On September 1, 2026, the District of Columbia's Office of the State Superintendent of Education published a model AI policy that puts determining eligibility for an IEP or a Section 504 plan in its prohibited category, and puts drafting IEP language in the restricted-but-allowed one (OSSE release). That line, between the writing and the deciding, is the only one a school psychologist needs. Everything below sits on the writing side of it.
Seven tools earn a place: PAR's AI Report Writer for the evaluation report itself, MagicSchool for recommendations and behavior plans, FrenalyticsEDU for the progress data you will need at the next annual review, Diffit for the family-facing summary, Glint for the visual supports you recommend, Read&Write for the accommodation you want to trial before you name it in a report, and StaffDraft for the message home. None of them decides anything.
This is written for school psychologists in US districts, and for the educational psychologists in the UK, Ireland and Australia who do a recognizably similar job on a different statutory clock. It is also for the special education directors who sign the purchase orders, because one finding below is about money and it is not the one you would expect.
What each tool does, and what it costs
| Tool | Best for | Pricing (checked September 21, 2026) | Free tier | Data hosted | Independent security assessment |
|---|---|---|---|---|---|
| PAR AI Report Writer | The psychoeducational report draft | $18 per report (5 minimum), $49 a month, or $549 a year | No | Private Azure OpenAI deployment (vendor site) | Not listed in the directory |
| MagicSchool | Recommendations and behavior plan language | Free; Plus $8.33 a user a month billed annually; Enterprise on quote | Yes, 80+ tools | US (vendor site) | Yes (9ine) |
| FrenalyticsEDU | Progress monitoring against IEP goals | Free; EDU Essentials $14.99 a month; EDU Extras $16.99 a month | Yes, 10 students, 1 staff account | US | No |
| Diffit | The summary a family can actually read | Free; individual teacher $14.99 a month or $149.99 a year; schools on quote | Yes, 90 days of your own resources | Not stated in the directory | No |
| Glint | Visual schedules, boards and social stories | $18 a month or $150 a year; enterprise on quote | Trial, plus free shared boards for families | US | No |
| Read&Write | Trialing an accommodation before you name it | Free for K-12 teachers; individual $184 a year; district $2.75 a seat | Yes, for K-12 teachers | Not stated in the directory | Yes (9ine) |
| StaffDraft | The routine message home | Free for 10 drafts a month; Individual £6 a month; organization £5 a user a month | Yes, 10 drafts a month | UK | No |
How this list was built
I started with the 29 tools in the directory's special education and accessibility category, then searched all 138 records for assessment, progress monitoring, report writing and communication functions, then went outside the directory for the one category it does not yet cover, which is software built for the psychoeducational report itself. Seven made the cut on five tests: does a job a school psychologist actually has, keeps the clinical judgment with the psychologist, has a safety posture I could check on the vendor's own page today, prices honestly, and does not need a district contract before you can see whether it works.
Nobody pays to appear here. Three near-misses are worth naming. SchoolAI is built around student-facing chat with teacher monitoring, which is a classroom tool rather than an evaluation tool. Sendix and Baobab both draft English EHCPs against the Children and Families Act, which is precise and useful and has no US equivalent, so they belong in a UK list rather than this one. And PsychAssist.ai, which does cover school psychologists, is priced per clinic at $1,499 a month for its Pro tier (pricing), which is private-practice money.
That last exclusion turned into the most interesting thing this research found.
The price gap nobody warns you about
Software for writing psychological reports has split into two markets with an order of magnitude between them, and school psychologists sit on the wrong side of the crack.
On one side, PAR sells AI Report Writer at $18 a report with a five-report minimum, or $549 a year for unlimited reports (pricing). On the other, PsychAssist prices Pro at $1,499 a month per clinic, with each additional psychologist at $349 a month (pricing). Both name school psychologists as a target user. A district running eight psychologists would pay roughly $4,400 a year for the first and roughly $47,000 for the second.
