It is July 5, 2023, and I have spent the first three weeks of summer break doing something my colleagues find either admirable or deeply concerning: I have been systematically testing every AI tool I can find that has potential educational applications. I have logged over 120 hours across more than forty platforms, and I want to share what I have found before the impressions blur and the landscape shifts again - which, at the current pace of AI development, could happen by next month.

A note on methodology: I tested each tool against five criteria - pedagogical utility, accuracy, ease of use, student appropriateness, and cost sustainability. I was not looking for perfection. I was looking for tools that could meaningfully improve some aspect of teaching or learning without requiring an unreasonable investment of time, money, or technical expertise.

The Large Language Models - A Comparative Landscape

The LLM landscape has transformed since ChatGPT's launch seven months ago. When I wrote my first ChatGPT article in December, it was essentially the only consumer-accessible large language model. Now there is a competitive ecosystem.

ChatGPT (GPT-3.5 free / GPT-4 via Plus subscription at $20/month): GPT-4, released in March 2023, represents a substantial improvement over GPT-3.5. It is more accurate, better at following complex instructions, and significantly less prone to the confident fabrication that plagued its predecessor. For educational use, the difference is meaningful - GPT-4 produces lesson plans that are more coherent, generates assessment items with fewer errors, and maintains context over longer conversations. The $20 monthly cost is the barrier; most teachers cannot justify it as a personal expense, and most schools have not yet established processes for institutional subscriptions.

Google Bard: Google launched Bard in March 2023, and its initial release was underwhelming - it made a factual error in its own launch demonstration. However, the integration with Google's ecosystem gives it a unique advantage for educators already working within Google Workspace. Bard can access real-time web information, which addresses one of ChatGPT's key limitations (its training data cutoff). The quality of its educational outputs is inconsistent - sometimes excellent, sometimes oddly off-target - but it is free and improving rapidly.

Claude (Anthropic): Claude impressed me more than I expected. Its responses tend to be more nuanced and balanced than ChatGPT's, and it handles ethical complexity with more sophistication. For generating discussion materials on controversial topics - which is one of my most frequent use cases - Claude produces outputs that acknowledge multiple perspectives more authentically. It also has a larger context window, meaning it can work with longer documents.

Bing Chat (GPT-4 powered): Microsoft's integration of GPT-4 into Bing search provides web-grounded responses with source citations. For research-oriented tasks, this is valuable because it connects AI-generated summaries to verifiable sources. The educational limitation is that Bing Chat is accessed through the Edge browser, and many school districts standardize on Chrome.

Purpose-Built Educational AI Tools

Beyond the general-purpose LLMs, a growing ecosystem of AI tools designed specifically for education has emerged. Here is where I found the most interesting developments.

Diffit (diffit.me): This tool generates differentiated reading materials from any source text, adjusting for reading level while preserving key concepts. I tested it extensively and found it remarkably effective for creating tiered materials for mixed-ability classrooms. It produces vocabulary lists, comprehension questions, and summary activities calibrated to different reading levels. Of all the tools I tested this summer, Diffit may be the one I use most frequently in the fall.

Curipod: An AI-powered interactive presentation tool that generates slide decks with embedded formative assessment activities - polls, word clouds, drawing responses, open-ended questions. The AI generates the initial content, which the teacher then reviews and modifies. The interactive elements transform a static presentation into an engagement tool. I found the AI-generated content to be a solid B+ starting point that required modest teacher refinement.

Khanmigo (Khan Academy): Khan Academy's AI tutor, built on GPT-4, takes a pedagogically intentional approach that distinguishes it from general-purpose chatbots. Rather than simply answering student questions, Khanmigo guides students through problems with Socratic questioning. It refuses to give answers directly, instead asking prompting questions designed to help students discover the answer themselves. This is precisely the kind of pedagogical scaffolding that educational AI should provide. The limitation is that it is currently available only through a pilot program.

