15 Things You Should Stop Doing the Hard Way
How to use AI at work in 2026 to save time, think better and get more done — with 15 practical prompts you can copy today.
Work Faster
Get more done in less time.
Think Smarter
Better insights and decisions.
Create More
Turn ideas into real results.
Unlock Opportunities
Build skills for a changing world.
There's probably a smarter way.
Work hasn't suddenly become easier. But the tools available to do it have changed dramatically.
People still spend hours summarising documents, rewriting emails, preparing presentations, researching unfamiliar subjects, organising meeting notes and staring at blank pages trying to figure out where to begin.
Generative AI changes that equation.
Research is beginning to quantify the difference. In a randomized field experiment involving 7,137 knowledge workers across 66 firms, workers given access to an integrated generative-AI tool spent roughly two fewer hours per week on email during the latter half of the experiment. Another field study involving 5,179 customer-support agents found AI assistance increased issues resolved per hour by an average of 14%, with substantially larger improvements among less-experienced workers.
That doesn't mean AI should make your decisions for you.
It means many of the mechanical steps between having a task and having something useful to work with can now be dramatically shortened.
AI can help you find the important information in a long report. It can challenge an idea before you present it. It can turn rough notes into a structured plan. It can help you understand a spreadsheet, prepare for a meeting or get past the intimidating blank page.
The advantage isn't simply having access to AI.
The advantage is knowing what to do with it.
Here are 15 places to start.
“It's not about doing less.
It's about achieving more.”
Less time.
Better results.
Turn a long document into something you can actually use
Reports, proposals, policies, research papers and contracts often contain far more information than you actually need for the decision in front of you. The problem isn't always reading speed. It's identifying what matters. AI can act as a first-pass analytical assistant: extracting decisions, risks, deadlines, obligations, contradictions and questions that deserve closer attention.
Read 40 pages from beginning to end. Highlight large sections. Make separate notes. Go back through those notes and try to determine what actually matters.
Tell AI exactly what you need to learn from the document and require it to distinguish between information explicitly contained in the source and its own interpretation.
Analyse the attached document for me. Identify: 1. The 10 most important points 2. Decisions that need to be made 3. Key risks or concerns 4. Deadlines and important dates 5. Actions required and who appears responsible 6. Financial implications mentioned 7. Anything that appears contradictory or unclear Separate facts stated in the document from your interpretation. For every important conclusion, identify the section or page containing the supporting information. If the document does not contain enough information to answer something, say so rather than guessing. Finish with a one-page executive summary.
For legal, financial, medical, regulatory or other consequential material, use AI to navigate the document, not replace it. Check important conclusions against the original source.
Stop spending ten minutes writing a two-minute email
A surprising amount of knowledge work consists of rewriting the same message until it sounds clear, professional and appropriately diplomatic. AI is particularly useful when you already know what you want to say but are struggling with how to say it.
Write. Delete. Rewrite. Add too much explanation. Remove half of it. Worry that the message sounds too abrupt. Rewrite it again.
Give AI the facts, recipient, objective and desired tone. Ask it to produce a concise draft while preserving your position.
Help me write the following email. Recipient: [WHO IS RECEIVING IT] Relationship/context: [BRIEF CONTEXT] What happened: [FACTS] What I need from them: [DESIRED ACTION] Deadline, if applicable: [DATE] Tone: professional, warm, confident and concise. Do not sound robotic, overly formal or apologetic. Do not invent facts. Keep the email under 180 words unless additional detail is genuinely necessary. Give me: * a subject line * the email * a shorter alternative version
AI works better when you provide the situation rather than simply saying, "write me a professional email." Context changes everything.
Turn a messy meeting into clear action
Meetings create conversations. Progress happens when those conversations become decisions and actions. Unfortunately, important commitments often disappear into notebooks, transcripts and chat messages. AI can turn rough notes into a structured record while you still verify the important details.
Attend the meeting. Take fragmented notes. Promise yourself you'll organise them later. Three days later, try to remember who agreed to what.
Give AI the transcript or notes and ask it to separate discussion, decisions, actions, owners, deadlines and unresolved questions.
Convert these meeting notes into a clear action summary. Create the following sections: 1. Executive summary 2. Decisions made 3. Action items 4. Owner for each action 5. Deadline for each action 6. Issues still unresolved 7. Questions requiring follow-up 8. Items to carry into the next meeting Do not invent owners, deadlines or decisions. Where something is unclear, mark it "Needs confirmation." Finish with a table containing: Action | Owner | Deadline | Status
If meeting notes contain confidential information, check your organisation's AI and data-handling policies before uploading them to an external service.
