AI tools for ESL teachers: 5 that earn their place in 2026

Every ESL conference now has a wall of booths promising AI will transform your classroom. Most of it is a chatbot with a lesson-plan skin. This list is shorter and more practical: five tools (well, four tools and one honest category), chosen by a single test — does it give you back hours, or does it give you another dashboard to feed? Prices are approximate as of 2026; free tiers noted.

1. sayit — speaking homework that grades itself

The biggest time sink in teaching speaking isn't teaching it — it's assessing it. Listening to 30 recordings, diagnosing each student's sound errors, writing individual feedback: several minutes per student, per assignment. So most of us assign less speaking homework than we know students need.

sayit's teacher workflow attacks exactly that. It looks like this:

1. 1.Write a paragraph — this week's vocabulary in context, the unit dialogue, anything. Or import a PDF of your own materials, or let the AI generate a leveled passage.
2. 2.Assign it to a class in one click — or differentiate: send the simpler paragraph to one batch, the harder one to another.
3. 3.Watch the gradebook fill in. As students record, you get completion rates, average scores, and a per-student, per-phoneme breakdown: not "Minh got 71%" but "Minh substitutes /s/ for /θ/ and drops final consonants." CSV export when you need it for reports; automated guardian emails if your program uses them.

Students get the consumer-grade feedback loop: record a sentence, see a color-coded per-word heatmap, tap a red word to see the target IPA next to the sound they actually made, plus a concrete articulation tip. The recognizer is language-model-free — it scores raw sounds instead of auto-correcting mistakes into dictionary words the way normal speech-to-text does, which is the difference between grading pronunciation and grading luck. It's also accent-tolerant: students aren't marked down for sounding Vietnamese or Brazilian, only for contrasts that change meaning.

Rostering is CSV import or join codes; there are IELTS/TOEFL-style exams for test-prep classes; privacy-sensitive programs can run it on-premise. Individual use is free in the browser, no card — try it as a student before you pilot it as a teacher. The full buyer's checklist for school plans is in our guide for language schools.

Honest limit: sayit grades sounds, fluency, stress, and completeness — not the pragmatic quality of a student's argument. Discussion skills are still yours to teach.

2. ELSA for Schools — gamified drills with a teacher dashboard

ELSA Speak is the biggest name in pronunciation apps (25M+ users, 8,000+ lessons across all 44 phonemes), and its Schools version gives teachers a dashboard over the same content, with CEFR level prediction built in. Where it shines: daily engagement. The gamified streak-and-score loop genuinely gets teenagers drilling on the bus, and the lesson library means zero teacher prep.

The flip side of zero prep is zero control: students drill ELSA's curriculum, not yours, so it runs alongside your syllabus rather than inside it. Feedback is percentage-scores per phoneme rather than a diagnosis, and the model targets a native-like accent. Consumer Pro runs ~$12/mo; school pricing is quote-based. Good fit if you want a motivation engine more than an assessment engine — more detail in sayit vs ELSA Speak.

3. YouGlish — real listening examples, free

Not new, not fancy, still one of the highest value-per-minute tools in ESL. Type any word or phrase and YouGlish plays real YouTube clips of real speakers saying it, jump-cut to the exact moment — thousands of examples across American, British, and Australian English.

Classroom uses: settle "how do natives actually say this" debates instantly; show a word's stress in connected speech rather than dictionary isolation; build listening micro-tasks ("write down the three words after 'nevertheless'"). It's listening-only — no feedback, no tracking — and completely free. Pair it with a production tool: hear it on YouGlish, then check students can say it (a quick lookup on sayit's pronunciation hub closes that loop).

4. Talkpal — conversation roleplay for homework

Talkpal does AI roleplay conversations — job interviews, hotel check-ins, debates — in 80+ languages, at ~$10/mo with 10 minutes a day free. For conversation quantity, it's hard to beat as homework: shy students who freeze in class will happily argue with a robot at midnight.

Know what it is and isn't. Feedback is transcript-based, so it inherits speech recognition's auto-correct problem: if the engine understood the student, the student passes — many pronunciation errors are invisible to it. Use it for fluency and confidence reps, not for pronunciation assessment. The free daily allowance is enough for meaningful homework without a school contract.

5. ChatGPT / Claude — the materials department you never had

Not an ESL product, but the general-purpose assistants have quietly become the biggest time-saver in most teachers' weeks. What they're reliably good at:

• Leveling texts: "Rewrite this article at CEFR A2, keep the topic vocabulary."
• Generating variations: ten more sentences drilling the past perfect, a cloze from any passage, three role-play cards per unit.
• Differentiation at speed: the same worksheet at three levels in one prompt.

Two cautions from hard experience. First, verify before you print — these models still occasionally invent a "common idiom" no native speaker uses or mislabel a grammar point, and they'll do it fluently. Second, don't trust them for pronunciation specifics: a text model can print /θ/ but cannot hear your student. Generate materials with a chatbot; assess speech with a tool that actually listens to phonemes.

What AI still can't do

A list like this reads vendor-first unless it's honest about the boundary, so: the parts of the job no tool on this page touches.

• Motivation. An app can streak-ify practice; it cannot notice that Amara stopped trying after her presentation went badly and have the two-minute corridor conversation that fixes it. Feedback works when a student cares about the person giving context to it.
• Culture and pragmatics. When "please" sounds rude, when directness reads as aggression, how small talk differs in Osaka and São Paulo — this is judgment about people, taught by a person who has misread rooms in a second language themselves.
• Reading the room. No dashboard tells you the class is lost but nodding. Skilled improvisation off a failing lesson plan remains the core professional skill.

The realistic framing: AI tools take over the mechanical layers — scoring sounds, generating variations, sourcing examples — and hand you back the hours to do the human layers better. The teachers getting the most from these tools aren't teaching less; they're grading less.

Start with the biggest time sink

If speaking assessment is what you've been rationing, that's the place to start. Try sayit free in the browser — record a sentence yourself and see the phoneme heatmap your students would get, no card required. Then, if it fits your program, book a demo for the class-management side (school pricing differs from individual plans). One paragraph, one class, one week: you'll know by Friday whether the gradebook filling itself in is worth it.