MagicSchool.ai: Designed specifically for teachers, MagicSchool offers over 50 AI-powered tools for common instructional tasks - lesson plan generation, rubric creation, IEP goal drafting, parent communication, assessment item writing, text complexity analysis. The interface is designed for educators who are not technically sophisticated, which dramatically lowers the adoption barrier. I found the quality variable across different tools (the lesson planner is excellent; the assessment generator needs more refinement), but the breadth of offerings makes it valuable as a one-stop resource.

Quillbot and Grammarly's AI features: Both writing assistance tools have integrated AI capabilities that go beyond grammar checking into paraphrasing, summarization, and style adjustment. For student writers, these tools occupy an interesting middle ground between independent writing and AI-generated writing. They enhance and refine human-written text rather than generating text from scratch. The pedagogical question - does this help students learn to write better, or does it create a dependency? - remains open.

Image and Multimedia AI

The text generation tools get most of the attention, but AI-powered image and multimedia tools have significant educational potential.

Canva's AI features: Canva, already widely used in schools, has integrated AI image generation (Magic Media) and AI-powered design suggestions. For creating instructional materials, presentation visuals, and student project templates, these features are genuinely useful. The AI-generated images are stylistically appropriate for educational contexts and can be produced from text descriptions - meaning a teacher who cannot draw can now create custom illustrations for any concept.

Adobe Firefly: Adobe's generative AI image tool, launched in beta in March 2023, produces high-quality images from text descriptions. Unlike some AI image generators trained on scraped internet data, Firefly was trained on Adobe Stock images and openly licensed content, which addresses some copyright concerns. For educational materials, the image quality is excellent, though the tool is less accessible to students given Adobe's licensing structure.

Eleven Labs and text-to-speech advances: AI-generated speech has reached a quality level where it is difficult to distinguish from human narration. For creating audio versions of text materials, pronunciation guides for world language classes, and accessible content for students with visual processing differences, these tools have immediate utility.

What I Am Not Convinced By

Honest reporting requires noting what did not impress me.

AI tutoring chatbots that claim to replace human instruction. Several startups are marketing AI tutoring systems as substitutes for human teachers or tutors. I tested four of them and found consistent problems: they fail to identify and address misconceptions, they provide explanations that are technically correct but pedagogically unhelpful, and they lack the ability to read student confusion and adjust in real time. Khanmigo is the exception because it was designed with pedagogical expertise, but most AI tutoring products I tested were built by technologists who understand AI but not learning science.

AI-generated full lesson plans used without modification. Every AI tool that generates lesson plans produces output that looks complete but is pedagogically shallow. The activities are generic, the timing is often unrealistic, and the differentiation is superficial. AI-generated lesson plans are useful as brainstorming tools and first drafts, but any teacher who uses them without substantial modification is delivering mediocre instruction.

"AI detection" services marketed to schools. I tested several AI detection tools against a corpus of student writing and AI-generated text. The false positive and false negative rates remain unacceptably high. Any school relying on these tools for academic integrity decisions is building policy on a foundation of unreliable technology.

What This Means for Fall 2023

As I prepare for the coming school year, I am approaching AI integration with three priorities.

First, I am building a curated toolkit of five to seven AI tools that I will use regularly, rather than trying to use everything. Depth of integration with fewer tools will produce better results than shallow engagement with many tools.

Second, I am designing AI literacy instruction into my curriculum from day one. Students will learn what AI is, how it works, what it can and cannot do, and how to use it critically and ethically. This is not an add-on; it is a foundational skill.

Third, I am connecting with other educators who are doing this work. As a Google for Education Certified Trainer and Coach, I am now positioned to facilitate professional development for colleagues, and I intend to use that role to build a community of practice around AI integration.

The summer of 2023 has been the most intellectually stimulating break I have experienced in nearly three decades of teaching. The tools are imperfect, the landscape is unstable, and the pedagogical questions are largely unanswered. But the potential is real, and the educators who invest in understanding it now will be the ones who shape how it is used.

Dr. Janette Camacho is a K-12 educator with 28+ years of classroom experience, a Google Certified Educator (Level 1 & 2), Google for Education Certified Trainer and Coach, Adobe Creative Educator, and Apple Teacher. She spent summer 2023 conducting an intensive survey of AI tools for K-12 education.