Best AI Tools for Students in 2026

By Samuel Rose, Founder of Agensi. Published September 28, 2026. Last updated September 28, 2026.

The two pages that rank first for this search each put their own product at number one, and those two vendors are also the sources Google's answer cites most. Nobody independent is winning this search. Agensi has no study tool to sell. What we have is a method, a gate on the one thing that decides whether a research tool is safe to cite from, and a price check against a student paying out of pocket.

The short version. For studying and notetaking, Gemini Notebook, the tool formerly called NotebookLM, because it only answers from what you upload and shows you the passage. Then Quizlet and Notion. For research and citations, Consensus, Elicit, Scite and Semantic Scholar, all of which show the sentence behind a claim, and not Perplexity, which links to pages instead. For writing and editing, Grammarly, ChatGPT and QuillBot, with a section on where editing ends. For tutoring, Khanmigo at $4 a month and StudyFetch. Which tool is best depends on the task, never on one winner.

Start here

Google's own answer ends by asking your academic level and the specific task. Use both.

High school: the free tiers cover almost everything; check first whether your school allows AI on assignments and whether a tool needs you to be 18. Undergraduate: the student discounts start here, and so does the integrity section. Graduate and researcher: the research group is yours, and the gate on source passages is why it exists. By task: summarise a lecture, studying; find and check sources, research; fix a draft, writing; learn something from scratch, tutoring.

How we picked these tools

This is the short form of the Agensi method, the same one on every ranking page we publish. A list without a stated method is an opinion, and an opinion does not get cited.

The longlist had 20 tools: the top search results for "ai tools for students", the tools Google's AI Overview names in its three groups, the student subreddits, and the Agensi collections. A tool stays if it passes four checks. Available: buyable without a sales call, or labelled sales-led. Alive: a release or price change in the last 12 months. On the job: it does its group's job. Evidenced: tested by us, or rated at least 4.0 from 50 or more reviews on a named platform. Student tools rarely clear the last check, because students do not write on G2. So this page uses the method's third tier: a tool that passes every other check is listed for completeness and unscored, and says so. Boterview and Brainly were cut because we could not verify a price or a policy for either.

The gate. A tool in the research and citations group is not listed unless it can show the source passage behind a generated answer, the sentence or figure from the paper, not just a link to the paper. This is a gate and not a change to the weights, which are fixed across every page we publish. It is why Perplexity is not in that group: its citations link to pages, and reviewers report links that go to the wrong ones. It is why four tools with almost no reviews are in it: they show the sentence.

We have not run our benchmark on these tools. The benchmark is one 90-minute recorded lecture and one 40-page PDF chapter turned into a revision set: a structured summary, twenty flashcards, a ten-question practice test with worked answers, and a citation list that survives a spot check against the source, costed at one student, one seat, free tier first. When we run it, each tool gets a score out of 100 across six dimensions, weighted in this order: output quality, total cost, job fit, sustained user experience, setup and integration, data and security. Until then there is no score column.

Disclosure: Agensi sells skills that install into Claude, Cursor, Codex CLI and Copilot, and this page links to our collections. We sell no study tool and no vendor paid to be listed. Reviewed by Samuel Rose, Founder of Agensi. Refreshed on a major release, a price change, or every six months.

Prices and ratings were checked on September 23, 2026. Student pricing is stated only where the vendor documents it.

How students actually use AI in a term

Four jobs: summarising and revising, finding and checking sources, drafting and editing, and learning something you missed. Most students use one general assistant for all four, which works until the sources are wrong. The groups below exist because the tools that are honest about sources are not the ones that write best.

AI tools for studying and notetaking

1. Gemini Notebook

Gemini Notebook, renamed from NotebookLM on July 16, 2026, ranks first because it does the one thing a study tool should: it answers only from the sources you upload, and when you hover a citation it shows the full quoted text and takes you to it in context. Google states your data is not used to train it unless you give feedback. It is listed unscored because it has 19 reviews on G2.

Watch: the rename. A Trustpilot reviewer wrote on August 25, 2026, "NotebookLM was really good. Gemini is not", and a G2 reviewer the same day accused it of being made worse to sell upgrades. The free tier caps you at 50 sources per notebook and 50 chat queries a day.

