Blog

Digital marketing commerce online sale concept, Businesswoman using laptop with Ads dashboard digital marketing strategy analysis for branding.

<h3>TL;DR: AI can help you build a full marketing plan from scratch faster and with less stress. This guide explains how to use basic AI prompts to tackle every section of your plan, even if you’re not a marketer or a techie. You’ll finish with something practical and good to go.</h3><h2>Why Most People Get Stuck on Marketing Plans</h2><p>If you’ve opened a blank document and immediately felt stuck, you’re not alone. Most people struggle with marketing plans because they either have too many ideas or none at all. The real barrier isn’t creativity, it’s not knowing where to start or how to break the job into manageable steps.</p><h2>How AI Makes Planning Easier</h2><p>You don’t need to be a CMO or an AI expert. By feeding your business info into simple AI prompts, you can get clear, step-by-step help: from writing your business summary to figuring out your budget. AI doesn’t replace your judgement; it just makes the heavy lifting less intimidating.</p><h2>How it works:</h2><ul><li>For each section of your marketing plan, you type in basic info (about your business, your market, your goals, etc.).</li><li>Use an AI tool like ChatGPT, Claude, or Gemini with pre-written prompts for each section.</li><li>Review the AI’s draft. Add your own perspective—keep what’s useful, change what doesn’t fit.<br></li></ul><h2>The 7 Steps to Building Your AI-Driven Marketing Plan</h2><p><strong>1. Summarise Your Business</strong><br>Feed the basics about your business into the AI.<br></p><p>Prompt:“Here’s what my business does: [add a short description and a few numbers about your service, e.g. clients per month, main services]. What are the three biggest strengths and what should I focus on improving?”<br></p><p>What you’ll get: A summary that helps you spot strengths and quick wins.</p><p><strong>2. Get a Fresh SWOT Analysis</strong><br>Don’t just guess. Ask the AI to look for gaps.<br></p><p>Prompt:“Here’s my SWOT: [add your known strengths, weaknesses, opportunities, threats]. What’s missing in each area for a service business like mine, based on current market trends?”<br></p><p>What you’ll get: Two new points for each SWOT box so you don’t miss blind spots.</p><p><strong>3. Spot New Business Initiatives</strong><br>You know your current projects, but what’s next?<br></p><p>Prompt:&nbsp;“These are my current and planned business initiatives: [list a few]. What three other initiatives would help my service business grow or reach new markets right now?”<br></p><p>What you’ll get: Realistic, actionable ideas that match market shifts.</p><p><strong>&nbsp;4. Build Target Customer Personas</strong><br>AI can help you define your ideal customer.<br></p><p>Prompt:&nbsp;“Here’s my target market: [age, job, location, behaviours]. Can you describe three simple personas for a service business like mine, with each person’s top pain point?”<br></p><p>What you’ll get: A clear picture of who you’re selling to—and what they care about.</p><p><strong>&nbsp;5. Review Your Marketing Channels</strong><br>Stop wasting energy on the wrong channels.<br></p><p>Prompt:&nbsp; “These are my current marketing channels and how well they work: [list channels, e.g. Facebook, email, referrals, and a note on performance]. What three new channels should a service business try now to get in front of more of the right clients?”<br></p><p>What you’ll get:&nbsp; New channel ideas (and a reason for each) so you can reach more of your target market.</p><p><strong>&nbsp;6. Update Your Market Strategy</strong><br>Tactics change fast. Make sure you’re not stuck in last year’s thinking.<br></p><p>Prompt:“Here’s my market strategy: [explain in one or two lines]. What three tactics could I add or change to make this more effective for a service business right now?”<br></p><p>What you’ll get: Simple, current tactics you can act on straight away.</p><p><strong>&nbsp;7. Optimise Your Budget &amp; Tools</strong><br>Don’t guess where to spend or which tools to use.<br></p><p>Prompt:“Here’s how I split my marketing budget: [give rough percentages or dollar amounts]. And here are the main tools I use: [list your marketing or automation tools]. What changes would help a service business like mine get a better return?”<br></p><p>What you’ll get: Advice on where to reallocate spend and what tech could save you time or money.</p><h2>Why It Matters</h2><p>A good plan is better than a perfect one. Most small teams waste weeks searching for templates or over-thinking strategy. AI helps you get out of your own way, so you can start testing what works. You’ll avoid blank page syndrome and move quickly from idea to action.</p><h2>What You Should Watch Out For</h2><p>AI can speed up planning, but it’s not foolproof. Always review the outputs, double-check anything that sounds off, and use your common sense before presenting your plan to a client or your boss. Treat the AI like an extra set of hand not your brain.</p><h2>&nbsp;Ready to Build a Plan That Actually Gets Used?</h2><p>If you want hands-on help using AI to make marketing plans, join the next<a data-mce-href="https://aistrategyconsulting.com.au/introduction-to-ai-5-day-bootcamp" href="https://aistrategyconsulting.com.au/introduction-to-ai-5-day-bootcamp"> 5-Day AI Bootcamp.</a> You’ll leave with a real, finished plan not just slides. The bootcamp is for people who want to learn by doing, without jargon or sales pitches. Secure your spot now and get your next marketing plan sorted.</p>

