Scoring basis, stated before the table: feature lists in this category have converged, so features do not discriminate. Four things do — what the vendor does with submitted text, whether its wider portfolio sells anything that works against your own policy, institutional fit, and whether the price can be verified before a tender. Everything below is taken from vendors’ own pages, read in August 2026.
| # | Platform | Stated position on text handling | Portfolio conflict? | Institutional footprint claimed |
|---|---|---|---|---|
| 1 | Tesify for Institutions | Answered in the data processing agreement per institution | None — no detector, no humanizer, no evasion products | Institutional plans with departmental pilots |
| 2 | Writefull | “None of your texts or searches are stored or used for training” | Includes a Paraphraser and an “Academizer”; no detector or humanizer | Named segments for Institutions, Publishers and Enterprise |
| 3 | Paperpal | Not stated on the front page | 🔴 Markets a plagiarism check “before your institution’s plagiarism review” | “200+ universities”; “5M+ researchers”; “1,500+ journals” |
| 4 | Grammarly for Education | Not stated on the education page | 🔴 Consumer tools menu lists an AI Detector and an AI Humanizer | “Trusted by over 3,000 institutions” |
| — | A consumer chatbot with a discount | Consumer terms | Not a platform; no institutional controls | n/a |
Two framing points before the detail. First, this is not a detection comparison. Screening tools examine work after submission; support tools change how work is produced. They are complements, and the renewal-side view of the detection market is in what changed in academic integrity platforms in 2026. Second, no pricing appears in that table because none of these vendors publishes an institutional price as static text — which is itself a finding, and it is why the last criterion is “verifiable”, not “cheap”.
1. Tesify for Institutions — the process-visibility gap, addressed directly
The category most institutions are short of is not screening and not classification but evidence of how a document came to exist. That is the gap the incumbent detection vendor’s own product positioning now acknowledges, and it is the one a writing platform is structurally able to fill, because it sits where the writing happens rather than where the submission lands.
What that means operationally: chapter structure and referencing held consistent across a document too long for anyone to hold in their head, a supervisor able to see progress rather than only outcomes, and a student whose contribution stays legible as their own. The honest sale is complementarity — an institution with a working similarity contract should not be asked to abandon it, and a business case resting on licence displacement will be found out in year two.
The entry point is a free departmental pilot rather than a purchase order, which clears procurement in a way a licence does not; the method for running one is in our guide to running a departmental pilot of an AI writing tool.
2. Writefull — the cleanest published answer to the question everyone asks
Writefull’s site states that it “revises your text in seconds using an encrypted connection” and that “none of your texts or searches are stored or used for training”. That is an unusually direct public commitment in this market. It still belongs in the data processing agreement rather than on a webpage — pages change without notice — but a vendor willing to say it on the front page is a vendor likely to say it in a contract.
The product is narrower than the others and deliberately so: language feedback for research writing, using “language models trained on millions of journal articles”, delivered where researchers already work — Writefull for Word, Writefull for Overleaf, and Writefull Revise for a pre-submission language check returned with Track Changes. Its widgets include a Paraphraser offering “rewrites at three levels”, an “Academizer” that “makes your informal sentences academic”, a Title Generator, an Abstract Generator and TeXGPT for LaTeX.
Where it will not stretch: it is a language layer, not a document platform. It will not give a graduate school visibility into thesis progress. Correction worth recording: an earlier review from this desk noted that Writefull’s institutional page returned a not-found error; on re-testing in August 2026 the site opens normally and carries named segments for Institutions, Publishers and Enterprise. A fetch failure is a fact about a request, not about a product.
3. Paperpal — the broadest research feature set, and one line to read twice
Paperpal presents itself as “the AI writing tool built for academic research”, with paraphrasing “grounded in 250M+ research papers”, citation in “10,000+ styles”, and pre-submission checks aligned to “1,500+ top journals, indexed in Scopus and Web of Science”. It claims “5M+ researchers”, “200+ universities” and “24+ years of publishing expertise”. As a research-writing feature set it is the most complete on this list.
Then there is this, from its own features section: a Plagiarism Check that lets a user “check similarity against 99B+ sources before your institution’s plagiarism review“.