Price: Free. Google AI Pro $19.99 a month raises the caps. US college students 18 and over get AI Pro free for 12 months if they redeem by December 31, 2026, card required. G2 4.6 from 19 reviews, below our bar.

2. Quizlet

Quizlet is the flashcard tool everyone knows, with AI study sets, practice tests, study guides and, since August 2026, spaced repetition. It ranks second because it clears the rating bar and the others in this group do not.

Watch: we could not read Quizlet's privacy policy, so its position on training is not stated here; check it before you upload notes. The AI quiz mode is uneven: a reviewer in March 2026 got answer choices that read "B. C. B. D.". The Plus price on Quizlet's own upgrade page looks stale and third-party pages give a different figure, so confirm before paying.

Price: Free with ads. Plus price to confirm on the page. G2 4.5 from 291 reviews, Capterra 4.6 from 146.

3. Notion

Notion is the notes-and-everything workspace, and its Plus plan is free for students with a college email. It is third because the free student plan does not include Notion AI; the AI needs the $20 Business plan. Notion states it does not use customer data to train models by default and covers AI under SOC 2 Type 2 and ISO 27001.

Watch: the learning curve. Reviewers in August and September 2026 call permissions "coarse" and the setup "steep".

Price: Free for students on Plus, one member. Business $20 a month for AI. G2 4.6 from 13,846 reviews.

4. Otter.ai

Otter records and transcribes lectures live, with summaries and chat over the transcript, and it gives students 20 percent off with a .edu address. It is fourth for one reason: its privacy policy says it trains its own models on de-identified recordings and transcriptions by default, with no AI-specific opt-out described.

Watch: 300 free minutes a month and a 30-minute cap per recording, so a lecture needs Pro. A reviewer in September 2026 said accuracy "drops considerably" on technical terms and non-native accents.

Price: Free for 300 minutes. Pro $8.33 a month annual, $79.99 a year for students. G2 4.4 from 504 reviews.

RemNote is the spaced-repetition notebook that serious revisers swear by, with 25 percent off for students and a plain statement that it does not train on your notes. It is not ranked because it has one review on G2. For Claude users, the free Flashcard Drill Maker skill turns Q and A pairs into spaced-repetition drills with Anki export, and Active Recall Study Coach runs a session and a timed mock. Both in Claude skills for education and learning.

AI tools for research and citations

Every tool here passes the gate: it can show you the sentence behind a claim. None has 50 reviews, so all four are listed unscored and the order is our judgement. Researchers: this is your group, and the fourth tool is free.

5. Consensus

Consensus searches peer-reviewed papers and answers with the quoted sentence next to the paper's PDF, highlighted. Its privacy policy, updated August 18, 2026, says your queries and content are never used to train models. It ranks first because it is the most usable of the four for a student who is not yet a researcher.

Watch: passage-level evidence appears only "when that evidence is available", and two Trustpilot reviewers in 2026 were charged for an annual renewal "with no warning". The student discount and the price come from a page that may be stale; confirm.

Price: Free with 10 pro analyses. Premium about $8.99 a month, 40 percent off with a .edu email, to confirm. No rating source above our bar.

6. Elicit

Elicit cites every generated claim with the exact sentence or figure from the paper and lets you click any cell in a data extraction table to see the supporting quote. It is the researcher's tool of the four, with an academic price tier you select on the pricing page.

Watch: the free tier's exact credit allowance is not published and usage does not roll over. Elicit's own limitations page says its models can "miss the nuance of a paper or misunderstand what a number refers to". Its privacy policy has no statement on training.

Price: Free with limited usage. Pro $49 a month. Academic tier price on the page. G2 4.5 from 1 review, below our bar.

7. Scite

Scite shows whether later papers supported or contradicted a finding, and its assistant links every claim to the specific sentence in the specific paper. Its terms state it does not use customer data to train AI systems. Since September 2026 it plugs into ChatGPT and Claude, and it added an academic option.

Watch: the free plan has no assistant or search, only 25 credits for use inside other tools, and the 7-day trial converts to a paid plan automatically.

Price: Free Connect plan. Basic $20 a month billed yearly. Academic price by request. G2 4.7 from 27 reviews, below our bar.

8. Semantic Scholar

Semantic Scholar is free, run by Ai2, and its Ask This Paper feature returns supporting statements from the paper with each answer. It ranks last only because its release notes stop in 2023 and Semantic Reader works on arXiv papers only.