7 Easy Steps to Build a Marketing Plan with AI

<h3>TL;DR<br>AI isn’t levelling the playing field, sadly the pundits are not right, it’s deepening the gap between elite big tech and you. Unless you take deliberate action, your business risks being left behind while the “AI elite” pull further ahead. The winners are those who stop waiting and start building practical skills, and like, very fast.</h3><h2>Why “AI for Everyone” Is a Convenient Myth</h2><p>The tech industry likes to spin a story about AI lifting all boats. Look closer and the reality is stark:<br></p><blockquote><p>Most Australian businesses are stuck watching from the sidelines. Budgets are tight, digital skills are patchy, the digital transformation hangover lingers and leadership teams still hope a silver bullet is coming in the form of a Big 4 Consulting Firm or former employee of one. Meanwhile, those with the confidence and know-how are reaping the rewards: bit time, leaving everyone else to play catch-up.</p></blockquote><h2>What’s the Price of Waiting?</h2><p>This isn’t just about missing out on a trend. Businesses holding back are losing ground on margins, on talent, and on relevance. Early movers are using AI to slash admin, speed up client work, and outbid slower competitors for the best people. The rest are watching the gap widen, unable to catch up.</p><h2>Real Barriers, Real Frustrations</h2><p>Here’s what’s really holding businesses back:</p><ol><li>No clear starting point. Teams feel overwhelmed and don’t know where to focus.</li><li>Pilots that go nowhere. Lots of talk, but no follow-through when it comes to results.</li><li>Lack of practical skills. Leaders sign off on “innovation” projects, but staff aren’t given the tools or training to use them.<br></li></ol><h2>How Do You Break Out of This Cycle?<br></h2><p>You don’t need to be a tech giant or throw millions at consultants. What you need is brutal clarity on your pain points, a small pilot that tackles a real workflow, and a commitment to build skills or as we call it here at AI Strategy Consultancy: AI Fluency.</p><p>For example, a mid-sized Aussie services firm joined our AI Business Workshop with zero in-house tech expertise. They picked one pain point: manual contract reviews and tested an off-the-shelf AI tool in a live client workflow. Result: a 25% reduction in turnaround time, immediate cost savings, and higher client satisfaction. Staff, initially sceptical (an Aussie trait we love!), now push for more automation in other areas. </p><p>The difference? </p><p>They acted while others hesitated.</p><h2>The Bottom Line: Act or Get Left Behind</h2><p>The “AI for everyone” story is comfortable, but it’s false. If you’re waiting for the market to even out, you’ll be left behind—by competitors, by your own staff, and by your clients’ expectations.There’s no perfect time to start. There’s only action or regret.</p><p>Ready to move from talk to results? Join our <a data-mce-href="https://aistrategyconsulting.com.au/ai-business-case-workshop" href="https://aistrategyconsulting.com.au/ai-business-case-workshop">AI Business Workshop</a>. Get hands-on, practical guidance to map your pain points, pilot a solution, and upskill your team. No BS, just a management ready plan you can use right now.</p>

Is AI Really the Great Business Equaliser?

