AI tools across the writing workflow
Generative AI tools can take work off a medical writer's day at every phase, from research to communicating the results, and much of what they take is not the writing itself. The tools differ in how much of a task they hand over and in what each is built to do: assistants inside the office suite, search tools with a language model attached, tools that answer questions about a document, drafting and rewriting assistants, note-taking tools, and specialist tools for regulatory and marketing work.
What the tools can do in the writing
Generative AI can assist medical writing. Generative AI applications can save time, streamline workflows and potentially improve the quality of the output.
One medical writer sorted the openings by the phase of the work in 2023. In research and discovery there is a lot of potential in literature search and reference management, summarising, extracting article highlights and explaining methods such as statistical tests. In conception and writing, the tools can help structure and draft outlines, abstracts and lay summaries, and reformat references into different styles. In communicating the results, they are excellent for repurposing the available information for different formats and audiences: the text is put in with a request to adapt it to a 9-year-old or a 60-year-old, and the tool provides a draft the writer then curates.
Generative AI, and ChatGPT in particular, can be used to assist with many tasks, including:
- paraphrasing
- reformatting references to different styles
- rewriting materials and methods sections
- explaining statistical tests
- rewriting abstracts
- extracting article highlights
- suggesting keywords
- writing submission letters
- summarising scientific articles or medical information for various audiences
- writing lay summaries
- repurposing available information for different formats and various audiences
- creating educational and other training materials such as courses, webinars, presentations and hand-outs
- compiling product and safety information
- generating responses to customer inquiries
- generating scripts for chatbots and virtual assistants
AI features suitable for medical writing include generating insights from trends, patterns and correlations in large datasets, feedback on clarity and readability for different audiences, analysing the style of existing documents and suggesting improvements, finding relevant research papers and summarising their findings, finding citations and analysing whether articles mention, support or contrast a claim, generating charts and other visual aids from data, automating the creation and quality check of clinical documents such as clinical study reports from source documents, proofreading, and automating routine tasks such as formatting and referencing. In 2023 AI support for tasks such as collaboration was not yet available at the same level. Collaborative technology using AI appeared to lag slightly behind the new wave of AI tools for medical writing.
In one writer's 2024 account, AI and machine learning tools had been essential to the writing process in medical writing, formatting documents, designing study protocols, analysing manuscripts and communicating complex scientific information to lay audiences. They had made preparing regulatory submissions more efficient and less tedious, for both nonclinical and clinical processes, by helping create error-free, organised and correctly formatted documents according to health authority guidance. Other uses in regulatory submission planning include outcome prediction, image analysis and the design and conduct of clinical trials. In biomedical publications AI had been used for manuscript preparation, citation management, proofreading and formatting, and for analysing a manuscript and suggesting journals for submission, which has ensured compliance with journal submission criteria and increases the likelihood of acceptance. AI makes the editorial workflow more efficient by finding peer reviewers, reviewing statistical data and screening manuscripts, and the tools also detect plagiarism and provide quality assessments. Writers whose first language is not English had used AI to improve the grammar, clarity and syntax of manuscripts, namely the methods, results and reference sections, which can increase the likelihood of acceptance by notable journals. AI algorithms can be used to develop lay summaries of complex scientific information for press releases, news stories and product descriptions.
Work beyond the writing
The tools can support a writer's day-to-day workload as well as the writing. A medical writer has many tasks to fulfil all day long that have nothing to do with the actual writing, and one medical writer saw the tools' potential there: in the literature search, organising the literature and note-taking. AI need not be used only with the writer's own data, and it can be used in many other ways that potentially do not conflict with the writer's restrictions and responsibilities for privacy. The tools also support training and education, marketing for freelancers, drafting emails, giving a structure to a contract or to terms and conditions before it goes to the legal department or a lawyer, and job hunting, by improving a CV and practising questions. Text the writer has written can also be put in for revision, which is closer to paraphrasing.
In one medical writer's experience, clients are not as advanced, so data shared with them still goes back to spreadsheets, because that is often how documents are shared with them; the potential lies instead in the note-taking, the literature review and the organising.
Within one publisher in 2024, staff were using AI all the time to improve their own processes. Elsevier's customer and market insights team used AI to analyse open-text responses from customers, synthesising masses of verbatims into a summary, allocating the comments to themes and then quantifying the themes, which people in the team had previously done by hand and might have found tedious, and which AI now does immediately. One speaker's colleagues built a tool that writes the minutes of meetings from the video or audio transcripts, and whoever attended the meeting then edits them.
