AI for Education Consultants: Seven Jobs, Proposals to Reports

An education consultant can hand seven jobs to AI: proposals, intake, visit write-ups, client reports, newsletters, invoice chasing and scheduling. Each one is first-draft work that you then check. Pupil data, staff records and unpublished school documents stay out unless the tool is a business plan with training switched off.

The split that matters is drafting versus judging. AI is quick at turning your notes into a structure and a first pass of wording. It cannot know what you saw in the corridor, which headteacher needs a blunt message or which finding the governors will fight. Keep the judgement; hand over the typing. For the general proposal workflow across consultancies, see how consultants write proposals in under an hour; this tutorial stays with the education practice.

Follow me on Instagram@sagnikteaches

Rules that apply to all seven jobs

Education consultants hold more sensitive material than most advisers: pupil outcomes, safeguarding concerns, staff capability issues, budget deficits and governor minutes. Before any job below, set three rules and write them on one page.

Connect on LinkedInSagnik Bhattacharya

• Tier your material. Public (published inspection reports, school websites, national guidance), internal (your own templates and methods) and restricted (anything naming a child, a teacher or a safeguarding matter). Public and internal can go into a paid assistant. Restricted never goes in, or goes in only after you replace names with labels such as "Pupil A" and "Teacher 3" and remove dates of birth and school identifiers.
• Use the right plan. Business and team plans from OpenAI, Anthropic and Google do not train on your content by default. Consumer plans have a model-training setting you must check yourself: on ChatGPT's Free, Plus and Pro plans it is on until you turn it off, and Claude and Gemini personal accounts each have their own switch. Read the data-controls page of whichever plan you use rather than trusting memory; the labels change.
• You own the output. A client hires you, not the tool. Every deliverable gets a human read against your notes before it leaves your desk.

If you want a template for telling clients, the wording in mentioning AI use in client contracts and proposals adapts well. Now the seven jobs.

Subscribe on YouTube@sagnikteaches

1. Proposals that reuse your best paragraphs

A school improvement proposal usually has the same bones: context, your reading of the problem, approach, outputs, timeline, fee. The time sink is rewording your approach for each school. Keep a library of your own paragraphs (how you run a curriculum review, how you moderate marking, how you coach a middle-leadership team) and let AI assemble and adapt them from a short brief.

Say a solo consultant is quoting for a primary school that wants help with maths mastery. The brief is five lines from the call: 280 pupils, maths results flat for two years, a new subject lead, six staff meetings available, a budget around $6,000. The consultant pastes the brief plus her approach library into the assistant with this prompt.

You are drafting a consultancy proposal for me. Use ONLY the approach paragraphs I paste below; do not invent methods or results.
Brief from the call: [paste five lines, no names].
My approach library: [paste paragraphs].
Write: 1) two sentences restating the school's problem in their words, 2) a three-phase plan across six staff meetings, 3) outputs the school will receive, 4) three assumptions I am making, 5) three questions I still need answered.
British English. No claims about outcomes.

The draft comes back in about a minute. What she fixes: the assistant tends to promise "improved outcomes", so she deletes that sentence and replaces it with what she controls (a diagnostic, a coaching cycle, a moderation pack). Her time falls from roughly two hours to forty minutes, most of it spent on the fee line and the assumptions. The assumptions list is the part that protects you later, so read it hardest.

A quick sum shows why the fee line deserves your time rather than the tool's. Suppose the maths project is six staff meetings at 90 minutes each, two preparation days, one day of lesson visits and a half-day for the written report. Nine hours of meetings is about one and a half days, so delivery comes to roughly five days. At a day rate of $1,100 the fee is $5,500, and a $400 allowance for materials and travel brings it to $5,900, just inside the $6,000 budget. Drop the report half-day by mistake and the quote falls to $5,350, which looks competitive until you write the report unpaid. AI can lay this out as a table, but it will happily round a day rate or forget the report half-day, so recompute every line on a calculator before the proposal goes out.

One more habit pays back quickly: ask the assistant to read your finished proposal as the head of finance and list the three questions that person would ask. Typical ones are "what happens if a meeting is cancelled", "who owns the materials afterwards" and "what is the payment schedule". Answer them in the document and the second call becomes a yes rather than a negotiation.

2. Intake notes and discovery questions

New enquiries arrive as long emails, voicemails and forwarded documents. Ask AI to turn the raw text into a one-page intake sheet with fixed fields: who is asking, what they say the problem is, what they have already tried, deadline, budget signal, decision maker and red flags.

A before and after shows the value. The before is a rambling email from a deputy head about "behaviour not where we want it, inspection due, new head arriving in January". The AI sheet reads: problem as stated (behaviour), likely underlying questions (consistency of routines, staff confidence, data on incidents), deadline (inspection date unknown, ask), decision maker (deputy head or incoming head, ask), red flag (the person asking may not be the person paying). Then it drafts six discovery questions. You will not use all six, but they stop the first call being a free-for-all.

