There's a lot more to generative AI chatbots than ChatGPT, but which one is best for medical writers?
Generative Artificial Intelligence (AI) tools have already transformed many industries. But, in MedComms, many medical writers are unsure if they should use these tools and how. Concerns about data confidentiality and potential breaches are understandable. But we can address these concerns with proper precautions. These include disabling web access, chat history and training.
Since the launch of Chat GPT in late 2022, there are now many alternatives. But keeping up with the latest developments can be confusing. So, here's my rundown on the available crop. I'll cover the key differences and some uses for medical writers.
ChatGPT
ChatGPT-3.5 remains the most popular AI chatbot. It has a simple user interface. One recent addition is the option to provide customs instructions. You can share any information with ChatGPT, such as your location or occupation. You can customise your ChatGPT's responses. It will use your preferences in new conversations.
ChatGPT-4 is the paid version ($20/month). The main difference between this and GPT-3.5 is the ability to browse, create and use GPTs. You can also access extra tools such as DALL-E, web browsing, and advanced data analysis.
If you don't fancy paying $20/month, you can now access ChatGPT-4 for free in CoPilot.
While ChatGPT-4 has its merits, I still prefer ChatGPT-3.5. It's faster for one. Also, a recent paper in Plos One found that GPT3.5 wrote more "readable" medical abstracts with less hallucinations than GPT4. The GPT3.5 abstracts were also more readable than the original authors' abstracts!
Tips for medical writers: Create a GPT with GPT-4 that applies editorial brand style guidelines to your copy.
Claude
Claude is a free chatbot launched by Anthropic over a year ago. Claude has one advantage. It can read, analyse, and summarise uploaded documents.
Tips for medical writers: Upload up to 5 RCTs papers and ask Claude to:
Extract and tabulate relevant data such as sample size, outcomes measures, statistical results
Synthesise the findings across the 5 papers to highlight common themes or differences
Appraise the strengths and limitations of each study
Compare the papers' results to clinical practice guidelines
CoPilot
Microsoft CoPilot (formerly Bing chat) offers three conversational styles. These are Creative, Balanced, and Precise. CoPilot can access the internet for more current information and provides links for sources.
Gemini
The AI formerly known as Google Bard, uses a different LLM (Gemini Pro) to ChatGPT. It's free to use, provides access to Google and can generate text and images. The paid version infuses Google's AI assistant into its productivity apps. These apps include Gmail, Docs, Sheets, Slides, and Meet.
Groq
If you need speed, then Groq is by far the fastest AI tool out there. It answers prompts around 90% faster than ChatGPT. It's also free.
Perplexity
Perplexity AI is good for accurate, factual and up-to-date research. The Pro version gives more in-depth answers to longer questions and provides references. The Pro version also includes the Claude 2.1 AI model.
Tips for medical writers: Select the Academic button to limit AI's output to published academic papers.
Currently available AI Chatbots*
Link | Pros | Cons | |
ChatGPT | Largest LLM available; ChatGPT 3.5 is free | Data cut-off is Jan 2022 for ChatGPT 3.5; free version not connected to the internet; ChatGPT-4 requires a subscription | |
Claude | Up to 200,000 tokens; allows file uploads | Claude Pro is $20/month; not widely available yet | |
CoPilot | Includes GPT-4; accesses the internet; different conversational style; links back to sources; it's free | Limited to 5 responses on a single conversation; 2,000 character limit per prompt | |
Gemini | Not limited to a set amount of responses; works well with Google apps | Slower than ChatGPT-4 | |
Groq | Very fast (500 tokens/sec); has internet access; free to use | Faster does not mean better | |
Perplexity | Links to sources; upload documents; Pro version allows longer searches | Pro version is limited to 5 searches/day |
*As of March 2024. LLM, large language model.
More useful online tools for writers
Here's some online tools I find useful, particularly for SEO medical blog writing:
Hemingway Editor Plus (https://hemingwayapp.com/)
Even the best writers make mistakes. Like the great man himself, the Hemingway Editor loves a short sentence. It also fixes copy issues. These include passive voice, adverbs, and hard sentences. It suggests simpler alternatives.
With the Hemingway app, you want to aim for a score of 6 or less.
Storybase (https://www.storybase.com/)
Storybase is a useful SEO tool for writing medical blog articles. It looks to see if your Google tag and meta description are working as hard as they can. Type in your SEO title and meta description. It will give you a score, with tips to improve SEO.
A score of 93% on Storybase is what you're aiming for.
Surfer SEO (https://surferseo.com/)
Surfer SEO analyses the top-ranking pages for your target keywords. It finds the key factors that help them rank well. It then gives specific recommendations. They aim to improve your content's structure, keyword use, and other SEO elements. This helps boost your chances of achieving higher search engine rankings.
Surfer SEO dashboard.
Summary
AI chatbots won't be replacing medical writers any time soon. And, used in the right way, they can be invaluable tools for speeding up your workflow.
Here's my three AI chatbot recommendations for medical writers:
Perplexity: for research tasks when accuracy is paramount. It uses recognised sources rather than hallucinating facts
Claude: great for writing creative and high-quality responses that need less reworking compared to ChatGPT
ChatGPT: general workhorse, good at creative writing tasks
FAQs
How do AI chatbots like ChatGPT check that the info they give is accurate and reliable?
Are there ethical concerns with using AI for medical research and writing
What does the future hold for conversational AI in medical writing?