TeachFX and McGraw Hill: Who Consents to the Recording?

The TeachFX McGraw Hill deal, announced September 24, 2026, looks like a routine ed-tech acquisition and contains one question worth a parent’s afternoon. TeachFX is an AI instructional coaching platform that records classroom lessons and gives teachers feedback on how they teach. McGraw Hill now owns it. The feedback also flows, through an integration with Claude for Teachers, into generated lesson plans. Somewhere in that pipeline is a recording of your child’s voice, and the agreement your district signed to permit it was signed with a company that is now a subsidiary.

Key Takeaways

• TeachFX analyses recorded lessons. The subject of the coaching is the teacher; the audio captures students too.
• Voice recordings of a child under 13 are personal information under the FTC’s COPPA Rule, which means consent questions are not hypothetical.
• Recording-consent law is state law, and all-party-consent states treat classroom audio differently from one-party states. Name your state before you argue.
• Your district’s data privacy agreement was with TeachFX, not McGraw Hill. Ask whether it has a change-of-control clause and whether it was reaffirmed.
• The district results on TeachFX’s own site are vendor-reported, not peer-reviewed. The underlying idea, getting students talking more, has better support than the specific percentages.

What was acquired, and what the product does

Instructional coaching is feedback given to a teacher about their teaching practice, traditionally by an observer in the room.

The AI-in-education tracker entry dated September 24, 2026 reports McGraw Hill acquiring TeachFX, described as an AI-native platform that analyses recorded lessons and provides feedback on teaching practices, with an MCP integration with Claude for Teachers that turns that feedback into lesson plans.

TeachFX’s own site describes what it measures: student talk time compared with teacher talk time, academic vocabulary, open-ended questions, specific praise, and standards alignment. It presents results from three districts: a 31% increase in student talk in Iron County School District in Utah, a 28% increase in Lincoln County Schools in North Carolina, and a 10-percentage-point increase in math proficiency at Miles Elementary in Atlanta Public Schools. The page carries no citations to peer-reviewed studies.

Those are the vendor’s numbers, published by the vendor. That does not make them false. It means they have not been through independent review, and a 10-point proficiency jump at a single school has many plausible causes. Hold them loosely.

The thing to understand about the pipeline

Here is the mechanism, because the privacy question only makes sense once you can see the hops.

A microphone in the room captures audio. That audio is transcribed, which converts voices into text. The text is analysed to produce metrics about talk ratios and question types. Those metrics become feedback for the teacher. Then, with the Claude for Teachers integration, the feedback is passed to a language model that drafts lesson plans.

Model Context Protocol, the “MCP” in that description, is a standard way for a tool to hand structured information to a model. It is useful engineering. It also means each hop is a point where data moves between systems, and the number of hops just went up. A parent asking “where does my child’s voice go” is asking a question with five answers now rather than two.

Note what is and is not sensitive at each stage. Raw audio is the most identifiable: a voice is biometric-adjacent and a child’s voice is recognisable to anyone who knows them. A transcript is less identifiable but still contains what your child said. Aggregate talk-ratio metrics are not identifiable at all. A district that retains only the aggregates has a very different privacy posture from one that retains the audio, and that is a question with a one-word answer if you ask it.

Data stage	How identifiable	Which rule reaches it	The question to ask
Raw classroom audio	High. A child’s voice is recognisable	COPPA for under-13s; state recording-consent law	”Is audio retained, and for how long?”
Transcript	Moderate. Contains what your child said	FERPA if maintained as a record about a student	”Are transcripts kept, and are they attributed by name?”
Teacher feedback metrics	Low. About the adult	District employment and evaluation policy	”Is this used for evaluation or only coaching?”
Generated lesson plans	None	Curriculum review	”Does a human approve these before use?”

The acquisition question most parents skip

Districts sign data privacy agreements with vendors. Those agreements name a company, describe what data it may collect, and commit it to deletion schedules and breach notification.

When that company is acquired, the agreement does not automatically evaporate, but the counterparty changes. Many agreements include a change-of-control clause, which may require notice, may permit termination, and occasionally requires renegotiation. Many do not include one at all.

So the single most useful email a parent can send this month asks three things: did the district have a data privacy agreement with TeachFX, does it contain a change-of-control provision, and has anyone reaffirmed it with McGraw Hill since September 24, 2026. The answer tells you whether your district’s privacy office is tracking its own vendor list or filing it.

This is not a TeachFX-specific concern. Ed-tech consolidates constantly, and every acquisition quietly transfers a set of permissions about children to a new owner with different incentives, a different retention policy and a different security posture. The pattern is the story; this deal is just a legible example of it. Our guide to reading a district AI policy covers where vendor lists usually hide.

Where the law actually lands

Three different regimes touch this, and they do not line up neatly.

COPPA. The FTC’s Children’s Online Privacy Protection Rule, whose current text was published in the Federal Register on April 22, 2025, applies to operators collecting personal information online from children under 13. Voice recordings are personal information. In practice, schools often provide consent on parents’ behalf under a school-consent framework, which is lawful and also means you may never have been asked.

FERPA. The U.S. Department of Education describes FERPA as giving parents rights of access, amendment and control over disclosure of personally identifiable information in education records, under 34 CFR Part 99. Whether a coaching recording counts as an education record depends on whether it is maintained as a record about particular students. A recording retained and attributable to your child is a stronger candidate than an aggregate metric about a teacher.

