Spelling and grammar checkers have, perhaps quietly, evolved from simple typo-spotters into more sophisticated writing assistants. Once limited to flagging basic errors, they can now offer stylistic suggestions, tone adjustments, predictive text, citation help, and more.
These advances are powered by natural language processing and language models, enabling tools to assess grammar in context, rewrite sentences, and provide smarter feedback. However, they are still far from perfect. Grammar, unlike spelling, depends heavily on context and intent, which remains challenging for automated tools. We enjoyed this evaluation of a range of grammar checkers, including ChatGPT, conducted by Daniel Kies from the College of DuPage in Illinois, concluding that “No grammar checker functions adequately overall”.
As the tech develops, AI-powered writing support continues to garner attention. With the integration of generative AI features and increasing questions around the difference between free versus paid access to popular tools, we thought it is a good time to take a more in-depth look at the different kinds of writing support on offer today. Though these tools aren’t perfect, they play a vital role in supporting students and learners through their academic journeys – understanding their growing capabilities is essential.
Taking a look at the tools
With this in mind, we set out to explore a few popular tools for writing support, including those with and without generative AI capabilities. For this comparison, we’ve chosen three widely used tools:
- Microsoft Word – A word processor with built-in spelling and grammar assistance with no generative AI features
- ChatGPT – A ‘general purpose’ Large Language Model (LLM) powered generative AI chatbot
- Grammarly – A writing support tool with optional generative AI features
This certainly isn’t a like-for-like comparison. These tools serve different purposes and were built with different goals in mind. But the aim here isn’t to decide which one is ‘best’ overall. Instead, we want to explore what each tool offers in terms of writing support and consider how they might meet the needs of different users.
We explored how each tool might be used to support writing across the following areas:
- Identifying and correcting spelling and grammar errors in a paragraph
- ‘Improving’ writing beyond what is spelling/grammatically correct e.g., tone, structure
- Rewriting full sentences
- User experience and personalisation options
Use the links to go directly to each section, or skip ahead to what we learned and some of our thoughts on finding the right tool.
Free vs Paid
For Grammarly and ChatGPT we have looked at both the free to access and paid versions, and we discuss the differences between these
At the time of this blog, our free ChatGPT account could use their latest model (GPT-4o) for a limited number of conversations and would then revert to GPT-4.1 mini based on demand. Therefore, we tested with both GPT-4.1 mini and 4o, however results between the two models were nearly identical. This was not too surprising, as these tests were relatively simple, and GPT-4.1 mini is still a powerful model.
1. Identifying and correcting spelling and grammar errors in a paragraph
We started by looking at how each tool handles the ‘correctness’ of spelling and grammar. To do this, we used the same paragraph containing five deliberate spelling and grammar errors across all three tools.
Test paragraph:
The study had a look at how differant amounts of sleep effected concentration levels in university students.
Participants track their sleep for, one week, and completed a short memory test each day.
Future studie’s will rewrite larger amounts of the test materials to explore more aspects of memory stuff.
Our text contained five intentional errors. We selected some of these from another of Daniel Kies’ blogs which provides guidance on twenty common usage errors based on Robert J. Connors and Andrea A. Lunsford’s 1992 and 2008 studies of writing errors in academic works by students:
- Spelling error: ‘differant’ should be ‘different’
- Incorrect comma placement: ‘, one week’ should be ‘one week’
- Wrong word choice: ‘effected’ should be ‘affected’
- Wrong tense: ‘participants track’ should be ‘participants tracked’
- Possessive apostrophe error: ‘studie’s’ should be ‘studies’
It’s worth pausing here to acknowledge that the idea of ‘correctness’ can be more straightforward for spelling than for grammar. Spelling typically follows standardised rules, but grammar is far more dependent on context, intent, and subtle nuances. These are areas that grammar checkers often struggle with, and where even human experts don’t always agree. This isn’t a quantitative study on how many errors each tool identified (though we do include those results). Our aim has been more to see how each tool approaches basic spelling and grammar support in practice.
Microsoft Word:
Word’s built-in Editor can check both spelling and grammar, offering corrections or automatically fixing common errors. Issues can be highlighted as the user types, with red zigzags for spelling, blue dotted lines for clarity, and double blue underlines for punctuation errors.
Word identified two errors (differant and studie’s) which it categorised as spelling issues and offered suitable suggestions for corrections. It did not pick up on the other three errors, flagging nothing under the grammar tab. Where errors are identified, Word does explain the issue and, in some cases, can provide optional ‘learning tips’ which give more detailed advice.