How much of the work a tool takes over
The tools continue a long line of writing technology, back through the world's oldest pencil, from the seventeenth century, to a poem from 1800 BCE about being too tired to talk, pressed into clay. The innovations of the physical world focused on the efficiency of writing and came over centuries or longer; those of the digital world came over years, months and sometimes weeks, and let the writer offload more of the cognitive load, working on the word, the sentence, the paragraph and then the entire essay.
From countless usability studies and his own use of the tools, one speaker found it helpful in 2024 to think of the current wave of AI tools in two categories, agents and copilots. A general chat tool presents an input box and seems to promise the entire universe, as if the world's knowledge were at the user's fingertips. An agent takes a natural-language prompt and figures out how to do the task automatically, so much of the decision-making is automated into the tool, and because the user is not really in the process, errors risk propagating to the end result. Like the intern who runs off with an entire project and hands it back at the end, an agent risks creating a lot more work. A copilot keeps the human in the loop: the user still gets from A to B but uses the tool to go through each step more efficiently, with the added benefit of being better able to document the steps. Different paradigms suit different tasks: an undergraduate writing an essay with a chat tool is usually willing to take the risk, but medical writers cannot make that gamble, because of compliance and legal review and because of who reads their work and who is affected by it, so they have to be more careful not to communicate things wrongly.
Choosing a tool for the task
Having the tools is not enough: a writer needs to know which tool to use for which task and how to use it effectively.
In 2023, besides ChatGPT and its plugins, hundreds of AI-assisted text-generating tools were being launched every week or built into existing applications as AI assistants. More than 100 new AI tools were released weekly, and it falls to the users to stay informed and check whether each tool can improve their workflow. In one medical writer's 2023 view, only the most valuable applications will be adopted and survive, and what comes next, and how the tools will integrate with and complement each other, cannot be predicted.
Assistants inside the office suite
Integrating AI into apps that writers already know and trust will aid their transition to working with it. Many of the tools are available inside Microsoft Office as add-ins, ChatGPT among them, already integrated and with no separate sign-up. Google Workspace offers the same as add-ons through the Google Workspace Marketplace, reachable from an application such as Google Slides, and many of them are based on large language models.
A custom GPT is created through a no-code chat interface inside ChatGPT and requires a subscription, whereas a plugin is built outside ChatGPT and connected to it through an application programming interface.
Microsoft's alternative to ChatGPT in December 2023 was Copilot, which needs no coding: it is reached through Edge or downloaded from Google Play, with Copilot Studio alongside. In Excel it generates formulas and creates visualisations; in Word it creates documents from prompts, such as a legal document, and structures material copied in from elsewhere, knowing even where to put a table copied from Excel; in Teams it takes notes and provides meeting recaps and insights.
Microsoft announced Microsoft 365 Copilot in March 2023: an assistant powered by GPT-4 and working with Microsoft Graph, an application programming interface that gives developers access to a broad range of Microsoft 365 services. One could think of it as ChatGPT built into the apps many medical writers use every day, and according to Microsoft it is even better than that. In 2023 it was not yet on the market, and it was being tested by selected business clients and in an invitation-only paid preview. In May 2023 Microsoft widened access from 20 to 600 enterprises worldwide and added Copilot in Outlook, OneNote and Viva Learning.
The typical Microsoft 365 user uses only a limited number of the features available across the apps, because the rest demand technical savvy and time to learn, and Copilot aims to open a wider range of them through prompts in plain text, which could give novices a smoother start and advanced users more speed and functionality. In Excel and PowerPoint a user can prompt it to present the data in a named sheet as a pie chart in shades of blue, or the information in a document as a presentation, and where the result is imperfect, modify the prompt and try again, quickly reaching something to edit instead of starting from scratch. In Word it can generate a first draft or a document structure with headings and subheadings, drawn from information on the internet used to train GPT-4 or from local documents: asked to write a 300-word abstract summarising a named document, it produces a draft within seconds, and asked for a 150-word paragraph about a topic, it writes from its training data. OneDrive, SharePoint, Teams and, to a lesser extent, Outlook offer document sharing, meeting and chat functions, and Copilot added on top of them was to summarise long email threads and draft suggested replies, produce meeting notes after Teams calls with summaries of the key points, and draft project updates from meeting notes and email conversations.
