If you are a SENCO looking for AI that actually helps, the useful tools split into two groups and you need both. Sendix and Baobab work on the statutory paperwork: the plan, the provision map, the annual review, the evidence trail behind all three. MagicSchool, Diffit, Read&Write and Glint work on what happens after the plan is written, which is the part that decides whether the plan was worth anything. SLT AI covers the seventh job, the reporting upward. None of the seven makes a decision about a child, and the good ones are built so they cannot.
The timing matters this term. The 2026 Schools White Paper proposes that every setting "create digital Individual Support Plans for any child or young person with identified SEND", inside a new four-tier structure of universal, targeted, targeted plus and specialist support (House of Commons Library briefing CBP-10550, 1 September 2026). The reformed system is not expected to take effect until September 2029, and changes touching EHCP support not before September 2030, so nobody is drafting a digital support plan for a new legal duty yet. What is happening now is that schools are counting how many plans they would have to produce, and realizing the number is larger than the current register.
The Line You Cannot Cross
Everything a SENCO does with AI sits on one side of a line: decisions about need, provision and placement on one side, documents and logistics on the other.
Wiltshire Council has written that line down more plainly than most vendors have. It uses "a secure, AI-powered digital assistant to support the drafting of EHCPs", limited to "summarising and organising information from professional reports, parental contributions, and educational advice". Then it states the boundary: "All EHCPs are reviewed, edited, and finalised by qualified SEND professionals", and "Only Local Authority decision-makers can make decisions about needs, provision, or placements" (Wiltshire Council local offer). The council also says information "is not used for training public AI systems", which is the sentence to look for in any vendor contract you sign.
That is a local authority describing its own practice, not a supplier describing its product, and it is the clearest statement of the boundary available in England right now. Copy it into your own SEN information report and you have answered the governor question before it is asked.
There is a second line, and it is the one people cross by accident. The National Education Union's guidance on AI for pupils with SEND puts it bluntly: "Never input sensitive or identifiable pupil data into personal AI accounts to comply with GDPR" (NEU, 30 June 2026). A SENCO with a personal ChatGPT tab open beside a draft plan is the most common data breach in this whole field, and it is free to avoid.
How I Picked These Seven
The pool was the special education and accessibility category on the aieducator.tools directory, 29 tools as of today, plus a check across all 138 listings for anything that handles provision or plans and sits under another label. Selection was on fit for the SENCO's own job, not the classroom teacher's, and no vendor paid to appear.
What I left out, and why. FrenalyticsEDU does automated progress monitoring across IEP, 504 and MTSS populations, which is the same job in American vocabulary. If you hold the US version of this post, the job title is special education coordinator and the plan is an IEP, and FrenalyticsEDU belongs on your list rather than mine; it is no use to you if your paperwork says EHCP. Khanmigo and Quizlet are pupil-facing study tools with nothing to say about statutory process. Tes Provision Map is a real competitor to Baobab and is not in the directory, so it is named here without a link rather than recommended on evidence I do not have. And any tool whose pitch is that it writes the plan for you is out on principle: a plan you did not think about is a plan you cannot defend at a tribunal.
Prices, free tiers and safety terms were checked against each vendor's own published page on September 30, 2026. Where a vendor publishes nothing, that is said in the section rather than estimated.
What the Directory Record Actually Shows
Three numbers from the directory, all counted today, that are worth knowing before you buy anything.
Of the 29 tools in the special education and accessibility category, 15 state that they collect student data, 10 state that they do not, and 4 leave the field blank. That third group is not evidence of anything except that nobody has asked them.
Five of those 29 carry an independent 9ine security assessment, against 36 of the 138 tools in the whole directory. Special education tools are therefore less likely than the average listing to have been assessed by anyone other than the vendor, which is the opposite of what you would hope for the category handling the most sensitive records in the building. You can see which ones on the independent security assessments page.
Data sits in the United States for 12 of the 29, the United Kingdom for 7, the European Union for 4 and Australia for 1, with 4 unstated. If your data protection impact assessment has a line about international transfers, that line decides your shortlist before any feature does.
Here is the pattern nobody advertises. Every tool on this list that works on the classroom publishes a price: MagicSchool, Diffit, Read&Write, Glint and SLT AI all put a figure on a public page. The two that work on the plan itself, Sendix and Baobab, publish nothing. So the half of a SENCO's toolkit that is specific to the role is also the half you cannot cost without booking a sales call, which makes it the half that never reaches a budget submission in time.
