After my last post on How to get started with Generative AI (available here), I heard from some readers asking: "What do I do next?" If you've used ChatGPT, Claude, or Google #Gemini or Microsoft Copilot even once in the last year, (as 50% of adults in major economies now have done according to a recent survey I read) then you've had a taste of what #GenAI can do (and perhaps what it cannot). This second newsletter builds on my previous one, so perhaps have a click through to that one first if you want.
So let's now go further and explore how to use different AI tools in harmony to get real work done.
Don't Chase the Shiny
Every week seems to bring a "new best model" with Kimi K2 Thinking newly released from China this week, with Gemini 2.5 Pro before that. Rather than trying to chase the latest and greatest, my suggestion is to gain competence and experience with two models and then use others only as you need more advanced capabilities. Stability and usability matter more than the very latest hype. I suggest two models because they all have strengths and weaknesses and I don't currently believe one model is best for everything (as yet).
So, a core skill of using AI is knowing when to bring which one to the table. As always, for all of the suggestions I offer below, you will need to ensure you look after your privacy and that you respect the IP of any content you share to a public cloud provider.
Microsoft Copilot or 'Open weight' models like #Llama, #Mistral, #Phi, or #Granite can run locally on a company server or even on a laptop so then you don't have these same privacy concerns, but you may still have moral restrictions that will limit you using even local AI with proprietary source materials. (I will cover installing and running these 'Open weight' models in the third primer in this series).
The safest way to proceed is to only use public cloud GenAI models for documents that do not have company details, personal names or any sensitive details in them, and for which there is no IP issues with such uploading. It's important to remove sensitive details from any document you choose to share unless the document is entirely in the public domain. Now let's get started!
1. Drafting or Improving a Document or Email
When you're staring at a blank screen or need to polish tone, you can now confidently draw upon the capabilities of the major models that are now all strong in language sense and style. Useful tools:
- ChatGPT 5 (Chat) gives balanced, concise, and reasonably dependable outputs. (The paid version has deeper capabilities). The free version has shallower 'thinking' than the paid version so I recommend the Pro version. This also enables you to personalise Chat for your style and background and enables you turn off the use of your data for training of future models.
- Claude is great at writing and particularly powerful for generating test case outputs (so-called synthetic data). I used it to generate a variety of simulated business case documents that I could test a policy against. You will quickly run against usage limits with the free version. I recommend the Pro subscription.
- Gemini Pro (available inside Google AI Studio) is wonderful for tuning documents and generating thought starters from a simple prompt. A limitation is that it only works with PDFs or inside Google Docs. (You can of course convert your Word, Excel or PPT docs to PDF and then upload them, or just copy and paste your text into it). It's free to use but the data you share may be used for training.
- Microsoft Copilot is integrated with Microsoft products such as Teams, Outlook and Word. It is now generally part of your included M365 subscriptions or business services. That integration is excellent and it is powerful for transcribing meeting notes from Teams (for example).
Prompt tip: You can ask all these models to "Rewrite this in my professional tone but keep my key message intact." (Upload a sample of your writing and ask it to keep its writing consistent to your tone.) Pro tip: if you use the paid versions of the tools you can personalise them with your background and tone to make your prompts even more powerful. You can also feed the output of one tool into the other for a second opinion. I find Chat great at a first pass improvement, and then I often put the result (if not confidential) into AiStudio and will prompt it along the lines "your AI colleague has tweaked this document. Can you further refine it for my intended audience of xxx with intent of making it easy to understand and more engaging."
2. Summarising or Querying Long Documents (Even Videos)
The big models are perfect for dealing with report overload or thematically working up workshop notes. All of the major models can now work well enough with long documents for most business purposes. Copilot can access a private knowledge store of company documents and NotebookLM (using Gemini) can access an uploaded set of documents for you to query or further explore or build upon (but note that uploaded data may be used for future training).
Check your T&Cs. While the major models are all capable at understanding long documents, Chat, Claude and Copilot particularly understand tables and attachments (including Excel spreadsheets). Gemini integrates directly with your Google Drive files, is great at handling big documents and reading handwritten documents. Gemini can even summarise YouTube video clips (up to about 20mins long) into easy to read themed outputs. All of them can transcribe audio files but Chat is particularly good at this.
Example Prompt: "Summarise this document in 200 non-technical words for an exec who has 60 seconds." Pro tip: You can ask Chat to examine a spreadsheet of data for quality issues and it will give you a report. Copilot natively integrates with Excel and will help you with fusing data columns, creating the right formula to use etc. All of the tools are good for document summarising or queries. Be aware it's important not to blindly trust numerical outputs as numeracy is not yet a strength of the models (see my first primer newsletter).
3. Researching a Topic or Writing a Brief for Management
Suppose you are asked to produce a report on some aspect of industry, or customer feedback on social media about your product. For formal reporting, my suggestion is to start with citation-based research tools before you move into generative synthesis. Here's a suggested flow:
Use ResearchRabbit.ai or Perplexity.ai for verified, citable sources (ResearchRabbit provides a visual method of finding the most-cited references and exploring the topic).
