A student hands in a clean, well-argued essay. You have a feeling. You run it through a detector, the detector says 74% AI, and now you have a feeling and a number, neither of which you can defend in a meeting with a parent. That is where a lot of teachers are in 2026, and it is a dead end. The question "did a machine write this?" has no reliable answer. The question "can you explain this to me?" always does.
This post is for teachers and department heads in middle and high school who want assessment that holds up now that most of their students use AI for homework. It covers six tools that ask students to say what they know, out loud, and it is honest about which ones are ready for a school and which are not. If you want a detector, this is not that post. I explained why in my piece on whether AI detectors work.
How This List Was Built
I started with the 44 tools in the directory's assessment and grading category and kept only those whose core job is a spoken or explained response rather than a written product. That left four. I added two from the special education and accessibility category that do the rehearsal side of the same job, because you cannot assess a student's speaking if they have never been given a safe place to practise it.
The criteria: does the tool make the student's reasoning visible, does it keep the judgement with the teacher, is the safety posture documented, and is the pricing honest. Nobody pays to appear in the directory or in this post, and vendors cannot buy a place on this list.
What I left out, and why. Sherpa, the Socratic oral assessment tool that came out of Georgia Tech, and FeedbackFruits, which runs oral assessment for universities, are not listed in the directory, so I have no verified data on either and will not guess. Modality Learning and Brisk Teaching show process rather than performance, and I covered them in the detectors post. Every AI detector is out on principle. Knowt and LearnForge have voice features, but voice is not their job.
Pricing and features verified 4 September 2026.
By the Numbers
Oral assessment is a small, young corner of the market, and the data says so. Of the 44 assessment and grading tools in the aieducator.tools directory, only 7 carry an independent 9ine security assessment, 7 do not state where they host student data, and 15 do not say whether they share data with third parties. Only 19 (43%) declare FERPA compliance and 15 (34%) declare COPPA. Four of the six tools below launched in 2025 or 2026. That is not a reason to avoid them. It is a reason to read the safety line under each one.
Comparison Table
The privacy column is the vendor's declaration, not an audit. The final column is the independent one.
| Tool | Best for | Pricing | Grades | Min age | Privacy (declared) | Data hosted | Independent security assessment |
|---|---|---|---|---|---|---|---|
| SayVeritas | Asynchronous oral exams, 6-12 | Paid, credit packs | K-5, 6-8, 9-12, Post-sec, Higher ed | 8+ | FERPA, COPPA, GDPR | US (vendor notice; directory says EU) | No |
| Vocina | Viva questions on submitted work | Freemium | 6-8, 9-12, Post-sec, Higher ed, Adult | 13+ | FERPA (GDPR in progress) | Australia | No |
| Snorkl | Spoken reasoning in math and science | Freemium | Not stated | Not stated | FERPA, COPPA (vendor trust page) | Not stated | Yes (9ine) |
| Readling | Reading fluency, elementary | Paid | K-8 | No minimum | GDPR | UK | No |
| EchoGuide Pro | Low-stakes speaking rehearsal | Paid, 7-day trial | All grades | 4+ | FERPA, COPPA, GDPR | UK | No |
| TikoNote | Students explaining to learn | Freemium | 6-12, Higher ed | 13+ | FERPA, COPPA, GDPR | US | No |
Best for asynchronous oral exams in grades 6 to 12: SayVeritas
SayVeritas is the closest thing on this list to an oral exam you can set for a whole class and grade by Tuesday. Students record spoken answers on any device, the platform asks a Socratic follow-up when an answer is thin, and two scoring agents grade the transcript against a Bloom's-aligned rubric with evidence snippets attached, so the teacher sees why a score was given rather than just the score.
The evidence for the approach is fresh. A 2026 field report from NYU Stern ran voice-AI oral exams across two cohorts and found grading agreement between models rose to 0.86 and 0.90 after the models compared reasoning, at a cost of between $0.29 and $0.96 per exam. The same report found 63% to 83% of students called it more stressful than a written exam, and about two-thirds said it tested their actual understanding. Both halves of that finding matter. (Ipeirotis and Rizakos, NYU Stern, 2026)
What sets it apart from Vocina is that it is built for schools from grade 6 up, with a stated minimum age of 8, whereas Vocina leans tertiary. Watch out for the hosting question: the directory record says EU, but the vendor's own privacy notice says its servers are in the United States and audio is retained for 30 days. Ask before you sign, and I have flagged the record for correction. Best for a department that wants a repeatable oral component on every unit. Not for elementary, or for a teacher who wants the AI's score to be the grade.