The gap is not a quality judgment; the products aim at different buyers. It matters because of who does the buying. School psychologists are usually funded out of special education, not out of the edtech budget, and the tools that reach them through district procurement are general teacher platforms that were never built for protected evaluation data. The tools built for that data are bought one license at a time, by individuals, outside the process that would have checked them. A special education director who has never seen either name on a purchase order should assume the second thing is already happening.
What the directory data says about the tools around this job
Of the 29 tools in the directory's special education and accessibility category, 14 declare FERPA compliance and only 5 carry an independent 9ine security assessment. Twenty-four state where they host data, which is better than the directory as a whole, where 90 of 138 do. Two declare that they share data with third parties. Sixteen have a free or freemium tier.
Read those numbers the right way. The compliance columns are vendor declarations, not audits; the 9ine column is the independent one, and it is thin. If you are about to enter anything from a student's file, the declaration is where you start looking, not where you stop.
Which tool should a school psychologist open first?
Four questions settle it, and they work on anything a colleague recommends at a district meeting.
Does a protocol, a score or a name go into it? This is the only question that changes the rules rather than the ranking. A tool you use to rewrite a paragraph you already wrote is a word processor with ambitions. A tool you feed WISC subtest scores, a teacher's behavior rating or a student's name becomes a recipient of an education record, and the Department of Education's Student Privacy Policy Office is specific about what has to be true before a vendor counts as a school official: it performs a service the district would otherwise use employees for, it meets the district's own published criteria for legitimate educational interest, it is "under the direct control of the agency or institution with respect to the use and maintenance of education records," and it "uses education records only for authorized purposes" without redisclosing them (Vendor FAQ). A consumer chatbot you signed up for with your work email meets none of those four.
Would you be comfortable reading the output aloud at the eligibility meeting? Not the polished version. The draft. If the answer is no because the sentences are generic, the tool is saving you the wrong work. If the answer is no because you cannot vouch for a claim in it, you have found the real risk, which is not that AI writes badly but that it writes plausibly.
Does it save minutes at the end, or at the start? Tools that draft save you at the start and cost you at the end, in checking. Tools that collect data while a student works, like progress monitoring software, save you at the end, on the day you would otherwise be reconstructing six months of trials from memory. The second kind is worth more and gets bought less.
Can you stop using it in June without losing anything? Progress data belongs to the district. Report templates belong to you. Before you put either into a product, find the export button. If there is not one, that is the answer.
PAR's AI Report Writer is the best fit for the report itself, because the company that publishes the tests built it
PAR is the publisher behind a shelf of the instruments school psychologists already use, and AI Report Writer sits inside PARiConnect, drafting a psychoeducational report from uploaded scores, interview notes and observations. You choose the tone, including a parent-friendly setting, and the detail level, and you can regenerate any individual section up to three times before exporting to Word (product page).
The privacy language is the most specific of anything in this list. PAR states that "data submitted through AI Report Writer is never used to train or improve any AI model," describes a dedicated private Azure OpenAI deployment where the data is not available to other customers, and says identifying information is tokenized, replaced with random character strings while you work online and restored when you download (product page). That is the architecture the job requires, described plainly, which almost nobody else here manages.
Watch out for: it drafts from what you give it, so a thin battery produces a confident, well-organized, thin report. The three-regenerations-per-section cap is real, and if you find yourself reaching it, the problem is usually the input. It is not listed in the aieducator.tools directory, so there is no independent rating, review count or 9ine assessment on it here.
Pricing: $18 per report with a five-report minimum, $49 a month for unlimited reports, or $549 a year for unlimited reports, all tiers with identical features (pricing). Checked September 21, 2026.
Safety: PAR states the data is never used to train a model and that identifiers are tokenized in the online view. It does not remove your district's obligations: you still need this on your approved vendor list, with a written agreement that meets the four school-official conditions above, before a real student's scores go into it. Ask your special education director whether one exists; if the answer is that a psychologist bought it on a department card, that is the conversation to have first.
