AI Home Education: Building Personalised Learning Portfolios

Personalised learning portfolios answer two questions every home-educating parent gets asked — by local authorities, by doubting relatives, and by themselves at 11pm: what is my child actually learning, and can you show me? A good portfolio is the evidence. A personalised one is evidence that clearly belongs to your child — their interests, their pace, their voice.

AI has quietly made this much easier. Not because AI writes the portfolio (it shouldn't — the evidence has to be real work), but because AI removes the two bottlenecks that used to make portfolios patchy: generating genuinely differentiated activities on demand, and structuring what you capture so it maps to learning goals instead of becoming a shoebox of photos.

This guide walks through the whole workflow: profiling the child, generating differentiated activities, and filing evidence into a portfolio you can actually export.

Step 1 — Build a learner profile before you generate anything

Every personalisation failure we've seen comes from the same place: asking an AI for "a Year 4 science activity" with no context. The output is generic because the input was generic. Spend twenty minutes once filling in an intake like this, and paste it at the top of every prompt (or save it as your child's profile in whatever tool you use).

Age, stage and interest intake template

Child: [first name or initial — you don't need full identifiers]
Age / year group: [e.g. 9, Year 4]
Curriculum or framework: [e.g. National Curriculum England,
  Common Core, EYFS, none — interest-led]
Current working level: [e.g. secure in Y4 maths, reading slightly
  below age expectations, EAL support for topic vocabulary]

Interests (the hooks that make work feel theirs):
  - [e.g. Minecraft, horses, dinosaurs, baking, football, space]

Learning preferences:
  - Works best: [e.g. hands-on, short bursts, discussion first,
    worksheets fine, needs movement breaks]
  - Avoids / struggles with: [e.g. long written tasks, timed tests,
    abstract maths without concrete examples]

Adjustments: [e.g. dyslexia-friendly fonts, chunked instructions,
  extra processing time, sensory considerations]

Subjects to cover this term: [list]
Compliance context: [e.g. annual LA review in England, US state
  portfolio review, no formal requirement — records for ourselves]

Two things matter here that parents often skip: current working level per subject (a child is rarely "at" their year group across the board) and the avoidance list (telling the AI what not to generate saves more time than any other single line).

Step 2 — Generate differentiated activities from the profile

With the profile saved, prompts get short and the outputs get specific. Here are ten we use, written to be pasted under a filled-in intake. They work in any AI tool — and they're the same patterns EdToolkit's plan and activity generation is built around.

Activity generation

1. Design a [45-minute] [subject] activity on [topic] for this child.
   Anchor at least two parts in their interests. Include what success
   looks like in one sentence.

2. Give me three versions of the same [topic] activity: hands-on,
   discussion-led, and worksheet-based. Same learning objective each time.

3. Take this [lesson/resource] and adapt it to two levels:
   [level A] and [level B]. Keep the key vocabulary identical
   across both so they can be used side by side.

4. Build a week-long mini-project on [topic] that touches
   [subject 1], [subject 2] and [subject 3]. Break it into daily
   30–45 minute chunks with one visible output per day.

Checking understanding

5. Write a 10-question quiz on [topic] pitched at this child's level.
   Mix recall and application. Include an answer key with a one-line
   hint for each question — hints first, answers last.

6. This child just did [activity]. Write five diagnostic questions
   that tell me whether they understood [concept] or just followed
   the steps.

7. Create a marking rubric for [task] with three levels:
   emerging, expected, exceeding. Criteria a 9-year-old could
   self-assess against, plus a parent column.

Stretching and supporting

8. My child found [activity] too easy. Generate the next step up —
   same topic, higher demand — without making it just 'more writing'.

9. My child is stuck on [concept] in [topic]. Suggest three different
   ways to approach it that aren't 'the same thing explained louder':
   one concrete, one visual, one through their interest in [interest].

10. Design a capture-the-evidence version of [activity]: what should
    the child produce, and what should I photograph or note down, to
    prove [learning objective] happened?

That last one is the bridge to the portfolio — generating the activity and its evidence plan at the same time.

What it looks like in practice

Literacy — Year 4, interested in horses, avoids long writing. Instead of "write a story", the profile produces: a labelled diagram of a knight's horse and tack (vocabulary), a short persuasive paragraph on whether horses should live in stables or fields (writing objective, genuinely interesting to this child), then reading a non-fiction text about cavalry at two levels. Same literacy objectives, none of the "I don't know what to write" stall.

Maths — Year 2, secure in number, needs concrete handling. Instead of a worksheet on money: a priced "tuck shop" with real coins, making change for purchases the child chooses. The prompt generates the price list at the right number range and the three extension questions for when it clicks early. Evidence is a photo of the shop plus the child's written totals.

Science — mixed-age siblings, one activity. One investigation (which material makes the best paper bridge?), each child working at their level: the 6-year-old predicts and photographs, the 10-year-old records variables, results in a table and a conclusion sentence. The differentiation prompt (#3) produces both versions in one pass instead of two separate planning sessions.

Practical learning — no classroom required. Baking scales to ratio and fractions; a walk becomes a map-reading and estimation exercise; a Minecraft build becomes perimeter and area. The intake's interest list is what makes these land — the AI is very good at wrapping the same objective around a different interest once you tell it the interest exists.

Step 3 — File the evidence where it becomes a portfolio

The difference between a folder of photos and a portfolio is linkage: every artefact attached to a child, a date, and what it was meant to show. The workflow that survives real life is boring on purpose:

1. Generate the activity with its evidence plan — what will be produced, what to photograph, what to note. (Prompt 10 does this automatically.)
2. Do the activity. This part is mercifully non-digital.
3. File it the same day. Photograph the physical work, save the generated resource itself, add two lines of context: what happened, what it showed. Two minutes while it's fresh beats twenty at the end of term.
4. Tag it to an objective. "Bridge investigation — working scientifically: results table" is retrievable at review time; "Tuesday science" is not.
5. Review monthly, not yearly. Fifteen minutes a month keeps the portfolio honest and turns up gaps while there's still time to fill them.

In EdToolkit this loop is the product: child profiles hold the intake from Step 1, curriculum generation turns it into weekly plans and activities (Steps 2), the quiz and rubric builders produce the assessment artefacts, and portfolio management files each piece of evidence against the child and their objectives — then exports collections when a reviewer, local authority or just a sceptical grandparent wants to see them. Each tool exists separately elsewhere; the point of keeping them in one system is that the evidence stays linked to the plan it came from.

If you're working with a general AI assistant instead, the same workflow holds — your "system" is a folder structure, a tagging convention in filenames, and a spreadsheet or doc for the objective links. It's more manual, but the discipline is identical.

What a portfolio is not

Worth saying plainly: a portfolio is not AI-generated essays filed as your child's work. Reviewers spot adult-perfect prose instantly, and more importantly it proves nothing about the child. The AI's job in this workflow is upstream — designing the activity, differentiating it, writing the quiz, structuring the rubric. The portfolio contents should be unmistakably the child's: their handwriting, their photos, their reasoning, wobbly corners and all.