Watch: TLDRs cover computer science and biomedicine; Ask This Paper is English-only and on a subset of papers.

Price: Free. No rating source.

Perplexity, the tool most students actually use for this, is not in the group. Its answers carry numbered citations to pages, and a reviewer in August 2026 described "phantom citations linking to wrong pages". It is a fine search engine at $10 a month for students; it is not a citation tool. Two free skills do the checking a citation tool does not: Research Claim and Citation Auditor finds the sentence your citation does not support, and Sourced Research Brief labels every claim with its source and a confidence. Both in Claude skills for research and analysis.

AI tools for writing and editing

More on the general tools in best AI writing skills.

9. Grammarly

Grammarly ranks first for students because of Authorship: it records what you typed, pasted and generated, and you decide whether to share the report with your professor. That is the only tool on this page built for the integrity conversation rather than against it. Grammarly, now under the Superhuman brand, uses your content to train its models unless you turn off product improvement in settings.

Watch: the free tier allows 100 AI prompts a month, Authorship works only in Google Docs, Word, the Grammarly editor and Canvas, and a Trustpilot reviewer in September 2026 was charged $144 for a cancelled account. Grammarly's own detector page says never to rely on a detector alone.

Price: Free. Pro $12 a month. No individual student discount. G2 4.7 from 14,073 reviews.

10. ChatGPT

ChatGPT is the general assistant and the one with the best student offer: four free months of Plus for US college students, claimed by October 31, 2026, then $20 a month. Consumer plans train on your content unless you turn off "Improve the model for everyone". OpenAI's own position on detectors is that its research did not find them reliable enough.

Watch: a reviewer in September 2026 wrote that it "can sometimes give an answer that sounds confident even when the information needs to be checked". Custom GPTs are being retired.

Price: Free. Plus $20 a month, four months free for US college students. G2 4.6 from 2,996 reviews.

11. QuillBot

QuillBot paraphrases, summarises and checks grammar, and it is third because of what the paraphraser is for. Its own detector page says: if you used AI in your work, disclose it rather than revising to change a score. QuillBot samples individual users' inputs to improve its products by default; Team plans are excluded.

Watch: the free paraphraser stops at 125 words and 2 modes. Reviewers in late 2025 say it "hallucinates, adding new words that don't fit the context".

Price: Free. Premium $8.33 a month billed annually. G2 4.4 from 81 reviews, Trustpilot 4.8 from 14,142.

Jenni AI, the academic editor with inline citations, is not ranked: no G2 rating, and Trustpilot's 3.6 from 154 reviews is thick with billing complaints, three of them in the summer of 2026. Free gets 10 autocompletes a day. For the citation list itself, Academic Citation Formatter audits APA 7 and IEEE formatting inside Claude; it is in Claude skills for writing and content.

AI tools for interactive learning and tutoring

12. Khanmigo

Khanmigo is Khan Academy's tutor, built into its practice problems, and since July 2026 it asks you to explain how you got an answer before it moves on. Khan Academy states it does not allow LLMs to train on student data and does not share names with providers, with annual SOC 2 Type 2 audits. It is listed unscored because there is no review listing for it.

Watch: the learner plan requires an 18-plus purchaser in the US, so under-18s need a parent. Khan Academy itself stays free; Khanmigo is the paid layer.

Price: $4 a month or $44 a year. No rating source.

13. StudyFetch

StudyFetch turns your uploaded notes and lectures into flashcards, quizzes and tests with a tutor called Sparky. It ranks second because its rating comes from Trustpilot, its site has no pricing page, and its privacy page did not load, so we cannot say what it does with your uploads.

Watch: paywalls. "Everything starts getting locked behind premium", a reviewer wrote in May 2026. The price below is a third-party figure.

Price: Free to start. About $19 a month or $96 a year, third-party figure. Trustpilot 4.2 from 289 reviews.

AI agents for students

A tool answers a question. An agent takes a goal and works through steps, and for a student the step that matters is file access. An agent that reads your lecture folder and builds the revision set is the benchmark done for you, and a program with access to everything in that folder. Read what a skill can reach before installing one, and browse Claude skills for agents and orchestration for the parts.