MagicSchool is the best fit for the recommendations section, because its free plan covers the writing without a purchase order
The part of the report that takes longest to write well is the part with no numbers in it: the recommendations, the behavior support language, the paragraph explaining what a processing speed score means for a fourth-grade classroom. MagicSchool's free tier carries 80-plus teacher tools including its student support set, which drafts behavior intervention language and family letters (pricing). It carries a 4.74 rating from 19 reviews in the directory and an independent 9ine security assessment, which only 5 of the 29 special education tools listed do.
Use it the way the evidence says practitioners already use it. A national survey of school psychologists using AI at work found that 69 percent used it to rewrite sections for clarity and 67 percent to generate recommendations, while none relied on it to produce an entire report, and 94 percent edited what it gave them (Journal of Applied School Psychology). Rewriting and recommending is exactly the band MagicSchool is good at.
Watch out for: the free plan caps generations, and the custom data privacy agreement is an Enterprise feature, so a free or Plus account means accepting the standard terms on your own authority. It is a teacher platform, so its vocabulary is instructional rather than clinical, and it will cheerfully draft a sentence about a disability category that does not match your state's criteria.
Pricing: Free forever with 80-plus tools; Plus at $8.33 a user a month billed annually or $12.99 billed monthly; Enterprise on quote with a custom data privacy agreement (pricing). Checked September 21, 2026.
Safety: MagicSchool's FAQ states it is "FERPA and COPPA compliant," that "student data is encrypted and never used to train AI models," and that data "is stored and processed in U.S.-based, FERPA-aligned data centers" (FAQ). The directory's own record leaves FERPA, COPPA, hosting and minimum age unstated, which is a gap in our record rather than in the vendor's disclosure. An independent 9ine assessment exists. Write recommendations in it from de-identified notes; do not paste a protocol into it.
FrenalyticsEDU is the best fit for IEP goal data, because it collects while the student works
Every annual review has the same weak point, which is the evidence. FrenalyticsEDU is built for special education, English learner and MTSS caseloads, and its selling proposition is automated progress monitoring: the student works through personalized sessions, and the platform records the performance data you would otherwise be tallying on a clipboard. Its EDU Extras tier reports individual question-level data specifically so it can be mapped to IEP goals (pricing).
For a school psychologist this is the tool that pays off in the spring rather than the fall. It is also the only one here that answers the third question in the framework above the right way round.
Watch out for: the free plan covers 10 student profiles and one staff account, which is a caseload sampler rather than a caseload. The paid tiers are priced per account, not per district, so a team of psychologists sharing one login is a data governance problem waiting to be discovered. MTSS, if you have not met the abbreviation, is the multi-tiered system of supports most US districts use to escalate intervention before a referral.
Pricing: Free with 10 student profiles and one staff account and no expiry; EDU Essentials $14.99 a month for 15 profiles and two staff; EDU Extras $16.99 a month for 20 profiles, five staff and question-level data, with extra students at $1 each; Enterprise for rostering and single sign-on on quote (pricing). Checked September 21, 2026.
Safety: our directory record lists FERPA compliance, US hosting and no third-party sharing. The vendor's own privacy policy, last updated February 21, 2025, says something more careful than that: it commits to taking "all steps reasonably necessary to protect the privacy of student data," and then states that unless a separate agreement such as a data privacy addendum says otherwise, personal data "may be transferred to, and maintained on, computers located outside of your state, province, country or other governmental jurisdiction" (privacy policy). It names Illinois SOPPA, New York Education Law 2-d and Tennessee SOPIPA compliance and offers data privacy agreements on request. There is no independent 9ine assessment. So the US hosting in our record is the default rather than a guarantee, and this is the tool on the list that most clearly needs a district agreement rather than a personal signup, whatever the free tier makes possible.
Diffit is the best fit for the family summary, because it rewrites at a reading level you choose
An eligibility report is written for a legal record and read by a parent. Those are different documents, and the second one is usually not written at all. Diffit takes a text and produces it at a reading level you select, with summaries, vocabulary and question sets, which makes it the fastest route from a finished report to a two-page summary a family can take home and argue with.