Pricing: Starter Pack $15 one-time for 250 assessment credits, Classroom Pack $29 one-time for 600 credits, school licence quote-based per year. A credit is one scored student response, so the Classroom Pack works out at roughly five cents a response. No free tier (verified 4 September 2026).
Safety: FERPA and COPPA declared, minimum age 8, hosting stated as US by the vendor, audio retained 30 days, no independent 9ine assessment, GDPR declared.
Best for viva-style questioning on submitted work: Vocina
Vocina turns a piece of written work into a five to ten minute spoken conversation about that work, which is the viva model universities have used for a century, made cheap enough to run every week. A student uploads their essay or project, the agent generates questions pitched at the level of that specific piece, then probes: explain this idea, justify that decision, apply the concept somewhere else. The teacher gets audio, a transcript, rubric-aligned feedback and an indicative understanding score.
A New Zealand vocational provider that redesigned its assessments with Vocina reported that spoken responses revealed reasoning that polished written work had hidden, and that transcripts made moderation easier, not harder. Their academic lead put it well: the point is not replacing writing with technology, it is designing assessment that captures thinking. (Scarlatti case study)
Against SayVeritas, Vocina's edge is the customisable agent: you write the prompt and the criteria, so a history viva and a chemistry viva behave differently. Watch out for the data record. Vocina declares that it collects student data and shares it with third parties, hosts in Australia, and lists GDPR as in progress. For a US district that is three questions for the vendor before anyone under 13 goes near it, and the minimum age is 13 anyway. Best for high school and post-16 teachers assessing coursework, essays and projects. Not for younger students or for a district that needs US hosting.
Pricing, in New Zealand dollars: free tier of 15 assessment minutes a month (around three short assessments) for one learner; Starter NZ$19 a month for 100 minutes; Standard NZ$65 for 350 minutes and five tutors; Pro NZ$199 for 1,200 minutes; enterprise quote-based. Unused minutes roll over for up to a year (verified 4 September 2026).
Safety: FERPA declared, COPPA not applicable, minimum age 13, hosted in Australia, student data collected and shared with third parties by the vendor's own declaration, no independent 9ine assessment, GDPR in progress.
Best for hearing the reasoning in math and science, including under-13s: Snorkl
Snorkl is the tool I would put in front of an elementary or middle school teacher first, because it captures a student talking through a problem while they draw it. The student records voice, whiteboard and text together; the AI gives instant feedback against criteria the teacher wrote and flags misconceptions across the class. It was built for the moment a child says "I got 12" and you want to know how.
The evidence here is the directory itself: Snorkl is the only tool on this list with an independent 9ine security assessment, and it holds a 4.75 rating from 12 educator reviews. Teachers in the directory's assessment discussion thread describe redesigning homework around "show me your thinking" because recall is easy to fake and method is not, and Snorkl is the tool that makes that redesign practical.
What sets it apart from SayVeritas is the drawing: for a maths proof or a science diagram, the pen is half the explanation (marking, for UK readers, gets faster because you are listening for one thing). Watch out for the directory gaps. The card is blank on FERPA, COPPA, hosting and minimum age, so I checked the vendor's trust page, which states the service is governed by FERPA and COPPA and carries iKeepSafe COPPA and FERPA badges; hosting location is still not stated anywhere I could find. Best for elementary and middle school maths, science and languages. Not for long-form essay defence.
Pricing: free teacher tier with unlimited classes but a limited number of activities; Premium for schools and districts adds the coach, dashboard and PD, quote-based (verified 4 September 2026).
Safety: FERPA and COPPA per the vendor's trust page (directory record blank), minimum age not stated, hosting not stated, independent 9ine assessment held, GDPR not stated.
Best for reading fluency in elementary: Readling
Readling is a reading-aloud platform where pupils read from any book and get instant AI feedback on fluency and expression, and it belongs here because reading aloud is the oldest performance assessment in the building. Teachers get a dashboard of progress, quick quizzes and pronunciation guidance; parents get a QR code so the same reading counts at home.
The evidence is thin but positive: it launched in 2026, holds a 5.0 rating from 8 directory reviews, and lists a dyslexia-friendly font, offline mode and high contrast among its accessibility features, which is more than most tools on this list bother with. What sets it apart from EchoGuide Pro is that it assesses rather than rehearses: there is a score and a record. Watch out for the market fit. It is built for UK primary schools, prices are not public, and FERPA and COPPA are marked not applicable, so a US elementary school is buying a product that was not designed with its laws in mind. Best for K-5 reading fluency, and any school in England. Not for older students or for anything beyond reading.