The Copilot for Microsoft 365 add-on was made available to all Microsoft 365 business customers on 15 January 2024.
In early 2024 Copilot's features and pricing varied by region and were subject to change. The minimum prerequisite was a Microsoft 365 Business Standard, Business Premium, E3 or E5 subscription at $12 to $57 per user per month, which included commercial data protection for private chat with Copilot at no additional cost, and the version compatible with the Office tools, Copilot for Microsoft 365, was an add-on at $30 per user per month, so a medical writer's minimum spend was $42 a month.
One contract medical writing company, which supplies clinical and regulatory writing to biotechnology, pharmaceutical and device companies through contract medical writers, chose Microsoft Copilot as its tool in 2024. The main reason was that it integrates into Word as a plug-in, as PerfectIt does, and Word is where, by its director of operations' guess, 99% of the writers he addressed write their clinical and regulatory documents. It was also accessible right away, whether to a freelancer or to a writer at a pharmaceutical or biotechnology company or a contract research organisation through its IT team. The company is judged on the documents it delivers, and its director held that building Copilot into its processes would help it deliver them without its having to define which of cost, quality and time to favour.
A freelance publications writer used Copilot in early 2024. It rapidly produced several title options from an article's content, giving her a fresh perspective from which she then created the title. For a presubmission letter, with the manuscript open in Word, it generated an engaging statement of the research's unique aspects and then a satisfactory one-page summary, which she edited and tweaked into the letter. Asked for three renditions of an awkwardly constructed sentence, it gave her options to select from and adjust, which she found both faster and better. Given a description of the situation, it produced good first drafts of the different abstract types clients need for manuscripts and conferences, which she then refined, and asked for a graphical abstract based on an actual abstract, it offered several illustrations that served as inspiration for her own. Its first responses were not tailored to her situation, and its performance improved as she honed her skill at writing more effective prompts.
The writer and the system administrator she works with also used Copilot for their business operations. With access to their data and their business context, it helped the system administrator write draft terms for a service contract for a customer, which let their lawyer review the contract quicker and saved money. It helped create promotional materials, presentations and sales proposals, which the writer says are not her forte. With Power Apps it generated an application for handling company emails from a description of what the app should do, which the system administrator reviewed and corrected without writing any code. Because Copilot can securely access her emails, files and chats, the writer found that it progressively understands her business operations more deeply, and she expected it to become an even more proficient assistant over time. The two expected their next major productivity gain to come from Copilot in Teams, and planned to create use cases in Excel, PowerPoint, Outlook and SharePoint.
Academic search and web search with a language model
Tools that combine search with a large language model split into academic search and general web search with language model capabilities.
SciSpace is an AI research assistant that reads, understands and explains uploaded scientific literature, with note-taking, connected papers, a citation generator that regenerates citations in a chosen style, a paraphraser and an AI detector. Its Chrome extension, SciSpace Copilot, helps the user understand technical language, mathematics and tables in PDFs, extract data from them to work with, and organise and annotate material.
scite helps discover and evaluate scientific articles through Smart Citations, which show the context in which a publication has been cited and classify whether the citing work provides supporting or contrasting evidence for the cited claim, so that a reader can assess the reliability and impact of a paper. The summary and the paper sit together on one screen, with no need to jump between tabs. In 2023 its integrated conversational assistant, based on GPT-3.5, answered questions from a database of 180 million articles, book chapters and data sets and could assist with writing.
Elicit, a free AI research assistant, draws on publications from Semantic Scholar to speed up the literature review: queried, it retrieves relevant papers and summarises their key information in a table. Consensus finds answers in research papers from the social sciences and biology. Which tool to use depends on what the writer is working on.
Perplexity was one writer's favourite tool in December 2023, and she expected it to become more useful than Google. It had free and paid plans, and in her account it answers queries accurately, adds sources and can be switched to an academic mode. Asked whether there is a reliable blood-based screening method that helps predict the risk of breast cancer, it returned an answer with references that made a starting point for the work.
Tools that answer questions about a document
ChatPDF and similar tools take an uploaded document, such as a scientific article, and answer specific questions about it or summarise its key points, which is very useful in an area full of difficult terms or one the writer does not work in every day. When the request is structured as a prompt rather than asked as a bare question, the tool follows the prompt's instructions and returns structured responses. PDF.ai, AskYourPDF and LightPDF are similar tools: through what is almost a chatbot interface they answer specific questions, find specific information in documents and simplify complex topics.