Use it on your own prose, not on the student's data. You paste the paragraph you wrote about working memory; it gives you the same paragraph at a sixth-grade reading level. Nothing about the student needs to go in.
Watch out for: the free plan only keeps access to resources you created in the last 90 days and locks export to Google and Microsoft formats behind the paid tier, so a free account is a drafting surface rather than an archive. The directory record does not state where Diffit hosts data. It will also flatten technical precision if you let it, and some of that precision is legally load-bearing.
Pricing: free Basic with 90-day access and no Google or Microsoft export; individual teacher $14.99 a month or $149.99 a year; Diffit for Schools is a flat annual subscription tiered by enrollment with no public figure (pricing, individual plan). Checked September 21, 2026.
Safety: our directory record leaves FERPA, COPPA, hosting and minimum age unstated, which undersells the vendor. Diffit's own privacy page states "we do not collect ANY student data," says the product is FERPA and COPPA compliant, and lists the company as a signatory of the National Data Privacy Agreement and of the Student Data Privacy Consortium's statewide agreements (privacy and security). For the use described here, where the only text is your own writing, the exposure is low. That changes the moment you paste a student's essay.
Glint is the best fit for the supports you recommend, because it makes them in under a minute
A report that recommends a visual schedule and a communication board leaves a teacher to build them. Glint generates visual supports, communication boards, visual schedules and social stories, exports to PDF in one click, and works in more than 50 languages on an iPad or Chromebook with nothing to install (product site). The vendor puts average generation time under 20 seconds against 20 to 60 minutes by hand, and says the platform reaches more than 300,000 learners.
The reason this belongs on a school psychologist's list rather than a teacher's is that it closes the gap between what you write and what happens. A recommendation a classroom teacher can act on before Friday is worth more than a better-worded one they cannot.
Watch out for: the vendor's own figures are the vendor's, not an independent measurement, and there is no 9ine assessment and no rating or review in the directory yet. The pricing page lists a trial and free shared boards for families without spelling out what the trial includes. The minimum age on the directory record is 18, which tells you the account is yours, not a student's.
Pricing: $18 a month or $150 a year, with enterprise pricing for schools and districts on quote; a trial and free access to shared boards for parents and families (pricing). Checked September 21, 2026.
Safety: the directory records Glint as declaring FERPA and COPPA compliance, hosting in the US, and offering deletion on request, and the vendor describes the product as "FERPA & COPPA aligned" on its own site (product site). Its privacy policy link currently resolves to the homepage rather than to a policy document, checked September 21, 2026, so the declaration is the vendor's word without a document behind it. Build supports from the functional description, not from the student's file, and nothing identifiable needs to enter it.
Read&Write is the best fit for the accommodation you want to trial, because it is free for K-12 teachers
Naming an assistive technology accommodation in a report is easy. Knowing whether it will help this student is not. Read&Write is the text-to-speech, word prediction and reading support toolbar that has been in special education rooms for years, and its publisher now makes it free for K-12 teachers, which means you can put it in front of a student during an observation window instead of writing it in as a guess (pricing).
Two naming notes, because both will confuse a search. The product now sits under the Everway brand rather than Texthelp, and texthelp.com still redirects to everway.com, checked September 21, 2026; neither page says when the change happened, so both names point at the same product. And Read&Write is not the same thing as the AI tools elsewhere on this list: most of what it does is decades-old assistive technology, which is exactly why it is worth trialing.
Watch out for: the free K-12 teacher license is for you, not for a class set, so a whole-grade rollout is the $2.75-a-seat district purchase. The directory record leaves hosting and minimum age unstated. And the tool only tells you something if you watch the student use it; installing it proves nothing.
Pricing: free premium subscription for K-12 teachers; individual license $184 a year; K-12 District $2.75 a seat with a minimum purchase; K-12 Group $18.30 a seat with a 150-seat minimum, all billed annually (pricing). Checked September 21, 2026.