Pricing: paid, no public pricing; quote from the vendor (verified 4 September 2026).
Safety: FERPA and COPPA not applicable per the vendor, no minimum age, hosted in the UK, collects name and performance data, no third-party sharing declared, no independent 9ine assessment, GDPR declared.
Best for low-stakes speaking rehearsal: EchoGuide Pro
EchoGuide Pro is not an assessment tool, and that is the point: it is where a nervous student rehearses before the assessment. Real-time listening highlights the next word, guides pace and clarity and gives fluency feedback, with a dyslexia-friendly mode and a low-anxiety design. If you are going to ask students to defend their work out loud, you owe the quiet ones a place to practise that is not in front of the class.
What sets it apart from Readling is scope: it runs from age 4 to adult and across subjects, while Readling is reading only. Watch out for two things. Pricing is not published, and the product launched in 2026, so evidence of use is limited to the vendor's own claims. Best for students with speech anxiety, English learners and anyone preparing a presentation. Not for grading anything.
Pricing: paid, no public pricing; 7-day free trial listed on the directory page (verified 4 September 2026).
Safety: FERPA and COPPA declared, minimum age 4, hosted in the UK, collects no student data by the vendor's declaration, no independent 9ine assessment, GDPR declared.
Best for students who need to explain it to learn it: TikoNote
TikoNote is a study app built on the Feynman method: the student turns notes or a PDF into a topic, then explains it in plain language and gets told where the explanation broke down. It is the rehearsal room for every other tool on this list, and it sits with the student, not the teacher.
The reason it is here is the RAND finding from March 2026 that 62% of American students from middle school to college now use AI for homework, up from 48% seven months earlier, while 67% believe heavier AI use harms critical thinking. (RAND American Youth Panel, 2026) Students know what the shortcut costs them. A tool that makes them explain rather than paste is AI use they can defend.
What sets it apart from Knowt is the single-mindedness: no flashcards, no games, just explain it. Watch out for the age line, 13 and up, and for the fact that pricing is not published on the site. Best for high school and college students revising for exams (revision, for UK readers) and for teachers who want to point students at something better than a summariser. Not for under-13s or for classroom assessment.
Pricing: freemium; no public pricing on the website; app store listing shows in-app purchases (verified 4 September 2026).
Safety: FERPA and COPPA declared, minimum age 13, hosted in the US, collects name and email only, no independent 9ine assessment, GDPR declared.
How to Actually Choose
Start from the assessment, not the tool. I use three Ps: Product is the essay, Process is how it was made, and Performance is the student showing you they own it, live or recorded. AI has made the first one cheap to fake. The second is covered by process tools. This list is the third. So the rule is: for any task where the product could have been generated in ten seconds, add a performance component worth at least a third of the grade, and let the student know in advance.
Then apply cognitive stretch, which is the amount of thinking a task asks of the student after the tool has done its part. A Socratic follow-up that asks "why does that step work?" stretches. A spoken read-back of the essay does not. Write the rubric before you pick the tool; the rubric generators post covers that job.
Three if-then rules a department head can act on. If your students are under 13, start with Snorkl and keep the others for later, because minimum ages and COPPA are not suggestions. If you assess coursework in high school, run Vocina's free tier on one class for a month before buying anything. If you need an oral exam for a whole grade with evidence attached to every score, SayVeritas is the one built for it, once the hosting question has a written answer.
A detector tells you what a machine might have done. An oral assessment tells you what the student can do.
What I'd Do This Week
Pick one class and one assignment that is already set. Add a two-minute spoken explanation on Snorkl or the Vocina free tier: "Explain your main argument and the one place you nearly got it wrong." Do not grade the audio this first time. Listen to five of them and ask yourself whether you learned something the essay had not told you. If you did, you have your evidence for the department meeting. If you did not, the task needed more stretch, not more software.
The Bottom Line
The way to assess students who use AI is to ask them to explain what they know, and in 2026 that finally scales: SayVeritas for asynchronous oral exams in grades 6 to 12, Vocina for viva-style questioning on submitted work, Snorkl for spoken reasoning in math and science and for anyone under 13, with Readling, EchoGuide Pro and TikoNote covering fluency, rehearsal and study. Check the safety line before you buy, because only one of the six has an independent assessment. Compare these against the rest of the assessment and grading tools in the directory, and read the security assessments page before any of them meets a student. Then go and listen to a child explain something. It is still the best assessment there is.
Dan Fitzpatrick is a Forbes contributor, three-time bestselling author, founder of The AI Educator and has trained more than 150,000 educators across 30+ countries. Read more about Dan.