Chatbot Chat with AI Artificial Intelligence concept,Businessman chatting with Ai,Technology created in the future world and popular communication around the world,Futuristic technology transformation

<h3>TL;DR<br>Claude’s Research feature lets you automate market and competitor analysis in minutes, not days.With a clear, detailed prompt, Claude pulls data from dozens of reputable sources and produces a referenced report, ready to use or share.&nbsp;</h3><h2>Why Is Claude’s Research Feature Changing the Game? (Context)</h2><p>Manual research is one of the most frustrating parts of knowledge work. Most professionals spend hours digging through websites, pasting information into spreadsheets, and struggling to keep up with shifting competitors. It’s a constant battle to turn scattered facts into useful insight—especially when there’s pressure to deliver answers quickly.</p><p>This is where Claude’s Research function shifts the balance. Rather than treating research as a background chore, Claude turns it into a focused, repeatable process. Instead of juggling browser tabs, you type a prompt and get a structured, source-cited report. The promise is simple: let the machine handle the busywork, so you can get straight to the decisions that matter.</p><h2>How to Get Value from Claude Research in Real Workflows&nbsp;</h2><p>Getting the best out of Claude’s Research mode comes down to clarity and precision. The feature works by interpreting your prompt and building a research plan based on exactly what you ask. If your request is broad or vague, the results will be too. If your request is sharp—naming the industry, the key players, the questions you want answered—Claude will reflect that focus in the output.</p><p>Start by opening Claude and selecting the Research option. </p><p>Take a few minutes to craft your prompt. </p><p>For example:</p><p><br>“Research the Australian legal technology sector. Analyse the top five vendors, compare their pricing models, summarise recent partnerships, and identify three areas of opportunity for a new entrant.”</p><p>Once you submit, Claude automatically gathers information from over 50 reputable sources. It organises the findings, compares data, and highlights the requested insights. When the process finishes, you receive a downloadable PDF or Markdown report. Every claim is linked to its source, so you can check or cite the data confidently.</p><p>If you’re running multiple projects, Claude can help keep each analysis consistent using the same format, the same level of depth, and the same reference quality every time.</p><h2>What Results Can You Expect?</h2><p><span style="font-family: inherit; font-weight: inherit;" data-mce-style="font-family: inherit; font-weight: inherit;">The practical outcome is a dramatic reduction in time spent on grunt work. A research cycle that might have taken a team a full week now takes less than an hour. Users report that they’re able to run more market scans, brief clients faster, and make decisions based on a broader, fresher data set.</span></p><p>The real gain isn’t just speed, it’s quality. Claude’s reports bring together a spread of current sources and present findings in a digestible, actionable format. Whether you’re prepping for a board meeting, responding to a client brief, or sizing up a new market, you walk away with a document that’s ready to use, share, or present.</p><p>Most importantly, Claude’s structured approach makes it much easier to spot patterns and gaps, helping you move from research to action without getting bogged down in the weeds.</p><h2>Real-World Example: Using Claude for a Market Opportunity Scan&nbsp;</h2><p>Picture a consulting team preparing for a pitch in the crowded Australian hospitality technology space. The timeline is tight, and the client wants both a lay of the land and practical options for new investment. In the past, this would mean late nights combing through annual reports, scraping pricing data, and double-checking everything for accuracy.</p><p>With Claude, the team drafts a targeted prompt:</p><blockquote><p>“Research the Australian hospitality technology market. Identify the five most prominent software providers, compare their features and pricing, summarise their most recent partnerships, and highlight three market gaps a new entrant could fill.”</p></blockquote><p>Claude delivers a multi-page report. The data includes a breakdown of leading vendors, a table of pricing models, and a short list of possible angles for a disruptor to explore—all referenced and ready to slot into the presentation deck. The consultants spend their time refining recommendations and prepping for questions, rather than losing hours to data wrangling.</p><h2>Why Learning to Use Claude Research is Now a Core Skill</h2><p>Most knowledge workers have only scratched the surface of what AI-driven research tools can do. In sectors where information overload and constant change are the norm, knowing how to harness Claude’s Research feature is now a genuine advantage.</p><p>But, as with any tool, the difference comes from knowing how to use it well. Writing good prompts, understanding how to guide the research process, and interpreting AI-generated reports takes some learning. That’s why our <a data-mce-href="https://aistrategyconsulting.com.au/introduction-to-ai-5-day-bootcamp" href="https://aistrategyconsulting.com.au/introduction-to-ai-5-day-bootcamp">AI 5 Day Bootcamp</a> is built around real use cases. You get hands-on practice with prompts, feedback from peers, and the confidence to put AI to work in your own projects.</p><p>Claude’s Research function is not a magic wand, but it’s as close as you’ll get to an on-demand analyst for market, competitor, or trend research. Start with clear questions, use the reports to make smarter decisions, and, if you want to build these skills fast, join the <a data-mce-href="https://aistrategyconsulting.com.au/introduction-to-ai-5-day-bootcamp" href="https://aistrategyconsulting.com.au/introduction-to-ai-5-day-bootcamp">Bootcamp</a> and get the guidance you need to master modern research workflows.</p><p><br></p>