State recording law. This is the one people forget, and it is the one that varies most. Consent requirements for recording conversations are set state by state, with some states requiring all parties to consent and others only one. A classroom in an all-party-consent state is a different legal setting from an identical classroom two states away. Ask your district which analysis it relied on.

What the evidence says about the underlying idea

Strip away the acquisition and the product rests on a defensible premise: classrooms where students talk more, in longer and more analytical turns, tend to produce better learning than classrooms where the teacher talks and students answer in three words.

That premise has support. UCLA’s Bjork Learning and Forgetting Lab lists generation, producing an answer rather than reading one, among the desirable difficulties that impede immediate performance while improving long-term retention. A student explaining her reasoning aloud is generating. A student listening is not. The UK’s Education Endowment Foundation has run a randomised trial of dialogic teaching, an approach built on structured classroom talk, which is the same family of intervention.

And the strongest adjacent evidence remains Nickow, Oreopoulos and Quan’s 2020 NBER meta-analysis, which found a pooled tutoring effect of 0.37 standard deviations with trained adults driving the result. Tools that improve what trained adults do are working on the right variable.

What none of that supports is the specific claim that a software dashboard caused a 31% increase in student talk at a particular district. Measurement changes behaviour, which is partly the point, and also makes clean attribution hard.

What to do at home

Send the three-part vendor email

Did the district have an agreement with TeachFX; does it have a change-of-control clause; has it been reaffirmed with McGraw Hill. Keep it short. One paragraph gets answered; three pages gets forwarded.

Ask whether audio is retained

This is the question with the biggest privacy delta and the simplest answer. If the district retains only aggregate metrics, most of the concern evaporates. If it retains audio, ask for the deletion schedule.

Ask whether it feeds teacher evaluation

A coaching tool and an evaluation tool are the same software with different consequences. Teachers behave differently when recordings count, and a classroom where the adult is being scored is a different place for your child to sit. Districts should be able to state which it is in one sentence.

Treat the lesson plans as drafts

The generated plans at the end of the pipeline are model output. Ask whether a human reviews them before they reach a classroom. Our look at teacher AI use in lesson planning covers how common this already is.

What not to do

Do not treat classroom recording as inherently sinister. Teachers have been observed by other adults forever, and a tool that tells a teacher she is talking 80% of the time is giving her information she cannot get any other way. The objection is not to coaching. It is to retention, attribution and consent happening without anyone naming them, and those are all answerable questions.

What to Watch For Over the Next 3 Months

• Week 4: Look for a vendor-list update or a reaffirmed agreement naming McGraw Hill. Privacy offices usually process these in batches at the start of a term.
• Month 2 red flags: A district that cannot say whether audio is retained. Coaching data appearing in a teacher’s formal evaluation without prior notice to staff. Generated lesson plans used without review.
• Month 3 self-check: Can you state what happens to a recording of your child’s voice from the moment it is captured to the moment it is deleted? Five steps. If you cannot, that is the records request to make.

Frequently Asked Questions

What did McGraw Hill actually buy?

TeachFX, an AI instructional coaching platform that analyses recorded lessons and gives teachers feedback on practice, with an integration into Claude for Teachers that converts feedback into lesson plans. The acquisition was reported on September 24, 2026.

Is my child being recorded?

If your district uses the tool, classroom audio is captured, and that audio includes students. Whether the recording is retained, transcribed or attributed to individual children depends on district configuration, which is why the retention question is the one to ask first.

Does COPPA cover voice recordings?

Voice recordings are personal information under the FTC’s COPPA Rule, which applies to children under 13. Schools frequently provide consent on parents’ behalf under a school-consent framework, so you may never have been asked directly.

Does the privacy agreement survive the acquisition?

It depends on the contract. Some data privacy agreements include change-of-control clauses requiring notice or permitting termination; many do not. Ask your district specifically whether its agreement was reaffirmed with the new owner.

Are the reported results reliable?

They are vendor-reported and not peer-reviewed. Increases in student talk and a 10-point proficiency gain at one school are plausible but unverified independently. The general principle that more student talk helps learning has better support than any specific percentage.

Should I ask the school to stop using it?

A more useful ask is configuration: no audio retention, no individual attribution of student speech, coaching rather than evaluation, and human review of generated plans. Those four conditions address most of the real risk without removing a tool teachers may genuinely benefit from.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Pursuit. (2026). “AI in Education: News, Policies, Innovations” (entry dated September 24, 2026). https://www.pursuit.us/news/ai-in-education-news-policies-innovations
2. TeachFX. “TeachFX: instructional coaching powered by AI.” https://teachfx.com/
3. Federal Trade Commission. (2025, April 22). “Children’s Online Privacy Protection Rule (COPPA).” FTC Legal Library. https://www.ftc.gov/legal-library/browse/rules/childrens-online-privacy-protection-rule-coppa
4. U.S. Department of Education, Student Privacy Policy Office. “What is FERPA?” 34 CFR Part 99. https://studentprivacy.ed.gov/faq/what-ferpa
5. Bjork Learning and Forgetting Lab, University of California, Los Angeles. “Research: Desirable Difficulties.” https://bjorklab.psych.ucla.edu/research/
6. Education Endowment Foundation. “Dialogic Teaching: trial.” https://educationendowmentfoundation.org.uk/projects-and-evaluation/projects/dialogic-teaching
7. Nickow, A., Oreopoulos, P., & Quan, V. (2020, July). “The Impressive Effects of Tutoring on PreK-12 Learning: A Systematic Review and Meta-Analysis of the Experimental Evidence.” NBER Working Paper 27476. https://www.nber.org/papers/w27476