Get oriented before disappearing down the research rabbit hole
Starting research on an unfamiliar subject can be inefficient because you don't yet know which questions matter. AI is useful at the beginning of research because it can help create a map of the subject. It is not automatically the final authority.
Open Google. Read the first result. Open another twelve tabs. Find conflicting information. Lose track of which source said what.
Use AI to define the landscape first: terminology, major questions, competing viewpoints and the sources you should investigate.
I need to research [TOPIC] for [PURPOSE]. Assume I am intelligent but not yet an expert in this subject. First give me: * a plain-English explanation * the 10 concepts I need to understand * important terminology * the major questions or debates * common misconceptions * what information is likely to be outdated quickly * the strongest types of primary sources I should consult Then create a research plan in priority order. Do not present uncertain claims as established facts. Clearly label anything that requires current verification.
For anything current — prices, regulations, statistics, product features, political developments, market conditions — verify against up-to-date primary or authoritative sources.
Stop beginning presentations with slide one
The hardest part of a presentation usually isn't PowerPoint. It's deciding what the audience needs to understand, believe or do when the presentation ends. AI can help build the argument before you build the slides.
Open a blank presentation. Choose a template. Type a title. Spend twenty minutes adjusting fonts. Still have no clear story.
Start with the audience, desired outcome and evidence. Ask AI to construct the narrative first.
Help me structure a presentation about [TOPIC]. Audience: [AUDIENCE] What they already know: [CONTEXT] What I need them to understand: [OBJECTIVE] What I want them to do after the presentation: [ACTION] Time available: [MINUTES] Build a persuasive presentation narrative. For each slide give me: * slide purpose * proposed headline * key message * supporting evidence/data required * suggested visual * speaker note Avoid generic headings such as "Introduction" and "Conclusion." Make every slide advance the argument.
Use AI for structure before decoration. A beautiful presentation with no argument is still a weak presentation.
Don't ask AI for ideas. Ask it to challenge yours.
"Give me 20 ideas" is one of the least interesting ways to use AI. A better use is creating a thinking partner that explores an idea from perspectives you may not naturally consider.
Brainstorm alone until the first reasonable idea feels like the best idea.
Ask AI to generate alternatives, challenge assumptions, expose weaknesses and then improve the strongest options.
I am considering the following idea: [DESCRIBE IDEA] Do not simply agree with me. Analyse it from five perspectives: 1. Customer 2. Competitor 3. Financial 4. Operational 5. Skeptical investor Identify: * strongest aspects * weak assumptions * hidden risks * opportunities I may have missed * reasons it might fail * ways competitors could respond Then suggest five materially different versions of the idea. Rank nothing yet. Ask me three questions that would materially improve the analysis.
Agreement feels good. Useful disagreement is often more valuable. Explicitly tell the AI to challenge your assumptions.
Stop staring at rows of numbers waiting for insight to appear
A spreadsheet can contain an answer without making that answer obvious. AI can help you decide what to examine, explain formulas, suggest analyses and translate results into plain language.
Scroll. Filter. Create a chart. Change the chart. Notice an odd number. Wonder whether it matters.
Begin with a business question and ask AI to help determine which calculations and comparisons can answer it.
I have a dataset containing [DESCRIBE COLUMNS/DATA]. My business question is: [QUESTION] Before analysing anything: 1. Tell me what data-quality problems I should check. 2. Identify the fields most relevant to the question. 3. Suggest the calculations or comparisons required. 4. Identify potential misleading interpretations. Then analyse the data I provide. Separate: * observations directly supported by the data * possible explanations * recommendations Do not claim causation where the data only shows correlation. Show the calculation behind important conclusions.
AI can make analytical mistakes too. For important numbers, verify formulas and totals independently.
The blank page no longer deserves an hour of your day
AI doesn't need to replace your voice to make writing easier. Often its most valuable role is producing the imperfect first structure that gives you something to react to.
Stare at the opening paragraph. Write two sentences. Delete them. Check email. Return to the blank page.
Give AI your argument, audience and raw thoughts. Ask for a structured first draft that you then improve.