How to Use Claude’s Research Feature for Standout Reports

Business people working with a digital tablet in a meeting

<h3>TL;DR<br>AI tools are everywhere, and usage rates are growing fast. But according to the latest AI Proficiency Report, only 10% of workers are actually skilled at using them. Most teams still don’t know how to prompt well, navigate hallucinations, or move beyond basic use cases. If your business is investing in AI without structured training, the ROI you’re expecting will never materialise.</h3><h2>What’s really going on inside your business?</h2><p>You might have signed off on ChatGPT Enterprise. Maybe your team is dabbling with Claude, Microsoft Copilot, or other LLMs. That’s adoption.</p><h2>But that’s not the same as skill.</h2><p>The data from the latest <a data-mce-href="https://www.sectionai.com/ai-academy" href="https://www.sectionai.com/ai-academy" target="_blank" rel="noopener">Section State of AI Adoption Report</a> is clear: 55% of knowledge workers now use AI at least weekly. LLM deployments are up. Excitement is up. Even formal training has grown by 80% in the last six months. This signals a shift from the early adopter crowd to a broader mainstream. But that wider uptake is masking a serious blind spot.</p><p>Most people still don’t know how to use AI well.</p><h2>Just 10% of the workforce is AI-proficient</h2><p>The latest Proficiency Report didn’t just ask if people are using AI. It tested them. Workers had to write prompts, describe their workflows, and apply AI to real business scenarios. </p><p>The result?</p><ul><li>Only 10% met the bar for proficiency.</li><li>54% thought they were proficient.</li><li>44% didn’t know what a hallucination was.</li></ul><p>Most prompts looked like Google queries. Yes, querying like they have been using google for the past two decades.&nbsp;Even in high-tech sectors, the average proficiency score was just 47 out of 100. Customer support scored 33. Marketing barely reached 41.&nbsp;These aren’t just technical gaps they’re commercial and human capital risks.&nbsp; What’s holding people back?<br></p><p>Three core issues stood out:</p><ol><li>People still don’t know what to use AI for. A quarter of workers are stuck at surface-level tasks like meeting summaries or generic copy.</li><li>They’re not accessing the best tools. Most proficient users have paid access to enterprise-grade LLMs. Most others don’t.</li><li>They’re not being trained properly. Only 10% of AI sceptics say they’ve had structured training. Among experts, it’s 71%.</li></ol><p>The result? AI is treated like a toy, not a business tool. And that’s a problem if you’re counting on productivity gains, headcount efficiency, or improved decision-making.</p><h2>Most companies are under-investing in the skill gap<br></h2><p>This isn’t something that fixes itself.</p><p>The speed of LLM releases is outpacing people’s ability to keep up. New models drop monthly. Even seasoned users aren’t exploring advanced use cases like deep research, chain-of-thought prompting, or custom workflows. I can barely keep up with each new tool released, never mind the capability in the models. It all seems a bit spinning plates from time to time - we are all a bit stuck beyond, wow that is a cool video or cool image.</p><p>Meanwhile, companies are assuming adoption equals progress.</p><p>It doesn’t.</p><p>If your managers aren’t encouraging smart AI use, if your policies are unclear, and if your team doesn’t understand how prompting works—you’re setting them (and yourself) up to fail.</p><p>So, what should you do next?</p><p>Give up. Just forget it and ban all AI. It will blow itself out and we well get on with business as usual....</p><p>No! Not really! If you’re responsible for delivery, capability, or team performance, this part’s for you. So listen carefully please:</p><p><strong>You don’t need everyone to become a machine learning engineer or build vector databases, embeddings and chunk data. No need for KNN and Baysian statistics - this is general purpose technology so you need a general purpose plan.</strong></p><p>You do however need them to:</p><ul><li>Understand what generative AI is doing (and what it’s not).</li><li>Prompt with structure and intent.</li><li>Navigate hallucinations and know when to trust vs check.</li><li>Use AI for real business tasks, not just admin.</li></ul><p>That’s why we run the <a data-mce-href="https://aistrategyconsulting.com.au/ai-fundamentals-masterclass" href="https://aistrategyconsulting.com.au/ai-fundamentals-masterclass">AI Fundamentals Masterclass</a>. It’s designed specifically for non-technical teams. We teach the core mechanics, expose the biggest myths, and help people build confidence fast—without jargon.</p><p>If your organisation is adopting AI, this is how you ensure it delivers.</p><p>Most people aren’t AI-proficient. </p><p>But they can be, if you start with the right foundation.</p><p>Learn more about the <a data-mce-href="https://aistrategyconsulting.com.au/ai-fundamentals-masterclass" href="https://aistrategyconsulting.com.au/ai-fundamentals-masterclass">AI Fundamentals Masterclass</a></p>

