For decodable texts that follow the phonics sequence you teach, start with Project Read or MagicSchool's decodable generator. To make a real article easier to read, use Brisk, Diffit or MagicSchool's Text Rewriter. To hear a child read aloud, use Readling. None of them replaces the check: Education Week has reported since late August that teachers found words in AI-written practice texts that students had not been taught.
| Tool | The job it does | Best for | Price (checked October 12, 2026) | Student data and privacy |
|---|---|---|---|---|
| Project Read | Writes decodable stories matched to a named phonics curriculum | Classroom and intervention teachers following UFLI, IMSE or CKLA (grade range not stated) | Free: 5 standard stories a week; School plan listed at $1,500 (first year $1,000) | Pricing page gives no student-data statement; privacy policy would not load (unverified) |
| MagicSchool | Decodable text generator and Text Rewriter inside a wider teacher toolkit | Teachers who already use MagicSchool for other tasks | Free plan with no numeric caps stated; Plus $8.33 a month billed annually | Custom data privacy agreement offered on the Enterprise plan |
| Diffit | Builds leveled reading passages with vocabulary and questions | Teachers who need several versions of one topic fast | Free Basic plan; no public dollar prices | Vendor states "zero student data collected" and no student logins |
| Brisk Teaching | Changes the grade level of an online article in the browser | Teachers adapting real articles; targets grades 1-12 | Free to install; plan limits unverified | Vendor states inputs are never used to train AI models |
| Readling | Listens to a child read aloud and gives feedback on fluency | Schools that cannot hear every child read as often as they would like | No public pricing; directory lists it as paid | Directory lists UK hosting and no minimum age; vendor page gave no privacy detail |
| General chatbots (ChatGPT, Gemini, Claude) | Write any text you can describe in a prompt | Teachers willing to write detailed prompts and check every word | Free tiers exist; not rechecked for this post | Depends on your district's approved list and the account type |
Which reading job are you trying to do?
Four questions sort most teachers to the right tool. Our view, not a vendor claim: answer them in this order.
- Must every word match a phonics sequence your school teaches? Then you need a decodable generator, not a leveler. Look at Project Read or MagicSchool, and read the Project Read section below before you assume either one knows your sequence.
- Are you starting from a real article or passage that is too hard? Then you need a leveler. Brisk works on a web page in the browser; Diffit and MagicSchool's Text Rewriter work from pasted text or a topic.
- Is the job hearing a child read? Then a text tool is the wrong category. Readling listens and gives feedback on fluency.
- Is none of that quite right? A general chatbot can write almost anything, but Education Week's reporting suggests it is the tool most likely to hand you words you have not taught.
Decodable text and leveled text are different products. A decodable text limits the letter-sound patterns a beginner meets. A leveled text adjusts vocabulary and sentence length to a grade. A text can be at the right grade level and still contain words a first grader cannot sound out.
What has Education Week reported about AI-generated reading texts since August?
Two Education Week stories by Sarah Schwartz frame the question. In "More Kids Are Learning to Read on AI-Generated Text. What Are the Pros and Cons?" (August 27, 2026), a Florida first-grade teacher estimates that 75 to 80 percent of the decodable texts her students read are AI-generated. That is one teacher's estimate, not a survey.
The same story describes a 2025 study led by Linling Shen of the University of Texas at Austin that tested six public language models, including Diffit. Leveled passages came out above the requested grade level, measured by Lexile and Coh-Metrix. Decodable passages contained the requested letter-sound patterns but often read as stilted and included multisyllable words that could trip a beginner. Shen told Education Week that newer versions are more accurate and that detailed prompts help.
The second story, "Using AI-Generated Text in Reading Class? Watch for These Pitfalls" (September 9, 2026), reports the practical side. The teacher Julie Highsmith said general chatbots often got phonics-constrained passages wrong, including words with patterns students had not learned. Thomas Juzek, a computational linguist at Florida State University, warned that rewriting can shift a text's tone or meaning without anyone noticing. Anne Murphy Karabell of American University advised leaders to ask vendors how their tools align with specific curricula.