Scholarcy summarises articles, reports and book chapters, highlighting key sections and letting the user save and export the summaries to return to later. Unlike Elicit it summarises only one publication at a time, but in much more detail, extracting key concepts, a synopsis of the full text, a comparative analysis and more. genei extracts key information from articles, analyses research and summarises articles, and its paid subscription had GPT-3 capabilities in 2023.
Drafting and rewriting assistants
In December 2023 one drafting tool for academic researchers was trained on millions of research manuscripts and runs language and grammar checks, offers rewriting options that avoid repetition, and has an integrated academic translation function. Its rephrase options rewrite, trim or make text more academic, and it generates outlines, titles, abstracts and keyword summaries. One medical writer recommended it as fast for a first draft and an overview.
Jenni AI, a very similar tool for academic writing, paraphrases, simplifies, shortens, lengthens, translates, writes the opposing argument and writes with more depth, with a choice of style from academic to friendly.
Lex is an AI tool developed for writers rather than translators. In May 2024 it gave feedback on a draft, offered prompts for an article idea and for sharpening an introduction, and carried AI commands to continue writing, shorten text, run checks, ask questions about the document and generate text from a prompt.
Quillbot is a comprehensive paraphrasing tool aimed at improving grammar and wording, and it also has a summariser. Its smaller alternative, Wordtune, rephrases, improves tone and word choice, and shortens or expands a text.
Draftsmith, from the creators of PerfectIt, uses ChatGPT to suggest improvements to a writer's draft before it goes to a copy editor. It is a Microsoft Word 365 add-in from the Microsoft Store that needs a Microsoft account, it does not work with Google Docs, similar office applications or earlier versions of Word, and with track changes on, its changes are tracked under the writer's username. Its buttons on the Word ribbon are grouped to follow, broadly, the drafting process:
- Pre-Edit: Fluency Enhancer and Dictation Fixer
- Revise: Engagement Tuner, Word Count Trimmer and Empathy Tuner
- Simplify: Readability Tuner and Plain English Converter
- Review: Editing Helper
A group's button shows three default functions for that stage in a panel, and a function applied to the sentence holding the cursor offers a suggestion that the writer accepts with Replace Text, which moves on to the next sentence, skips with Next, or refreshes for a different one. It was sold by the month for ad hoc use, such as while self-editing or on a specific project, or by the year for regular use. Two reviewers held that it can be particularly useful for simplifying dense sentences and paragraphs, rewording complex concepts and reducing word count, and that working inside Word, without toggling to a browser and copying and pasting into ChatGPT, may save the writer valuable time.
Note-taking tools
In one medical writer's experience, note-taking while researching or in meetings is what takes the most time. Simple tools for it in December 2023 included Notion, a productivity and note-taking application, AudioPen, which turns voice notes into text, and Google Keep, and each can be tried out quickly.
Languages the tools work in
At the end of 2023 one medical writer used the tools in French, German and English without problems, with translation integrated, and held that while some languages are surely left out, the main ones are covered.
Specialist tools for regulatory and marketing work
DistillerSR is one tool for regulatory and clinical work. For post-market work it keeps adding whatever it finds over set periods without the writer doing anything. For generative AI in regulatory intelligence, one 2023 review suggested considering Dr.Evidence, Huma.AI, Orca1.ai and Yseop.
Yseop's Copilot application, presented in 2023, extends a company that began with financial report writing into clinical study reports, and it works with Microsoft Word but is distinct from Microsoft's Copilot. It combines an expert system with generative AI based on a large language model using open-source data. As its presenters described it, like other proprietary applications it is constrained to one document type and trained on synthetic data or data the client owns or selects, generating a specific type of document from specified input, which addresses concerns about data traceability and reliability, with the aim of speeding up the regulatory submission process through faster report writing. Yseop's longer-term goal was to cover as much of the submission process as possible with separate automation packs bought à la carte, with bundling possible for larger clients.
For marketing content, one medical writer recommended Jasper in 2023, with Copy.ai as a valid alternative. She named Writesonic alongside Jasper, and found Jasper a bit expensive but able to do everything. Jasper is an all-in-one platform trained to write original, creative content for all sorts of marketing assets, building on OpenAI's models and adding others for specific content marketing needs, with templates that help structure and detail a prompt and a plagiarism checker. AIwriter searches, writes content and adds valid references to it.