The Seven at a Glance
| Tool | Best for | Strengths | Limitations | Price / free tier | Privacy considerations |
|---|---|---|---|---|---|
| Sendix | Auditing targets in a plan you already have | Flags every target that is not SMART and drafts the fix from the file's own evidence | Works on plans, not provision or timetabling | No public pricing | Directory record: staff-facing, states it does not collect student data, UK GDPR stated, hosting UK |
| Baobab | Running provision maps and annual reviews in one place | Living EHCPs, provision mapping, plan creation, annual reviews, SEND finance, statutory returns | Whole-system purchase, so a slow decision | No public pricing | Directory record: 18+ accounts, states it collects student data, UK GDPR stated, hosting UK |
| MagicSchool | Turning a written plan into classroom-ready adaptations | 80+ staff tools behind one login, including plan drafting and text leveling | Every output is a first draft with your name on it | Free forever tier; Plus $8.33 per user per month billed annually, $12.99 monthly | Directory record leaves FERPA, COPPA, minimum age and hosting unstated; vendor publishes its own privacy pages |
| Diffit | Adapting one specific text to one specific reading age | Re-levels any text or PDF and translates it, no pupil login needed | Adapts the text, not the task | Free version available; school and district pricing by quote, no public figures | Directory record leaves FERPA, COPPA, minimum age and hosting unstated |
| Read&Write | Pupil-facing reading and writing support | Text-to-speech, dictation, word prediction, works across a learner's whole day | An access tool, not a planning tool | Free premium subscription for K-12 teachers; paid plans from $2.75 per seat | Directory record: no minimum age stated, states it does not collect student data |
| Glint | Visual supports, communication boards and social stories | Generates boards, schedules and stories from one prompt | Narrow by design, one job only | $18 per month or $150 per year; enterprise tier for schools; free trial offered, contents not published | Directory record: 18+ accounts, states it does not collect student data |
| SLT AI | The governor report and the SEN information report | 200 leadership tools including a SEND and inclusion set, built for UK leadership writing | Leadership writing only, never touches a plan | £10 per month, or £95 per year for one leader; £85 per user per year for 2 to 5; £75 for 6 to 10 | Directory record: 18+ accounts, states it does not collect student data, hosting UK |
Best for Auditing a Plan You Already Have: Sendix
Sendix is the best fit for checking the quality of targets in an existing plan because it does one narrow thing: it "flags every target that isn't SMART and drafts the fix from evidence in the file" (Sendix). SMART here is the ordinary planning test, a target that is specific, measurable, achievable, relevant and time-bound, as in "will read 60 words per minute by February" rather than "will improve reading". The vendor states it works for "SEND and ASN teams in England and Scotland" and covers EHCPs, Child's Plans and CSPs, the co-ordinated support plans used in Scotland, which makes it the only tool here that names the Scottish documents. It holds a 5.00 rating in the directory from a single review, so treat that number as one person's opinion rather than a trend.
The reason this is the first tool rather than the fourth is that it works on what you have. You do not migrate anything. You take the plan that is already written, in the form the local authority already accepted, and you get back a list of the targets that will not survive a review meeting.
Watch out for: it audits targets, and a plan can have excellent targets and hopeless provision behind them. Nothing here tells you whether the intervention named in section F is actually running on a Tuesday.
Pricing: no public pricing. The directory records it as contact-only and the site publishes no figure or trial (Sendix). Checked September 30, 2026.
Safety: the directory record lists it as staff-facing with no minimum age applied, states that it does not collect student data, and records UK GDPR compliance with UK hosting. FERPA is marked not applicable, which is right for a UK product and is not the same as being cleared for a US district. The privacy notice did not render for machine reading during this check, so ask for it in writing before you upload a plan, and ask specifically whether uploaded files are used to train models. Your school remains the data controller for anything you put in.
Best for Provision Maps and Annual Reviews in One Place: Baobab
Baobab is the best fit for running the whole SEND administrative cycle from one system because it covers "Living EHCPs, provision mapping, IEP creation, annual reviews, SEND finance tracking, statutory returns" (Baobab). That list is the SENCO job description with the teaching taken out. The phrase worth pausing on is "living": a plan that updates as provision changes, rather than a document that is true on the day of the review and drifts for eleven months afterward.
Where Sendix inspects one plan, Baobab is the system the plans live in. That makes it the bigger decision and the slower one, because it replaces a spreadsheet the school has grown attached to.
Watch out for: whole-system tools earn their keep in year two, not term one. Budget the migration time honestly, and expect to keep the spreadsheet running in parallel for a term.
Pricing: no public pricing. The directory records it as paid and the site publishes no figure or trial (Baobab). Checked September 30, 2026.
Safety: the directory record sets a minimum account age of 18, so pupils never hold logins, states that it collects student data, and records UK GDPR compliance with UK hosting. The privacy policy at app.baobab.education is blocked to automated readers, so request it directly. Because it holds student data, this one needs a written data processing agreement naming sub-processors and retention periods before a single pupil record goes in, and your local authority may want to see that agreement too.