Free tiers: what you get, where it stops

Gemini Notebook: 50 sources and 50 queries a day; AI Pro free for a year for US college students. Quizlet: free with ads. Notion: Plus free for students, no AI. Otter: 300 minutes, 30 per recording. Consensus: 10 pro analyses. Elicit: limited credits. Scite: 25 credits, no assistant. Semantic Scholar: everything. Grammarly: 100 AI prompts a month. ChatGPT: free chat, plus four free months of Plus for US college students until October 31, 2026. QuillBot: 125 words a paraphrase. Khanmigo: none, $4. StudyFetch: free to start. The best free AI agent skills page lists free skills.

Where studying ends and cheating begins

Facts, not sermons. Three things decide it.

What the tool does to the work. Gemini Notebook and the research group produce summaries and citations from sources you give them; the words are yours to write. Grammarly edits your words and records that it did. ChatGPT and QuillBot can produce the words, and both vendors say so: OpenAI tells students to put shareable chat links in the bibliography, QuillBot says disclose rather than revise to beat a score.

What a typical policy says. Most institutions now permit some AI use with disclosure and prohibit undisclosed generation of submitted work, and many run detectors. The two detector vendors on this page both say a detector score should never decide a case alone, and false positives land on real students. Grammarly's Authorship report exists because of that.

What yours says. Policies differ by institution, by course and by assignment, and this page describes a typical one, not yours. Read your handbook and the brief for each assignment, and when the brief is silent, ask before you submit. Teachers reading this from the other side of the desk have their own page: AI tools for teachers.

What students actually report

For each tool we looked for dated reviews from the last 12 months on G2, Capterra and Trustpilot, retrieved September 23, 2026, and coded each against the six themes in our method. Too small to score, and we do not score it.

Two themes. Billing: annual renewals with no warning at Consensus, charges after cancellation at Grammarly and Jenni, "everything locked behind premium" at StudyFetch. Students are the audience most exposed to a card left on file. And output that needs checking: phantom citations at Perplexity, broken quiz answers at Quizlet, confident wrong answers at ChatGPT. The first theme is why every price above carries its paywall point. The second is why the research group has a gate.

Why one tool gives good or useless output

Disclosure, again: Agensi sells skills, so we have a commercial interest in this argument. Judge it on the mechanism.

Take the benchmark. Give ChatGPT a lecture transcript and a chapter and ask for a revision set. With nothing else it guesses what the exam will test, what counts as a good flashcard, how a worked answer should look, and it produces a citation list that looks right until you check it. Two students with the same tool and the same prompt get the same plausible, generic set.

Now give it a procedure. The exam format. The rule that every flashcard is one fact. The worked-answer structure your course uses. The rule that every citation carries the page number and the sentence it supports. The set changes because the inputs changed, not because the model did. That is the difference between a tool and a tool with a skill, and it is also the difference between a citation list you can defend and one you cannot.

A skill is a folder of instructions and reference files the assistant loads when the job comes up. It installs into Claude Code, Cursor, Codex CLI or Copilot in about 30 seconds, and every listing on Agensi passes a security scan before it goes live. We have not yet run this as a scored before-and-after. When we do, the result goes here with the date and version.

How to choose your study stack

Beginners: one tool, Gemini Notebook, with this week's readings in it, for two weeks. Add the second tool for the second job only when the first one is a habit. Undergraduates: claim the ChatGPT and Google student offers before their deadlines, and put a reminder in your calendar for the month they start charging. Researchers: Elicit or Scite, and the citation auditor skill on everything before submission. Final year: the writing group plus AI for resume writing, and the free Resume Adapter skill, which tailors one resume to each job.

Mistakes to avoid

Citing from Perplexity. The cost is a reference your marker cannot find.

Leaving a card on file after a trial. Consensus, Jenni and Grammarly all have 2026 reviews about the charge that followed. The cost is $107 to $144.

Uploading a lecture to a tool that trains on it. Otter does by default; Gemini Notebook, Consensus, Scite and Notion state they do not.

Paraphrasing to pass a detector. QuillBot's own page says not to. The cost is a misconduct hearing.

Buying a second assistant. ChatGPT free plus Gemini Notebook free covers the benchmark. The cost is $20 a month you did not need.

The six skills named above are a start. The planners and trackers that hold a term together are in Claude skills for productivity.

Samuel Rose is the founder of Agensi, a marketplace for AI agent skills built on the SKILL.md open standard. Published and last updated September 28, 2026.