Help me create the first draft of [DOCUMENT]. Audience: [AUDIENCE] Purpose: [PURPOSE] Main argument/message: [MESSAGE] Important points that must be included: [POINTS] Tone: [TONE] First create the outline. Then draft the document. Preserve my ideas rather than adding unsupported claims. Avoid clichés, corporate filler and generic AI phrases. Flag places where stronger evidence or examples would improve the document.
Don't judge AI by whether its first draft is publishable. Judge it by whether the draft gets you to a better final version faster.
Turn AI into a tutor instead of an answer machine
There is a major difference between asking AI for an answer and asking it to help you understand something. The second is usually more valuable.
Read an explanation you don't understand, search for another explanation and hope the second one is clearer.
Tell AI what you already understand, ask it to teach progressively and make it test you before moving on.
Teach me [TOPIC]. My current level is: [BEGINNER / INTERMEDIATE / ADVANCED] My objective is: [WHAT YOU WANT TO BE ABLE TO DO] Teach this progressively. Start with the mental model I need before introducing technical detail. Use practical examples and analogies where useful. After each major concept: * ask me one question * wait for my answer * identify any misunderstanding * explain it differently if necessary Do not move to the next concept simply because I responded. Help me actually understand it.
Try explaining the concept back to the AI in your own words and ask it to identify what you've misunderstood. Active recall beats passive reading.
Make better comparisons before making bigger decisions
People often compare options inconsistently. One product wins on price, another on quality, another because somebody recommended it — and suddenly the decision has no coherent logic. AI can help structure the comparison.
Create a mental pros-and-cons list while unconsciously changing the criteria depending on which option you're considering.
Define the criteria before evaluating the options and force uncertainty into the open.
Help me compare these options: [OPTION A] [OPTION B] [OPTION C] The decision I am trying to make is: [DECISION] My constraints are: [BUDGET / TIME / LOCATION / OTHER] First propose the criteria that should matter. Do not choose for me. Create a comparison showing: * factual differences * advantages * disadvantages * costs * switching/implementation difficulty * risks * information still missing Separate verified facts from assumptions. Finish by showing which types of user or situation each option appears best suited to.
AI should improve the quality of your decision process, not make every decision on your behalf.
Turn "we should do this" into something executable
Ideas are cheap. Execution requires sequencing. AI is useful for converting an objective into workstreams, dependencies, milestones, risks and next actions.
Create an ambitious to-do list containing 47 items with no obvious starting point.
Give AI the destination, deadline, resources and constraints and ask it to work backwards.
Help me turn this objective into an executable project: Objective: [OBJECTIVE] Deadline: [DATE] Resources available: [PEOPLE / BUDGET / TOOLS] Constraints: [CONSTRAINTS] Work backwards from the deadline. Create: * major milestones * workstreams * tasks * dependencies * likely bottlenecks * decisions required * risks * measurable completion criteria Identify the critical path. Then tell me the five actions that should happen first. Do not assume unlimited resources.
After creating the plan, ask: "What would make this plan fail?" That second conversation is often more useful than the first.
Stop treating every platform like a blank page
A strong idea can often become several useful pieces of content without simply copying and pasting the same words everywhere.
Write the article. Then start from zero for LinkedIn. Start again for email. Start again for social.
Create one strong source asset and ask AI to adapt the idea — not merely shorten the text — for different audiences and formats.
Use the following original content as the source: [PASTE CONTENT] Preserve its core ideas and factual meaning. Adapt it into: 1. A LinkedIn post 2. A short email 3. Five social posts 4. A 60-second video script 5. Three headline/hook variations Do not merely summarise the original. Adapt the opening, structure and emphasis to each format. Do not add facts that are not present in the source. Keep the voice intelligent, natural and specific. Avoid exaggerated claims and generic motivational language.
Repurposing should multiply the idea, not duplicate the wording.
Find the signal hiding inside hundreds of comments
Customer feedback is valuable precisely because it is messy. Different customers describe the same problem in different words. A recurring frustration can be hidden across reviews, support tickets, surveys and emails. AI can help group those signals.
Read comments individually and remember the loudest complaint.
Analyse feedback as a dataset: recurring themes, frequency, severity, sentiment and representative examples.