A preprint points the same way. Huang, Wei and Huang's 2024 paper on leveled-text generation tested 100 educational materials and found, on manual inspection, that rewritten text sometimes introduced misinformation.
Is Project Read the best fit for decodable texts that follow a phonics curriculum?
Project Read is the best fit for a teacher who needs decodable stories tied to a named phonics curriculum, because it is the one tool here built for that job. Its decodable generator page says it is "automatically aligned to major curricula (UFLI, IMSE, CKLA)" and cites "95% average decodability." Both are vendor claims. The page does not explain how decodability is measured, and I found no independent test of it.
Education Week reports the company says at least 100,000 educators use Project Read. The generator also accepts student names and interests as inputs.
Watch out for: The decodability figure is the vendor's own, so treat each story as a draft. Leave student names out unless your district has approved entering them. Project Read is not in the aieducator.tools directory, so I have no directory rating for it.
Pricing: The pricing page, checked October 12, 2026, lists a free teacher plan with up to 5 standard stories a week and no customization. The School plan is shown as $1,500 and then $1,000 for the "first year," without an explanation. The generator page says "unlimited stories for $999/school." Those figures disagree, so ask for a quote in writing.
Safety: The pricing page makes no statement about how student data is collected or used, and the privacy policy would not load when I tried it. That is unverified. Before a school buys, ask for the privacy policy and a data agreement, and confirm what happens to any student names typed into the generator.
Is MagicSchool worth using for decodable texts and text rewriting?
MagicSchool is the best fit for a teacher who already uses it for planning and wants a leveler and a decodable generator in the same place. The directory rates it 4.74 from 19 reviews. Its Text Rewriter page says it turns "one source text into multiple versions at different reading levels," with easy, medium and hard settings. Education Week reports the company says more than 180,000 educators used the Text Rewriter over the past year.
MagicSchool lists a decodable text generator. The page I could load describes the toolkit, not the generator, so how it aligns to a given phonics sequence is unverified.
Watch out for: The Text Rewriter offers three difficulty settings, not a grade-by-grade scale. If you need a third-grade version, you will check the result yourself. Choose Project Read instead if curriculum alignment is the purchase.
Pricing: The pricing page, checked October 12, 2026, shows a free plan for individual teachers and a Plus plan at $8.33 a month billed annually ($99.96 a year) or $12.99 billed monthly. Unlimited generations are listed as a Plus feature, but the free plan states no numeric cap. The page also lists Quizzes as both included and not included on the free plan.
Safety: The privacy page shows a COPPA badge and says customizable data privacy agreements are available; the pricing page puts a custom agreement on the Enterprise plan. Neither page says whether user data trains AI models, so that is unverified. A district still needs to sign its own agreement.
Is Diffit the right choice for several versions of one reading passage?
Diffit is the best fit for a teacher who needs the same topic at several reading levels, with vocabulary and questions attached. Its plans page says the free Basic plan includes differentiated reading passages, vocabulary and comprehension questions. The directory rates it 4.20 from 5 reviews, a small sample. Diffit was one of the tools in the study Education Week describes, and the story notes Diffit did not respond to a request for comment.
Watch out for: Free Basic users lose Google Docs export, longer text inputs and standards alignment, according to the plans page. Diffit's home page does not describe a phonics-constrained feature, so do not use it for decodable practice.
Pricing: The plans page, checked October 12, 2026, shows no dollar amounts. It describes a flat annual school license tiered by school size and an individual teacher license. For both, no public pricing.
Safety: Diffit's home page says "zero student data collected" and "no student logins, no student AI." That is the vendor's claim. Because students never touch the tool, the main exposure is what you paste into it, so keep student work out.
Can Brisk change the reading level of any article?
Brisk Teaching is the best fit for a teacher who finds a good article online and needs a version students can read. The directory rates it 4.77 from 35 reviews. Its reading-level page says it adjusts "any online article or text" to "1st through 12th grade levels" after detecting the original level, through a browser extension.
Watch out for: It levels text; it does not write phonics-constrained text, so it is the wrong tool for a beginning reader. The same page says "more than 58 languages" for translation, while other copy says "over 50," so I would not quote either number.
Pricing: The page offers a free install but does not describe what the free tier includes, and the pricing page would not load when I tried it. Plan limits are unverified.