Best for Turning a Plan into Classroom Work: MagicSchool
MagicSchool is the best fit for converting a written plan into material a teacher can use on Monday because it puts more than 80 staff tools behind one login, including plan drafting and text leveling, and it carries the strongest review evidence of anything on this list at 4.74 from 19 directory reviews. The practical use for a SENCO is not writing the plan. It is producing the eight versions of a task that the plan implies, for the eight teachers who will each otherwise ask you what adapting means.
Watch out for: every output is a first draft with your name going on it. The free tier is genuinely free rather than a trial, which means the temptation is to work at speed through a pile of plans without reading what came back.
Pricing: a free forever tier including "80+ teacher tools", then Plus at $8.33 per user per month billed annually or $12.99 billed monthly, with custom Enterprise pricing for districts needing data privacy agreements and single sign-on (MagicSchool pricing). Checked September 30, 2026.
Safety: the directory record leaves FERPA, COPPA, minimum age and hosting unstated, so those are open questions rather than failures. The vendor publishes its own privacy and security pages, and the Enterprise tier is where custom data privacy agreements sit, which tells you the free tier is not the place for identifiable pupil detail. Write the plan content yourself and use MagicSchool on the anonymized task, not the named child.
Where Special Education Math Software Fits in a Provision Map
Special education math software sits in the provision map as an intervention with a named frequency, a named adult and a measurable outcome, which is the only form in which any software belongs in a plan. The distinction that matters to a SENCO is between a tool that adapts the material and a tool that delivers an intervention: the first is ordinary quality-first teaching and does not need to be named in section F, the second is provision and does. If you are choosing between products for pupils working several years below their year group, the special education math software comparison goes through them properly, and the AI tools for special education teachers guide covers the classroom side of the same question.
Best for Adapting One Text Today: Diffit
Diffit is the best fit for the adaptation you need in the next ten minutes because it re-levels any text or PDF for reading level and translates it, without a pupil login anywhere in the process. Paste in the Year 9 history source, get it back at three reading ages and in the home languages on your register. It holds 4.20 from 5 directory reviews.
Watch out for: it adapts the text and not the task. A source rewritten at a lower reading age attached to an unchanged exam-style question is a harder piece of work, not an easier one.
Pricing: a free version remains available; Basic and Diffit for Schools tiers exist with no published figures, and school pricing is described as a "flat-rate annual subscription, tiered based on student enrollment" (Diffit pricing). No public figure for either paid tier. Checked September 30, 2026.
Safety: the directory record leaves FERPA, COPPA, minimum age and hosting unstated. Nothing in the workflow requires a pupil's name, which is the reason it is on this list: the safest tool is the one with no reason to know who the child is.
Best for Pupil-Facing Reading and Writing: Read&Write
Read&Write is the best fit for support that follows the pupil rather than the lesson because it "works alongside every learner" across text-to-speech, dictation, word prediction and comprehension, and because a K-12 teacher can have the premium version at no cost. It holds 4.50 from 2 directory reviews.
One piece of housekeeping found in this check: the product now publishes under the Everway name, and texthelp.com/products/read-and-write-education redirects to everway.com (Read&Write). Training slides and access arrangement paperwork naming Texthelp are pointing at a company name the vendor has stopped using.
Watch out for: access technology only works if it is on the device the pupil actually uses in an exam, and access arrangements have their own evidence requirements that no tool satisfies on your behalf.
Pricing: "K-12 teachers get a premium Read&Write subscription for free"; institutional plans are advertised from $2.75 per seat (Read&Write). Checked September 30, 2026.
Safety: the directory record states no minimum age and that it does not collect student data. Because this one runs on pupil devices, the question to ask your technology lead is not about the vendor at all: it is whether the deployment reaches the pupils named in plans, or only the ones whose parents asked.
Best for Visual Supports: Glint
Glint is the best fit for communication boards, visual schedules and social stories because it generates all three from a single prompt, which turns a job that used to eat a Sunday into one that fits between two meetings. The vendor describes it as an "AI-powered visual supports generator" and aims it at districts, speech and language therapists, and families.
Watch out for: a generated social story carries the generator's assumptions about the child. Read every one with the adult who knows them, and expect to rewrite the ending.
Pricing: $18 per month or $150 per year, with an enterprise tier for schools and organizations; a free trial is offered but the pricing page does not publish what the free level includes (Glint pricing). Checked September 30, 2026.
Safety: the directory record sets a minimum account age of 18 and states that it does not collect student data, which fits a tool that staff use to make resources rather than one pupils log into. The vendor's enterprise tier advertises FERPA and COPPA alignment; that claim belongs to the enterprise tier and should be read as a starting point for your own checks, not a clearance.
Best for the Governor Report: SLT AI
SLT AI is the best fit for the writing that goes upward because it is built for UK school leadership specifically, with a SEND and inclusion tool set sitting alongside governance, compliance and school improvement, and it holds no directory reviews yet. A SENCO writes to three audiences: parents, teachers and the people who hold the school to account. This is the only tool here that understands the third one, and it knows what a governor report and a SEN information report are without being told.