ChatGPT:
We used the following prompt with ChatGPT along with the test paragraph:
Check this paragraph for spelling and grammar errors. Don’t correct it, just list the issues and suggest corrections
ChatGPT generated a list correctly identifying all five of our issues, explaining them, and offering suggestions for each. We did need to be specific in our prompt that we only wanted suggestions, as ChatGPT tended to rewrite the text (sometimes even when asked not to!). It also offered to rewrite the paragraph for us and implement the suggestions, which it could do effectively.
We saw that ChatGPT also identified a couple of additional issues relating to word choices and tone. Picking up on the phrase ‘memory stuff’ as being informal and vague, and the awkward phrasing of ‘rewrite larger amounts of the test materials’.
Paid ChatGPT users also have access to custom GPTs. We tested one called Grammar Checker by studyx.ai, which ranked in the top 10 of writing-related GPTs. It flagged the same issues but gave much more detailed explanations and suggestions. Depending on the user, this might be helpful or overwhelming, but there is always the option of prompting it to be more concise or finding another GPT that suits better.
One important note: ChatGPT and most custom GPTs do tend to default to American English. Specifying British English within the prompt resolves this, or it can be added for all a user’s chats using ChatGPT’s custom instructions feature.
Grammarly:
Both free and Pro Gammarly accounts offer spelling and grammar checks, with suggestions users can accept or dismiss.
In both free and Pro accounts, spelling and grammar errors are flagged under the ‘Correctness’ tab. In our free account four issues were identified under correctness, the only error not identified was the misplaced comma. Interestingly, when inputting the same paragraph on the Pro account six issues were flagged under corrections – all five of our planted errors and an additional suggestion to remove the second comma in the second sentence. It seemed that the Pro account was actually providing more thorough spelling and grammar checking. Both versions could also provide detailed explanations of the errors, for instance, explaining the difference in meaning between effect and affect.
With generative AI features enabled, Grammarly also offers a chatbot-style prompt interface. We tested the same prompt we used with ChatGPT and got very similar results, it identified the five spelling and grammar issues correctly and some additional ones for tone and word choice.
Generative features are included in both plans, but free users are limited to 100 prompts per month (compared to 2,000 for Pro), so using a prompt for basic error checking may not be efficient. However, the chat interface could help users ask follow-up questions and get deeper explanations.
2. ‘Improving’ writing beyond what is correct
Next, we wanted to consider how these tools might support writing beyond the correctness of spelling and grammar. This could relate to suggestions for appropriate tone, word choices, structure or even content.
Microsoft Word:
Though Word can’t receive prompts or generate text it does have considerably detailed options for suggesting ‘refinements’. The refinements options cover a range of categories from clarity and inclusiveness to sensitive geopolitical references. Users can set a tone – Formal, Professional, or Casual – to guide these suggestions or they can manually choose these in the grammar settings menu.
These settings menus do present a potentially difficult user experience for some compared to prompting or to Grammarly Pro’s preferences list. Users face a long list of small tick boxes, which can take time to explore, and they may not find it easy understand what each option actually means. Further, while the depth of options looked comprehensive, we struggled to get Word to flag issues even when trying to make mistakes based on the settings available.
For example, we did expect that with ‘Formal’ tone selected, we would generate a suggestion for some of the informal phrasing in our example, such as ‘had a look at’ and ‘memory stuff’, but these were not flagged.
ChatGPT:
With ChatGPT there is plenty of potential to generate more advanced feedback and suggestions. We already saw that even when using a basic prompt looking for spelling and grammar errors ChatGPT also identified issues with tone and word choice. In the AI team, we’ve been using generative AI tools like this for some time and we’ve found a lot of flexibility in how feedback can be requested.
We can use broad prompts like “Help me improve this paragraph” and “Give me feedback on my writing”, or take a more targeted approach – e.g., “Make this sound more professional” or “Adapt this for a 10-year-old.”As before, if you only want advice rather than a rewritten version, it’s important to ask ChatGPT not to make those changes automatically.
For our example, we used this prompt:
Help me improve this paragraph, it is for an academic report. Don’t make changes just provide suggestions for improvement.
In this way, we provided some context to guide the output from ChatGPT, which we felt was on a similar level to using the ‘Formal’ tone setting in Word.
The output was a comprehensive list of suggestions on how to make the language more formal, change word choices to improve clarity, fix spelling and grammar errors, and even improve the content itself.
Grammarly:
This is where we saw a key distinction between free and paid Grammarly accounts. Grammarly Pro offers advanced suggestions under the categories ‘clarity’, ‘engagement’, and ‘delivery’. These are not available to free users, though they might be offered two or three of these a day at no charge to entice them into upgrading their subscription.