Safety: the directory record leaves FERPA, COPPA, hosting and minimum age unstated, so the vendor's own privacy page is the check; texthelp.com/privacy now redirects to Everway, whose privacy page displays FERPA, COPPA, GDPR, ISO/IEC 27001 and SOPPA badges and links the policy documents themselves (privacy). It carries an independent 9ine assessment, one of only five in the special education category. Because students use it directly, this belongs on the district's approved list rather than on a personal account.
StaffDraft is the best fit for the message home, because it checks for identifiers before it drafts
StaffDraft turns rough notes into school emails, letters and references, and it does one thing no other tool here does: it runs a personal identifier check before generation, and it detects language that suggests a safeguarding concern, blocks the draft, and tells the user to contact their designated safeguarding lead instead (product site). It also states that drafts are not retained after processing.
For a role where the most common data breach is a tired professional pasting too much into a text box at 6pm, a tool that refuses the paste is doing more for you than a tool that encrypts it.
Watch out for: it is British, priced in pounds, and its vocabulary is UK school vocabulary, so the safeguarding routing will name a role your district does not have. The free tier is 10 drafts a month. There is no independent security assessment and no rating in the directory yet.
Pricing: free for 10 drafts a month with all six tools and the identifier check; Individual £6 a month for unlimited drafts; an organization-wide plan at £5 a user a month with a 20-user minimum; pay-as-you-go credits from £2.99 (product site). Checked September 21, 2026.
Safety: the directory records UK hosting, an 18-plus minimum age, deletion on request, and FERPA marked not applicable, which is accurate for a tool that is not meant to receive student records at all. The vendor states drafts are not retained after processing and advises removing student names before generating.
What about ChatGPT, Gemini Notebook and the other tools everyone asks about?
Short verdicts, because these are the names that come up in every district conversation and most of them are the wrong answer for this job.
ChatGPT is the one school psychologists already use and the one the survey evidence flags as the risk. It will rewrite your paragraph well. A consumer account is not under your district's direct control in the sense the Department of Education's vendor guidance requires, so it fails the school-official test the moment a name or a score goes in. If your district has bought an enterprise agreement, that is a different product with different terms, and you should know which one you are logged into.
Gemini Notebook, the tool Google renamed from NotebookLM in July 2026, is genuinely good at holding a set of documents and answering questions across them, which sounds exactly like a cumulative file review. Same objection: the documents you would want to load are the ones you are least free to upload.
MagicSchool and Diffit both appear above, on the writing side of the line, and neither should see a protocol.
Khanmigo and Quizlet are student learning tools. They have nothing to do with an evaluation and appear here only because AI answer engines keep listing them under every education query.
9ine is not a productivity tool at all; it is the vendor assessment service behind the independent security column in our directory, and if your district is choosing between three products for a caseload of real children, the security assessments page is a better starting point than any feature comparison.
What this looks like in one evaluation
A third-grade initial evaluation for a suspected specific learning disability in reading, with parental consent signed on a Monday. Federal regulation gives the district 60 days from that consent, or a shorter state timeframe where one exists (34 CFR 300.301). Here is where each tool sits in those 60 days, and what never leaves your own hands.
Week one, the file review. You read the cumulative file, the intervention records and the teacher's referral yourself. No tool. This is the part where you notice that the reading intervention started in January and the attendance dropped in November, and nothing that summarizes documents will notice it for you.
Week two, the observation and the testing. Also no tool. You write field notes by hand or in your district's system.
Week three, progress data. If the student has been working in FrenalyticsEDU during intervention, the question-level data is already there and you pull it rather than reconstruct it. If not, this is the week you learn why it matters, and the next student on your caseload starts in it.
Week four, the draft. Scores, notes and observations go into PAR's AI Report Writer, under your district's written agreement, and you get a structured draft back. You set it to the technical tone for the record and generate a second pass at the parent-friendly setting for the summary. Then you read every sentence, because 94 percent of the psychologists in that survey did and the 6 percent are not a group to join.
Week five, the recommendations. The instructional recommendations get drafted in MagicSchool from a de-identified description of the profile: "a third grader with average verbal reasoning, low phonological processing and slow rapid naming." No name, no scores, no file. You take what is useful and cut what is generic.