Keep the sheet in one place, such as a shared folder or CRM note, so the proposal in job 1 starts from it. If chasing missing documents is where intake stalls, chasing missing client records with AI covers reminder wording.

3. Turning visit notes into a structured write-up

School visits produce scrappy notes: lesson observations, learning walks, meetings with leaders, a book look. Typing these into a report is the least skilled hour of the week. Dictate or jot your notes on the day, then have AI sort them under the headings of your framework.

The restricted-data rule bites hardest here, because visit notes name children and teachers. The workable habit is to write notes with labels from the start ("Y4 teacher A", "Pupil 2") and keep the key linking labels to names in a separate file that never goes near the tool. Then the paste is already safe.

Sort these visit notes under my headings: Teaching and learning, Curriculum, Leadership, Pupil behaviour, Evidence I still need.
Keep my wording where it is clear. Do not add observations I did not make. Mark anything ambiguous with [CHECK].
Notes: [paste labelled notes]

A realistic output for a secondary geography visit: five bullets under Teaching and learning, a Curriculum heading with a single line "no evidence seen of sequencing across Years 8 and 9 [CHECK]", and an Evidence I still need list with two items (the scheme of work, the last two assessment results). The [CHECK] flags are the point: they show you what the tool could not support from your notes instead of letting it smooth the gap over.

4. Client reports that start from evidence, not a blank page

The report is the product. Do not let AI write the findings; let it build the scaffolding. Give it your structured notes from job 3 and ask for an executive summary draft, a findings table with columns for finding, evidence, strength of evidence and recommendation, and a first pass at an action plan with owners and dates left as blanks for you to confirm.

Then do the work only you can do. Rank the recommendations. Delete the ones that are true but not worth the school's attention this term. Add the sentence that says what you would do first if it were your school. A useful test: read the executive summary aloud to someone who has not seen the notes. If every sentence could describe any school, rewrite it with a specific detail from your visit.

Here is a filled-in findings row, illustrative, so you can see the standard to aim for. Finding: "Pupils in Year 5 can recall number facts but cannot explain why a method works." Evidence: "Three lessons observed, two pupil conversations, a sample of ten books." Strength of evidence: "Moderate; one class only." Recommendation: "Add a weekly explain-it-back task and review it in the January staff meeting." The AI's first draft of that row usually reads "Pupils show gaps in conceptual understanding" with no evidence column filled. If the evidence cell is vague, the finding is not ready, and you go back to your notes rather than asking the tool to sound surer.

Ask for a second output too: a plain-language one-page version for governors or parents, if the contract allows it. Tell the assistant the reading level and that it must not add any claim missing from the full report. Compare the two side by side, because shortening is where softened or exaggerated wording creeps in.

Build the report as a template with fixed headings so every client report looks like yours and the assistant fills the same shape each time. Over a term, the template also lets you compare schools without rereading everything.

5. A newsletter that shows your thinking between projects

Education consultants win work through trust, and a short monthly newsletter keeps heads and trust leaders in mind. The usual failure is that it never gets written. Make AI the editor, not the author: you record a two-minute voice note on what you noticed this month across your visits (nothing identifying), and the tool turns it into 300 words with a subject line and one practical tip.

A realistic example: your note says that three schools you visited all had assessment data nobody used. The draft newsletter has the headline "Data nobody opens", a three-step test (who reads it, what decision changes, when is it reviewed) and a closing question. You cut the generic opening, add one line of your own opinion, and send it. Fifteen minutes, not two hours.

Two guardrails. First, never describe a visit in a way a reader could identify, even without names ("a coastal primary with a new head" is still identifiable to the people involved). Second, do not let the tool invent statistics to sound authoritative. If a number matters, source it yourself, and read using ChatGPT as an adviser without being misled for the habit of checking claims.

6. Invoices, reminders and the awkward late-payment email

Schools and trusts pay on their own cycles, and purchase-order numbers go missing. AI will not issue invoices, but it will draft the three messages that eat your goodwill: the invoice cover note quoting the purchase order, the polite reminder at 14 days and the firmer one at 30.

Say a consultant invoices $1,800 for two days of leadership coaching and the trust's finance team replies that the purchase order is missing. A good prompt gives the facts and the tone:

Write a short, friendly email to a school business manager. Invoice [number] for $1,800 was sent on [date] and is now 14 days old. Their finance team said a purchase order number was missing. Ask them to confirm the PO number or tell me who to send it to. Under 90 words. No threats. Sign off "[your name]".

The output is usually fine on tone but sometimes adds a sentence about "late payment charges" you never set. Delete anything that states terms you have not agreed. Keep a small file of the three messages as reusable templates, and let your accounting software send the reminders on a schedule if it can. The amount is never typed by the AI; it comes from the invoice.