Perplexity is a 'mixture of experts' model which will use different models in the background depending on the task. It is excellent at producing well-researched reports (but always double check all references it offers). If your research is not formal academic, you can just use Perplexity. While you can use AiStudio, Chat or Claude to generate a deep report, it's my experience that Perplexity is very good at choosing the best of those models to use (in the background) and is more reliable at not hallucinating invented references.
Now send THAT generated output into Claude or Chat or NotebookLM for further research exploration. Then ask those tools to build on THAT material with references to additional sources if required.
Example Prompt tip: "From these articles, what are three emerging trends?" Pro tip: Upload your generated research report into Claude and ask it to produce an interactive dashboard about the topic based on the uploaded report. You can also upload a style guide for it to use and you can specify the kinds of graphs and widgets you want. Similarly NotebookLM will automatically produce a MindMap, interactive podcast and a video based on your uploaded document(s). And amazingly you can interact with the NotebookLM artificial agents in the podcast to further explore the themes. Limitations: I generated such a podcast on a higher mathematics topic by a senior researcher and he felt the podcast made significant errors in understanding his esoteric topic. But for more general business topics it should be fine. Just listen with a critical ear before you further share the automatically generated podcasts.
4. Creating Presentations (and Practising Delivery)
There are great tools for creating presentations. In terms of producing actual content, then I like to start with my own as a first pass, but you can use Chat, Claude or AiStudio, and you can also use them to create outlines tailored for your purpose. Once you have your desired content, you can generate slides with visuals for example using AI Studio (Gemini) which outputs to Google Slides. (There are other tools like Gamma or Beautiful.ai that can turn outlines into visuals, but you can certainly make do just with AI Studio.) Or you can just use PowerPoint and use Copilot to help you find a good template and relevant images that you can use to fill with your content. I did that recently for a presentation and it worked well. I'm sure Copilot can probably do this pretty automatically, but I haven't yet found the knack. (Feel free to share your experience).
Example prompt tip: "Make this engaging for a 5-minute presentation. Tell it your audience, give it a template of a winning presentation and ask it to use a similar style." Pro tip: Use your phone to make a small video of you practicing your presentation and upload it to Chat or AiStudio (if not confidential) and then ask it for feedback. It will tell you about your pace, let you know about your waving hands (note to self) as well as the content and flow (it's good to use a different model to the one you used to help you build the presentation).
5. Building a Simple App Prototype (Advanced Topic)
You don't need to code anything anymore just to test out a simple idea. Chat Codex, Claude Code, AiStudio and Github Copilot can all generate full end to end apps suitable for testing an idea. These are advanced topics. But Claude and AiStudio are particularly adept at making quick interactive apps that can be easily shared (e.g. I used AiStudio to interactively build a game for new parents during my keynote presentation. I input into the build prompt an audience suggestion of "24 hours in the life of a new baby". It was built in 5 minutes by AiStudio while I was giving the presentation. I recorded the laptop screen while I was using the game and then uploaded the video of the screen recording to Chat 5 and then asked it how it would improve its compliance with Australian Midwifery standards. I then took the 9 improvement suggestions from Chat and put those back into AiStudio and it greatly (and rapidly) improved the app. It did a great job - on the fly.
Final Thought
You don't have to be a technologist to lead with AI but you do need to think like a conductor. Each model brings its own instrument. You can possibly make do just with a Chat Pro subscription but I find that I get most value in my professional life by using Chat, Claude, Perplexity and Google's Gemini suite (AiStudio, NotebookLM and Gemini itself) in sequence or combination.
For example, Perplexity plus Claude is a great combo for most idea research purposes (Perplexity is best for ground truthing, Claude is great for generating test case studies, writing, interactive dashboards and managing whole projects). Chat and AiStudio are two tools that I switch back and forward between to build and riff off each other when tuning up complex reports or analysing data.
For me, Copilot is my least-used tool, but that's probably because I don't currently work in a corporate environment. I'm sure I would use it a lot more if I needed to keep my materials totally confidential and had lots of material in Teams and SharePoint and corporate emails for it to use as source materials.
Please let me know any corrections, or further thoughts on this newsletter. Next newsletter will be about running models in a way that is totally confidential - on your own machines. Paul
PS I haven't covered the use of #MCP and #Agents as that is getting into advanced AI. If there's interest I will do that as a fourth in the series - let me know if you'd like that.
Resources
Here are some wonderful YT content developers to follow:
- Grace Leung youtu.be/zFcxA9T_BWs?si=70sLawPlqfuauHSe (general tool use, particularly for business digital marketing)
- Elle Wang youtube.com/watch?v=Jut6WTytoew (a bit technical - but check her earlier posts on her 3:2:1 research technique for business report researching)
- Paige Bailey (works for Google DeepMind and mostly posts on LinkedIn) linkedin.com/in/dynamicwebpaige/
- Antonina youtu.be/XUuVnH6gSa8?si=YXJiC-rxS0S4vq-S - mainly posts related to the Google ecosystem but great examples of multiple tools working together
- Rob the AI guy youtube.com/@realrobtheaiguy - he posts about a variety of tools
- Enovair youtube.com/@EnovairPartners - a lot of generally useful AI tips covering most of the tools here and also how to use two or more together for efficient workflows
#genai #aitools #aiworkflow
Leave a Comment