With the test paragraph Grammarly Pro identified an additional suggestion under clarity, relating to the informal wording in the first sentence.
Pro users are also able to tailor the aspects Grammarly gives feedback on through an extensive preferences menu with categories including consistency, clarity, formality, inclusivity and more. Users can also set up a ‘style profile’ to receive custom suggestions.
In addition, they can also define ‘Goals’ for each document, such as intended audience and tone, which further refine the suggestions provided. While the number of settings may feel overwhelming, we found they are well-explained and usually accompanied by helpful examples.
With generative AI features enabled, free users can access more advanced support and suggestions through prompting. Grammarly’s selection of preset prompt options could be especially helpful here for users who might be less confident in prompting. Users who do not or cannot use generative AI though, are much more limited in this respect within the free version of Grammarly.
3. Rewriting full sentences
Here, we were looking for the tool to rewrite our test sentences in full rather than just provide suggestions for individual words or short phrases.
Microsoft Word:
Word cannot achieve this. Though it can offer word and phrase suggestions it does not have the capability to rewrite entire sentences.
ChatGPT:
ChatGPT can easily be prompted to rewrite full sentences. As we encountered with the previous tests we found it often overdelivered as well, pasting in our test sentences with no prompt at all resulted in ChatGPT automatically providing a reworked version.
The reworked version fixed the spelling and grammar issues and made some improvements to the informal wording in our example.
Grammarly:
With a free Grammarly account this can only be accomplished with the generative AI features enabled.
Users can choose from preset options like “simplify it”, “make it more descriptive”, or rewrite for a specific audience. Custom prompts are also supported.
Pro users gain access to sentence rewriting options which are not part of the generative AI feature set. These are toggled in the preferences settings and include options like rewriting for clarity, stronger impact, and effectiveness.
However, Grammarly Pro’s non-generative AI suggestions only appear if the tool identifies an issue. In our example, we could ‘rewrite’ by inserting the suggested changes, but we couldn’t get a full rewrite of an entire sentence at will. The generative AI features, by contrast, let users request rewrites for any sentence.
4. User experience and personalisation options
The final aspect we looked at was around the user experience of the tools, particularly how each tool might fit into different writing workflows and any personalisation options they offer.
Microsoft Word:
Word, being a word processor, is the built-in experience which many of us are accustomed to. The application itself is for processing text and the spelling, grammar and writing support is integrated. This allows the user to receive corrections and suggestions as they type and make any changes all in the same program. Users can disable real-time checking and run it manually if preferred, which may be useful if live suggestions feel distracting.
As we saw, Word offers a broad range of settings for customisation. These are split across a few menus, with some accessible under ‘Grammar and Refinements’ and others through the main ‘Options’ menu in the ‘Proofing’ and ‘Advanced options’ sections. As we noted, these aren’t too well explained to the user within the program and as the options are spread across different menus the actual experience of customising these settings isn’t the easiest.
ChatGPT:
ChatGPT isn’t a word processor, so using it for writing support involves writing elsewhere and copying sections into the tool for review. It can’t provide real-time suggestions as you type, which some users may find limiting.
Personalisation options include setting what ChatGPT should know about you and what traits it should display. You can also control whether past chats inform future responses and remove individual chats from it’s memory.
These settings can shape the tone and style of responses. For example, adding a trait like “adopt a sceptical, questioning approach” leads to responses with more rhetorical questions and a more critical tone. Providing personal details such as the course you study or your writing preferences can help to tailor responses further. For instance, you might request that all responses be in bullet points.
That said, how it interprets that information will be up to the model’s training and development. As with all outputs, there is a potential that ChatGPT will be working on a stereotypical interpretation of that information.
Grammarly:
One of Grammarly’s key benefits is that it offers its own platform for word processing, or it can be used across most other word processors and platforms through its browser and desktop integrations. These integrations are available to both free and Pro users, providing a lot of flexibility for fitting Grammarly into different workflows.
Grammarly also offers personalisation through ‘personalised insights’, where users can opt in to allow Grammarly to analyse more of their writing data, such as frequently used phrases or writing habits. This feature is optional; users can disable it at any time and have any stored data deleted. Additionally, for those not wanting to use generative AI, Grammarly allows users to disable all generative AI features entirely.
What did we learn?
Although this was quite an informal exploration, taking a closer look at these tools and their differences helped us think more critically about how we consider the writing support that is available today.