Watch out for: fluent leadership prose is the easiest thing in education to generate and the easiest to see through. A governors' SEND report that reads better than the provision it describes is a problem you have made worse.
Pricing: £10 per month or £95 per year for one leader, £85 per user per year for 2 to 5 leaders, £75 per user per year for 6 to 10, with quotes above that; the pricing page lists no free plan (SLT AI pricing). Checked September 30, 2026.
Safety: the directory record sets a minimum account age of 18, states that it does not collect student data, and records UK GDPR compliance with UK hosting. FERPA is marked not applicable, correctly for a UK product. The risk here is not the platform, it is the habit: leadership writing pulls named case studies toward it, and a governor report does not need a pupil's name to make its point.
The Big Names Search Assistants Keep Suggesting
Ask a chatbot for AI tools for SENCOs and it will offer you four things that are not.
ChatGPT is the most capable general drafting tool available and the riskiest place for this particular job, because the boundary between a work account and a personal one is invisible from inside the tab. Use it on anonymized text or not at all, per the NEU's guidance above.
NotebookLM, now Gemini Notebook, is the one general tool genuinely worth a SENCO's time, because it answers only from documents you upload and cites where each answer came from. Load the professional reports for one pupil and ask what evidence exists for a specific need. It is not SEND software; it is a way of reading forty pages fast with the citation attached.
Khanmigo is a pupil tutor. It has no view of a plan and no statutory awareness, and it belongs in a conversation about intervention delivery, not paperwork.
Quizlet is revision. It earns no place in a provision map beyond what any flashcard does.
Four Questions That Decide It
Work through these in order and the shortlist chooses itself.
1. Is this a document job or a decision job? If the answer is decision, close the tab. Need, provision and placement are determinations made by people, and Wiltshire Council's own statement is the wording to borrow when you explain that to staff.
2. Whose account does the child's name go into? A school-contracted account with a data processing agreement, or nobody's. If the tool has no agreement available, the workflow has to work on anonymized material, which rules out about half of what you would like to do with it.
3. Does it know the statutory form, or only the English? Sendix and Baobab know what section F is. A general writing tool produces something that reads like a plan and is not one. For anything that will be quoted back at you in a meeting, that difference is the whole purchase.
4. What happens to the evidence if you leave the school? A provision map inside a subscription one person set up is a single point of failure, and SENCO turnover is the reason SEND records go missing. Buy at school level or accept that you are the backup.
A Fortnight Planned Out: One Year 8 SEN Support Review
Here is how the tools fit a single real task: reviewing the SEN support plan of one Year 8 pupil with a specific literacy need, across two weeks, without a new system going in.
Days 1 and 2. Gather what exists: last year's plan, two subject teacher comments, the reading assessment from September. Load them into Gemini Notebook and ask one question, which is what evidence exists for the target that is already written. You are looking for the target nobody can show progress against.
Day 3. Run the plan through Sendix. Take the list of targets that are not measurable, and rewrite two of them yourself using the evidence Notebook surfaced. Two is the right number. A plan with twelve new targets is a plan nobody will deliver.
Days 4 to 6. Turn each surviving target into something a teacher can do. For the reading target, take the actual text the English department is teaching next and run it through Diffit at the pupil's reading age, then use MagicSchool to produce the adapted task that goes with it, not just the adapted text. Send both to the two teachers who asked what adapting means.
Day 7. Check that Read&Write is deployed on the device this pupil uses in lessons, not just available in principle. If it is not, that is the week's real finding and no plan wording fixes it.
Days 8 to 10. Write the review record yourself, in your own words, naming what changed and what the evidence was. Use SLT AI only for the summary paragraph that goes to governors, and only after the record is finished.
What you measure. One thing: whether the reading target has a number beside it in December that was not there in September. If it does not, the plan was paperwork.
The Bottom Line
Start with Sendix if your plans are written and you doubt their quality, and with Baobab if the plans are fine and the provision behind them lives in a spreadsheet only you can read. Add MagicSchool and Diffit for the week-to-week work of making adaptation real, because a plan that never reaches a lesson is the most expensive document in the school. Keep every one of them on the document side of the line, and borrow Wiltshire's wording when somebody asks you why. The rest of the field is worth a look in the special education and accessibility category before you commit a budget line, if only to see how many vendors leave the privacy fields blank.
Seventy-four percent of SENCOs told the national workload survey that administration takes up the majority of their allocated SENCO time (Bath Spa University, nasen and the National Education Union, October 2019). That figure is six years old and no SENCO reading it will think it has improved. The tools above can take a real share of that majority off you. What they cannot do is the thinking, and the thinking is the job.
Dan Fitzpatrick is The AI Educator. More on Dan.
