.

③ Solutions — so he can self-check

• $\dfrac{x}{8}+\dfrac{x}{12}=1 \Rightarrow \dfrac{3x+2x}{24}=1 \Rightarrow x=\dfrac{24}{5}$; the same move with letters gives $x=\dfrac{12ab}{2a+3b}$.
• $15-3(a^2b-2ab^2)+4(ab)^2 = 4a^2b^2-3a^2b+6ab^2+15$ — the fix he kept missing: $-3\times(-2ab^2)=+6ab^2$.

I’ll be honest about what that workflow actually is, though: it’s a manual rig of three separate tools — an AI assistant, a voice-typing app, and Obsidian — and it works because it’s me doing it. It does not scale to ten tutors each wiring up their own version. That’s exactly the gap ClassQuill is built to close: the same pre-session intelligence, auto-marked and consistent across every tutor, with nobody rigging up a tool stack.

2. Auto-mark the routine work so your time goes to the teaching

Marking is where tutor hours quietly disappear. A tutor who marks a set of practice questions by hand is spending teaching-grade time on data entry — and the feedback lands a week late, when the student has forgotten the question.

Auto-marking flips that: the student submits, it’s marked instantly, and the result feeds their progress picture without anyone typing it in. The feedback is immediate, which is when it actually changes behaviour. And the tutor’s session time goes to the two questions the student got wrong, not the eight they got right.

The caveat that keeps this honest: auto-marking is only as good as the question bank behind it. For structured, curriculum-aligned questions (the kind a VCE maths question bank is built from) it’s reliable. For free-form essay marking, AI is a drafting aid, not a verdict — you still read it.

3. Generate targeted practice at the student’s exact weakness — not generic worksheets

Once you can see the gap, the next move is practice aimed at it. The old way is hunting through a textbook for “more questions like that one.” AI shortens it to: here are six questions on this specific skill, at this level. (In my pre-session pass above, that’s the second of the three Obsidian notes — two fresh practice questions for every question they got wrong.)

The value isn’t volume — anyone can generate a thousand questions. It’s aim. Practice that targets the misconception the student keeps hitting beats a generic worksheet every time, and it’s the difference between a student plateauing and a student moving.

4. Turn a messy worked solution into an explanation the student can revise from

Tutors explain things brilliantly out loud and then the student goes home with nothing to revise from. AI is genuinely good at turning a worked solution into a clean, step-by-step explanation the student keeps — if you give it the correct working and have it format, not solve.

This is the honest line on general AI: it’s a strong explainer and a weak solver. Hand it the right answer and ask it to explain the steps clearly, and it’s excellent. Ask it to do the maths and trust the output, and it will confidently hand a student a wrong method. The tutor stays the source of truth; AI is the formatter.

Here’s how that plays out after a session for me. I talk — again through Wispr Flow — through what the student did well and where they came unstuck, and have the AI turn it into something they can actually revise from. One of mine answers fast but never shows his working, so on trigonometry — solving for the hypotenuse or an adjacent side — he’d freeze. I described exactly that to the AI and had it build him a fixed process to follow every time:

1. Label every side of the triangle
2. Write out the formula
3. Write down what you know and what you’re solving for
4. Rearrange
5. Solve

The AI produces that in seconds. I export the Obsidian note to PDF and drop it into the student’s Google Drive folder, so the next-step plan is waiting for them. (The Obsidian → PDF → Drive part is my own manual flow, not something the product does — same caveat as the pre-session rig in #1.)

Here’s a real one of those fixed-process notes (anonymised) — same student, for perpendicular lines rather than trigonometry, but the idea is identical: a fixed set of steps he follows every time so he never freezes on where to start.

Perpendicular lines — the process, every time

Key fact: perpendicular gradients are negative reciprocals — “flip the fraction, change the sign,” so $m_1 \times m_2 = -1$.

1. Rearrange the line to $y = mx + c$ and read off $m_1$.
2. Flip and change the sign → $m_2 = -\dfrac{1}{m_1}$.
3. Write $y - y_1 = m_2(x - x_1)$ using the given point.
4. Expand and tidy into the form the question asks for.
5. Check the point fits.

Worked example: perpendicular to $y = 2x + 9$ through $(2, -4)$ → $m_2 = -\tfrac{1}{2}$, so $y = -\tfrac{1}{2}x - 3$ (check: $-\tfrac{1}{2}(2) - 3 = -4$ ✓).

5. Let the parent update write itself

Parents don’t renew because sessions ran on time. They renew because they can see their child improving. But the thing that proves it — a clear, regular progress update — is exactly the thing tutors don’t have time to write after a long day.

AI lesson summaries close that gap: the post-session update is generated from what actually happened — topics covered, where the student is strong, where they’re slipping — so the parent gets a professional update and the tutor doesn’t lose an evening writing it. The owner outcome here is blunt: this is retention. The companies that show evidence keep clients through the moments when budgets get tight.

And the time cost is genuinely small. Per student I spend about ten minutes before the session and five after — fifteen minutes all up — and that covers the lot: the pre-session diagnosis, the targeted practice, the after-session plan, and the parent-facing update. For fifteen minutes, the jump in how prepared the session feels and how visible the progress is to the parent is the best-value quarter-hour in my week. It’s also how you drive repeat business: satisfaction climbs, and parents can see what they’re paying for.

6. Give the student an AI tutor for the gap between sessions

Most learning loss happens at 9pm on a Tuesday, when the student hits a wall on a question, can’t reach their tutor, and gives up. A curriculum-grounded AI tutor keeps the momentum: the student asks, gets an answer aligned to what they’re actually studying, and keeps going.

The word that matters is grounded. A general chatbot will answer — and sometimes invent. An AI tutor tied to the curriculum and the student’s own course is a different thing: it’s the safety net between sessions, not a replacement for them. Fewer “I got stuck and stopped” gaps means fewer stalled students, which means fewer cancellations.

7. Diagnose the misconception, not just the wrong answer

A mark tells you a student got it wrong. It doesn’t tell you why. The high-value use of AI is spotting the pattern across many attempts — this student isn’t bad at probability, they’re consistently misreading “without replacement” — and putting that in front of the tutor.

That’s weakness detection, and it’s the part no spreadsheet does. For a single tutor it’s a nice-to-have. For an owner with ten tutors, it’s quality control: you can finally see whether a cohort is struggling with a topic because it’s hard or because it was taught poorly.

I’ll be straight: I haven’t hit this one myself yet — solo, you don’t have a cohort big enough for a cross-tutor pattern to surface. But it’s the obvious next gear. The moment you’ve got several tutors, this is what tells you whether a topic is genuinely hard or whether it was taught poorly — and that’s a question no count of “sessions delivered” can answer.

The rule underneath all seven: AI does the volume, you keep the judgement

Notice what none of these are: “ask ChatGPT to teach the lesson.” Every genuine use above has AI doing the high-volume, low-judgement work — marking, formatting, summarising, pattern-spotting — and the tutor doing the part that needs a human who knows the student and the curriculum.

That’s also the honest test for any “AI tutoring tool” you’re evaluating: does it ground its AI in a real curriculum and your students’ actual work, or is it a general chatbot with a badge on it? The first makes your tutors more effective. The second makes them busier cleaning up confident wrong answers.

And notice the other thing: everything above, I do by hand, across four separate tools, because I’m one operator who happens to be technical. It works — but it’s a rig. The reason I’m building ClassQuill is to turn that rig into one place: auto-marking, pre-session intelligence, parent updates, and a curriculum-grounded AI tutor, consistent across every tutor on a team, with nobody having to wire up Claude Code, Wispr Flow and Obsidian themselves. See how it works →