Week five, the supports. Anything you recommend that a teacher would have to build, you build in Glint and attach. Read&Write goes in front of the student for two sessions before you write it into the accommodations, so that the report says what happened rather than what might.
Week six, the family. The parent-friendly section goes through Diffit to land at a reading level the family can use, and the cover email is drafted in StaffDraft, which will stop you if you put a name in it.
The eligibility meeting. Nothing on this list attends. The determination is made by the team, from the data, in the room.
What can AI legally do in a special education evaluation?
Less than the vendors imply and more than most districts currently allow, and the gap between those two is where school psychologists are improvising.
The clearest written line so far is the DC model policy of September 1, 2026, which sorts AI uses into categories and puts determining eligibility for an IEP or a Section 504 plan among the prohibited ones, alongside physical surveillance of students and staff, student discipline decisions and evaluations of teacher performance. Drafting IEP language, reviewing and grading student work and coaching educators sit in the restricted category, allowed with safeguards. The same policy requires staff to complete training before using AI, to renew it annually, and to "maintain human oversight and accountability for all outputs" (OSSE release).
Underneath that sits FERPA, which does not mention AI at all and does not need to. A vendor receiving education records has to satisfy the four school-official conditions quoted above, of which "direct control" is the one a consumer subscription cannot meet (Vendor FAQ). The US Department of Education's July 2025 guidance on AI in schools stresses engaging parents in decisions about adopting new technologies and attending to user privacy, without setting rules specific to evaluation (press release).
So the practical answer, in one sentence: AI may draft what you would have written and organize what you have already collected, and it may not decide who qualifies, what the disability is, or what a score means.
What the research actually found
The most useful evidence on this so far is a national survey of school psychologists using AI at work, published in the Journal of Applied School Psychology. Among the 100 respondents, 45 used AI in report writing. Sixty-nine percent used it to rewrite sections for clarity and 67 percent to generate recommendations. Few used it for interpretive or diagnostic purposes, and none relied on it to generate an entire report. Ninety-four percent edited the output before use. Those using it for report writing reported saving 6.3 hours a week, with a 95 percent confidence interval of 4.6 to 8.1 hours, more than double the roughly three hours reported by those using AI for other tasks (abstract).
The same study found the other half of the picture, and it is the reason this post spends so long on the line between drafting and deciding: some practitioners entered student-identifying information into AI platforms, which would breach FERPA, and some used AI for diagnostic or eligibility purposes for which it has no empirical validation. The authors' conclusion is that formal oversight is needed rather than reliance on individual editorial care.
Read that as a description of a profession under time pressure rather than as a scandal. The National Association of School Psychologists recommends one school psychologist per 500 students and puts the current national ratio at 1:1127 (NASP). A ratio of that size is what makes a 6.3-hour saving worth a risk somebody should not have to take alone.
Three jobs to keep off AI
Interpreting a score. A standard score of 82 on a processing speed index means something different for a student who has been in three schools this year than for one who has not. The number is the same and the meaning is not, and no tool on the market has access to the difference.
Deciding eligibility. Prohibited in the DC model policy, unvalidated in the research, and, more simply, it is the one part of the job the law gives to a team of people who know the child.
Talking to a family about a disability. The summary can be drafted. The conversation cannot. A parent hearing the word dyslexia for the first time is not reading a document, they are watching your face.
Where this leaves you
Pick the cheapest of the seven and use it on a report you have to write anyway. If your district has no written agreement covering AI in special education documentation, that is the more urgent job, and it is worth taking to your special education director this month rather than after the first complaint. The directory's special education and accessibility category lists all 29 tools with their pricing and data handling side by side, and our guides to AI tools for IEP writing and AI tools for speech-language pathologists cover the two neighboring roles in more depth. AI tools for school counselors covers the other confidential caseload in the building.
The work that made you want this job was never the writing. Give that part away carefully, and keep the part where you sit with a nine-year-old and work out what is actually going on.
Dan Fitzpatrick is The AI Educator. Read more from him at aieducator.tools.