Safety: Brisk's privacy page says "student and teacher inputs are never used to train, fine-tune, or improve AI models" and that it signed the National Data Privacy Agreement with 25 U.S. states as of November 2025. It states no minimum age. Check whether your state is on the list it publishes.
Does Readling fit if the job is listening to children read?
Readling is the best fit for a school that cannot hear every child read aloud as often as it would like, because the student reads and the app gives feedback on fluency and expression. The directory rates it 5.00 from 8 reviews, a small sample, and lists UK hosting and no minimum age. The vendor describes it as built for UK primary schools.
Watch out for: It does not generate texts. It also takes a child's voice as input, so a school needs to know where the recordings go and for how long. Its home page, as I could load it, did not say.
Pricing: No public pricing. The directory lists it as paid.
Safety: The directory's hosting and age entries come from the listing, not from a vendor policy I could read. Ask for the privacy notice, the retention period for audio recordings, and whether a data agreement is available before any student uses it.
Should you use ChatGPT, Gemini or Claude to write reading texts?
Use a general chatbot only when you will check every word, and know that the researchers Education Week quotes found them weakest at phonics constraints. ChatGPT (4.35 from 23 reviews), Gemini (4.59 from 41) and Claude (4.57 from 21) are the options the ASCD authors Barbara Culatta and Lee Ann Setzer name in "How to Create Custom Decodable Texts in Minutes" (March 2026), along with SchoolAI and MagicSchool. They describe a workflow of about ten minutes per text, split between prompting and editing, and warn that the AI told them "sled" lacks a short-e sound.
Watch out for: The chatbot will not tell you which of its words you have not taught. You supply the word list.
Pricing: Free tiers exist for all three; I did not recheck limits for this post.
Safety: Use the account type your district has approved, and keep student names and work out of the prompt.
What does a grade 1 decodable lesson look like with these tools?
This is a plan, not a record of a classroom. Suppose a first-grade teacher is on a lesson about short a and wants a four-sentence story for a 15-minute small group.
- Write the allowed list first. Short-a words with single consonants and the blends already taught, plus the high-frequency words students have learned, say the, a and is.
- Set the limits in the prompt or the tool's settings. Four sentences, present tense, no names of real children, no words outside the list, and no short-o or short-i words yet.
- Generate, then read every word. Take a draft such as "Sam had a cap and a map. The cat sat on a mat." Every word passes except on, which uses short o and has not been taught. Replace the sentence with "The cat sat. The cat had a hat." and read every word again.
- Read it aloud once as a child would. Cut anything that sounds stilted, because the Shen study found decodable text often did.
- Print it and keep the word list beside it. Next week's text starts from this list.
With Project Read, steps 1 and 2 become menu choices of curriculum and lesson; you still do steps 3 to 5.
What are the four checks before students read AI-generated text?
These checks come from the sources above, not from a vendor.
- Decodability. Read each word against what students have been taught. Culatta and Setzer's "sled" example shows why a chatbot's own assurance is not evidence.
- Level. Do not trust a tool's grade label. Education Week reports leveled passages from general models ran above the requested grade, so check with a separate readability measure and your own read.
- Meaning. Compare the rewrite with the source. Juzek warns that tone and meaning can drift, and the Huang study found introduced misinformation on manual inspection.
- Variety. Education Week reports that researchers worry children who read mostly AI text meet less varied language, though one researcher called the longer-term effects speculative. Keep real books in the mix.
How do these tools compare on cost?
Only two vendors publish dollar prices for a school or teacher plan: Project Read ($1,500 or $1,000 for School, depending on which page you read) and MagicSchool ($8.33 a month on annual billing for Plus). Diffit, Brisk and Readling publish no figure I could verify. A free tier is enough to run the checks above before you ask a school to pay.
For related reading, see our guides to AI tools for intervention teachers, AI tools for students with dyslexia and AI tools for kindergarten teachers.
Prices and plan details checked October 12, 2026 from the vendor pages linked above. Directory ratings are from aieducator.tools on the same date. Education Week, ASCD and the Huang paper are the only research sources; no vendor result in this post was tested by us.