Analyse the following customer feedback: [PASTE OR ATTACH FEEDBACK] Identify: * recurring complaints * recurring positive themes * requested features * confusion points * objections * unmet needs * phrases customers repeatedly use Group similar comments into themes. Where possible, show the number or proportion of comments supporting each theme. Distinguish widespread issues from isolated comments. Do not treat frequency as proof of importance without explaining the context. Finish with the five questions I should investigate next.
Don't only analyse unhappy customers. Positive feedback can reveal what customers value most — and what you should protect.
Prepare for opportunities with evidence, not generic answers
Career applications often become exercises in repeatedly rewriting the same history. AI can help you connect your actual experience to the requirements of a particular opportunity without inventing qualifications you don't have.
Send the same CV everywhere, change the opening paragraph and hope the recruiter connects the dots.
Compare the opportunity with your actual experience, identify evidence and gaps, and tailor the presentation accordingly.
Compare this job description with my CV. Do not invent experience, qualifications, achievements or responsibilities. Identify: 1. Requirements I clearly meet 2. Requirements I partially meet 3. Requirements not demonstrated in my CV 4. Relevant achievements that deserve more prominence 5. Keywords that genuinely apply to my experience 6. Questions I should prepare for in an interview Then suggest how I can restructure my CV to make the relevant evidence easier to find. Flag anything you are uncertain about rather than filling the gap yourself.
The objective isn't to trick an applicant-tracking system. It is to make real evidence easier to see.
Ask a better final question: What do I do next?
AI becomes considerably more useful when conversations end with action rather than information. A good analysis that changes nothing is still just an interesting document.
Research. Save links. Create notes. Think about it again tomorrow.
Ask AI to convert what you've learned into a small number of specific next actions, starting with something you can actually do.
Based on everything we've discussed about: [TOPIC / PROJECT] convert this conversation into an action plan. Separate: DO TODAY DO THIS WEEK DO LATER For every action specify: * what needs to be done * why it matters * what information or resources are required * what completion looks like Identify anything that is still an assumption and needs verification. Keep the immediate plan realistic. Finally answer: "If I can do only one thing next, what should that thing accomplish?" Explain the reasoning without making the decision for me.
AI is at its best when it reduces the distance between thinking and doing.
The real skill isn't prompting
You may have noticed something about the prompts in this guide.
None depends on secret words.
There is no magic phrase.
The strongest prompts simply make the job clear. They tell the AI what you're trying to achieve, provide the context it needs, define the desired output and establish boundaries around what it should not assume. That aligns with current prompting guidance: clarity, context and explicit instructions generally produce more useful results than vague requests.
But there is an even bigger lesson.
The real AI skill isn't memorising prompts.
It's learning to look at a task you've performed the same way for years and ask:
Is there now a smarter way to do this?
Sometimes the answer will be AI. Sometimes it won't. Knowing the difference is part of the skill.
AI should make you more capable, not less thoughtful
There is a temptation to measure AI adoption by how much work can be handed over to a machine.
That is probably the wrong measure.
A better question is whether the technology allows people to spend less time on mechanical work and more time applying judgement, experience, creativity and attention where they matter.
Research already suggests that generative AI can improve productivity in some workplace contexts, but the effects vary substantially according to the task, worker and way the technology is used.
So don't automate something simply because you can.
Use AI where it makes you faster without making you careless, more informed without making you dependent, and more capable without outsourcing your judgement.
That's the advantage worth pursuing.
Your next step
You don't need to change the way you work overnight.
Choose one task from this guide that you regularly do the hard way.
Try the prompt. Change it. Add your own context. Challenge the answer. Improve it.
Then decide whether it actually made the work better.
If it did, you've learned something considerably more useful than another AI trick.
You've learned where AI belongs in your workflow.
The advantage isn't AI. It's knowing what to do with it.
Galmo AI — AI for those ready to level up.
15 AI Prompts for Work Cheat Sheet
Keep all 15 prompts from this guide in one practical reference.
No spam. Unsubscribe anytime.
Sources & Further Reading
Brynjolfsson, Li and Raymond studied the deployment of a generative-AI conversational assistant among 5,179 customer-support agents. The study found an average productivity increase of roughly 14%, with larger gains among less-experienced workers.
A randomized field experiment involving 7,137 knowledge workers at 66 firms examined the effects of a generative-AI tool integrated into everyday work applications. Among other findings, active users spent approximately two fewer hours per week on email during the latter half of the experiment.
Practical guidance on writing clearer instructions, supplying context and specifying desired outputs when working with generative-AI systems.