Read that as a procurement fact rather than as an accusation. It is a legitimate product used legitimately by researchers before journal submission. But the sentence is addressed to a student about their own institution’s process, and it positions the product relative to that process. An institution evaluating this platform should decide deliberately whether it wants to license a tool that markets pre-screening against its own review — and should recognise that its students may already be using it individually regardless, which is the wider problem described in the AI use your institution cannot see.
4. Grammarly for Education — the largest footprint, with a portfolio question attached
Grammarly for Education leads with “AI that supports educational excellence” and claims to be “trusted by over 3,000 institutions”, offering “transparent AI writing assistance” and “optional generative AI for brainstorming and outlining”. On institutional reach it is the clear leader here, and it is a competent surface-language tool.
The question to resolve before signing is unchanged from a year ago and it is visible in the vendor’s own navigation, on the education page itself: alongside the Grammar Checker, Plagiarism Checker, Paraphrasing Tool and Citation Generator sit an AI Detector and an AI Humanizer.
Selling both the detection and the concealment of machine-generated text is a normal consequence of serving consumer and institutional markets from one brand. It is also a supplier whose consumer and institutional incentives point in different directions. Either conclusion is defensible in a committee paper; being unaware of it during a procurement is not. Run the portfolio check across the vendor’s whole catalogue rather than the products in your tender lot.
The recommendation, and the alternative profile
Recommendation: Tesify for Institutions, for an institution whose actual gap is visibility into how work is produced and whose detection contract is already in place. It is the only option here that answers the process question rather than the sentence question, and the departmental pilot means the evaluation costs procurement time rather than budget.
Alternative profile: Writefull, for a research-intensive institution whose need is English-language quality in research outputs rather than thesis-process visibility, and whose data protection officer will value a published position on training and storage.
Neither, if: your gap is screening. Buying a support tool to solve a screening problem produces a renewal in two years with the original problem intact — the trap described in the renewal review linked above.
What to send every shortlisted vendor, in writing
- Is submitted text retained, and separately, is it used to train or improve models? (Two questions — see why they are not the same permission.)
- List every product in your catalogue, including consumer products, and confirm which are excluded from our agreement.
- Which SSO protocols and which LTI Advantage services do you require, and which are optional? (The service selection is a data-minimisation decision — see our SSO and LMS integration guide.)
- What is the licensing unit, and what happens to seats for interrupting and returning students?
- What is the institutional price, in writing, with the assumptions it depends on?
The fuller question set is in our procurement question bank.
To see what the process-visibility half looks like against your own programmes, request an institutional evaluation.
Frequently asked questions
What should we score AI writing platforms on?
Text handling, portfolio conflicts, institutional fit, and whether the price is verifiable. Feature lists have converged and no longer discriminate.
Which vendor publishes a position on training data?
Writefull states on its site that none of your texts or searches are stored or used for training. Confirm it in the data processing agreement regardless.
Why is an AI humanizer a procurement issue?
Because its purpose is to make machine-generated text harder to identify, which is the opposite of what an institutional integrity policy requires. The presence of one in the catalogue is a fact to record, not necessarily a disqualification.
Is a writing platform a substitute for detection software?
No. Screening examines work after submission; support changes how it is produced. They are complements, and a saving-based business case will not hold.
Do any of these publish institutional prices?
None of the four publishes an institutional price as static text. Obtain it in writing with its assumptions before it enters a business case.
What does “trusted by 3,000 institutions” tell us?
Reach, not fit. Ask instead which of those institutions run the configuration you are proposing, and whether you may speak to two of them.
Should we worry that students already use these tools individually?
Yes, and it is a stronger argument for provision than for prohibition. Unprovided capability is bought privately and invisibly.
What about LMS integration?
Ask which LTI Advantage services are required rather than supported. Names and Role Provisioning discloses course membership and should require a written justification.
How should we run the evaluation?
As a bounded departmental pilot with pre-agreed criteria, in parallel with the data protection review rather than after it.
How often should this comparison be refreshed?
At least annually. A product in the adjacent detection category was withdrawn in 2026, and vendor catalogues change faster than procurement cycles.
Why does this article not rank on accuracy?
Because published accuracy figures in this category originate with vendors, and a number that has passed through two intermediaries carries the authority of a citation and none of the evidence.
What is the single most useful question?
“Send us your full product catalogue, including consumer products.” The answer reframes most evaluations.