Generative features can provide advanced support, but they aren’t the only way
Looking across the three tools – one with no generative AI, one with optional generative AI features, and one fully generative- it’s clear that generative AI can offer considerable capability to support the writing process. The ability to ask directly for the kind of support you need is a real strength, especially for users who are confident in shaping their own prompts or who want more interactive feedback.
However, generative AI isn’t a requirement for stronger writing support. Grammarly Pro offers comprehensive spelling and grammar checking, and more in-depth suggestions on tone, clarity, and delivery using rule-based features, not generative ones. One drawback of generative AI too, which we didn’t encounter here but remains a possibility, is the potential for inconsistent outputs and hallucinations, which could affect the quality of the support given.
Free tools can go further than we might expect, but paying undoubtedly offers more
Both ChatGPT and Grammarly offered strong free options with capabilities that could cover basic spelling and grammar checking and also stretch into more advanced support. However, much of the more advanced writing support and personalisation features required paid subscriptions.
Free accounts also come with some danger of relying on features which may not be free forever. It has not been uncommon for providers to change which features are available to free users, limiting their amount of access or removing them entirely. Notably, ChatGPT’s model for this works on demand, so if there are more paid users accessing their latest model, then free users will be more limited or moved back to a previous model.
Ultimately, for users who require reliable and frequent access to stronger support, free options are unlikely to meet their needs.
Different tools suit different workflows
Each tool offered a different user experience. Word has the benefit of being a word processor with built-in spelling and grammar checking, the basics are automatic and easy to access, but it’s not sufficient for more comprehensive support. With ChatGPT the user must be more active and intentional with copying in text, crafting prompts, and reviewing generated responses. Then Grammarly, with its interface that blends a word processor with layered writing support and optional generative features, often felt like a hybrid between the Microsoft Word and ChatGPT experiences.
So, how do we find the ‘right’ tool?
We have only looked at three tools in this piece, but with so many more available, students and educators alike face a complicated landscape when it comes to choosing a tool.
As with choosing any tool, a good starting point is to identify which features are most important to the individual user. For example, a learner who needs consistent spelling support across different platforms may benefit from a tool like Grammarly, which can run in the background across multiple programs. Meanwhile, users who need more advanced support and are comfortable exploring more open-ended tools may find free large language model (LLM) platforms such as ChatGPT, Google Gemini, or Microsoft Copilot offer flexible, powerful support.
We know that AI tools often raise concerns around data protection and privacy, particularly when free accounts are being used. Understanding these risks and ensuring users have proper guidance on the appropriate use of tools is essential. As well, where institutions can provide licensed versions of tools many of these concerns can be mitigated.
There are concerns too, around how much these tools can do – particularly when it comes to generating or rephrasing large amounts of text. However, with the right training and guidance, these tools can be used effectively and appropriately to support learning. We’ve seen that generative AI can provide significant opportunity to support disabled and neurodivergent students, and making opportunities for this to be explored is key. We believe that adopting a truly effective approach to AI use means ensuring tools can be leveraged responsibly to support all learners.
Join the conversation
These conversations are ongoing, and we continue them at our AI in Education community meetups, with monthly discussion sessions for FE and HE Jisc members.
If you are interested in exploring writing support and AI from an accessibility perspective too then we encourage you to join Jisc’s Accessibility and Assistive Technology communities.
Find out more by visiting our Artificial Intelligence page to view publications and resources, join us for events and discover what AI has to offer through our range of interactive online demos.
Join our AI in Education communities to stay up to date and engage with other members.
Get in touch with the team directly at AI@jisc.ac.uk
One reply on “From Typos to Tone: Exploring Current Tools for Writing Support”
I love this post! The findings chime with my experience exploring the use of writing tools with students in proofreading workshops. Using genuine mistakes from the students’ previous writing, we compare the efficacy of MS Word, Grammarly, Copilot, Gemini, Claude (and any other tool students choose, including their unassisted brains) in identifying and explaining errors. I also ask the LLM chatbots to suggest independent study resources related to the identified errors.
Yes! GenAI tools rewrite text even when you ask them not to. Yes! The MS Editor often fails to flag errors (e.g. gender-non-inclusive language) even when you’ve selected them through the customisation tool. Yes! Grammarly offers good advice, even when the GenAI feature is disabled.
Doing this activity with different subject/cohorts since 2023, I’ve noticed that the relative performance of Copilot/Gemini/Claude varies across error correction worksheets and has been improving over time. But they’re still not as good as me when it comes to recommending error-specific, reputable follow-up resources for language study. Writing teachers aren’t redundant … yet.