For the firmer 30-day message, the same prompt works with one change: add the fact that the invoice terms say payment within 30 days and ask for a payment date. Sample output, illustrative: "Hello [first name], invoice [number] for $1,800 is now 30 days old and, under the terms in our agreement, was due today. Could you let me know when it will be paid, or whether anything is holding it up? Thank you, [your name]." That is firm without being cold. If a school has a history of slow payment, put a deposit or a stage payment into the next proposal instead of writing sharper emails.

7. Scheduling visits around school calendars

Term dates, staff training days, trips, exam periods and staff meetings make scheduling a puzzle. AI is good at the puzzle part if you give it constraints in plain text: your available days, the school's blocked dates, travel time between sites, and the order of phases.

A consultant with three schools in a half-term needs 9 days of visits. She pastes each school's blocked dates and her own constraints (no more than two schools in one week, a day between visits for writing up). The tool proposes a calendar and flags a clash she missed: School B's exam week falls inside the phase meant for lesson observations. She moves it. The tool cannot see a calendar it has not been given, so the check is that every date in the proposal appears in a source she pasted.

If you use an online booking page, a shared calendar and a short confirmation message template do most of this without AI. Use the assistant for the planning conversation, and the calendar for the actual bookings.

A filled-in confidentiality rule you can copy

Here is a one-page rule a small consultancy could adopt as written. Replace the brackets with your own tools.

• We use [tool] on a [plan name] plan with training switched off.
• We never enter names of pupils, staff or governors; we use labels, and the key stays in [location].
• We never enter safeguarding information, health information, or unpublished HR or finance documents.
• Published documents and our own templates can be used freely.
• Every deliverable is read in full by [name] before it is sent, and any figure is checked against its source.
• If a client asks us not to use AI on their work, we do not, and the proposal says so.

Review it every term. The vendors change their settings, and so do your habits. A rule that was true in September can be wrong by January.

Where consultancy drafts slip: pasted names, unsupported advice

A realistic mistake: a consultant pastes a whole governors' report to "summarise" it, and the report contains a named member of staff under a capability process. The data is now in a consumer chat history. The fix is the tiering rule above, plus a habit of scanning pasted text for names before sending. Another failure is quieter: a report that reads fluently but contains a recommendation unsupported by any note. The [CHECK] habit in job 3, and a final pass where you tick every recommendation against evidence, catch most of it.

How to know it is working after a month: time per proposal, time per report, and the number of times a client asked a question the report should have answered. If drafting time falls but client questions rise, you are shipping fluent but thin work. Slow down on the judgement sections.

Start with one job, then widen

Most solo consultants gain the most from jobs 3 and 4 together, because visit notes into a report is the biggest block of unbilled time. Pick one job, run it on your next two projects, and measure the hours. If you would rather map all seven with someone, that is what my AI implementation consultation covers; get in touch. For the wider picture of AI jobs across sectors, see the AI hub.

Questions education consultants ask before starting

Can I paste a school's inspection report into ChatGPT?

Only if the report is public and you are comfortable with the plan's data terms. Anything marked confidential, any pupil or staff detail, and any draft the school has not published should stay out of a consumer chat tool. Use a business plan, which does not train on your content by default, or strip names and identifiers before pasting.

Should I tell clients I use AI in my practice?

Yes, in one plain sentence in your terms or proposal: what you use it for (drafting, tidying notes), what you never put into it (pupil data, staff records), and that you check and own every deliverable. Clients mostly want to know that a person is accountable for the final work.

Which tool should a solo education consultant start with?

Start with one general assistant on a paid individual or small-team plan, because that is enough for all seven jobs here. The privacy terms of the plan matter more than which model is marginally better. Add scheduling and invoicing features inside the tools you already use before buying anything new.

Will AI-written reports sound the same as everyone else's?

They will if you let the tool choose the voice. Feed it two or three of your own past reports as a style sample, give it your headings, and rewrite the judgement paragraphs yourself. The evidence, the recommendations and the tone of challenge are the parts clients pay for.

Further reads

• AI Readiness Checklist for Accountants, Solicitors, Consultants — Score your practice before adding more tools.
• Questions to Ask Before Buying AI That Touches Client Data — Questions to put to any vendor before client data goes in.
• Should You Charge Clients Less When AI Speeds Up Your Work? — Decide what happens to your fees once drafting gets faster.
• What Does an AI Implementation Consultant Actually Do? — See what an adviser on the other side of the table actually does.
• How to Evaluate an AI Implementation Proposal or Quote — Judge any AI quote you are handed, including your own.
• AI Implementation for Education in Australia: A Starter Plan — A 90-day AI plan for an education business: choose two jobs, check the privacy principles, pilot one tool, measure, and hand over.
• College Counselling Business: Can You Use AI on Student Essays? — Where AI feedback on student essays stays honest, where it becomes ghost-writing, and a disclosure and review routine for counsellors.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI data controls FAQ; Anthropic privacy help articles on model training (consumer and commercial); Google Workspace generative AI privacy hub; Gemini Apps privacy hub; checked 4 November 2026.