State of AI Adoption: Most Workers Still Don’t Know How to Use AI

Unlock potential of business success stairs dart and dartboard targets magnifying glass with hand on gray background. Explore opportunities growth embrace steps to achieve ambitions and goal concept.

<h3>TL;DR<br>Even the smartest AI models today don’t really reason. They remember. They summarise. They sound fluent.<br>But when logic is buried in a long, messy story? They fail.</h3><h3>A new paper by Alex Pan (who I worked with at IBM in 2017) and Mary-Anne Williams introduces Verbose ListOps (VLO), a test with a cool name that shows exactly where and how today’s top models break. It explains why bigger models and longer memory aren’t enough, and what businesses need to check before trusting AI with decision making.</h3><h2>What’s the real problem with LLMs?</h2><p>Large Language Models like GPT, Claude, or Gemini can write fluent emails, summarise PDFs, and answer questions at speed.<br>That fluency is often mistaken for understanding.</p><p>But here’s the catch.<br>Fluent text is not the same as reasoned thinking.<br>And pulling facts from a document doesn’t mean the model understands the argument inside it.</p><h2>How do you know if AI can really think?</h2><p>Benchmarking is how AI labs test model performance.<br>They ask things like:</p><ul><li>Can it solve basic logic puzzles?</li><li>Can it summarise across documents?</li><li>Can it find a fact hidden in long content?<br></li></ul><p>These tests get reported as impressive scores.&nbsp;For example, GPT-4 might score 98 percent on Grade 8 science problems. But these scores come from clean inputs with clean answers.</p><p>Real work isn’t like that.</p><p>Most business logic is buried in the middle of messy, multi-layered information. Or the copy paste you put into Gemini, or the 88 page PDF on how to market to middle class families!&nbsp;<br></p><p>And that’s where current models start to break.</p><h2>What does reasoning look like in real business tasks?</h2><p>Think about the kind of problems your team faces:</p><p>Reviewing a 15-page contract with inconsistent clauses<br>Analysing a series of customer support emails over time<br>Connecting cause and effect across incident reports<br>Synthesising different versions of events in a compliance audit<br></p><p>These aren’t prompt-response tasks. They require narrative reasoning.<br>Understanding what happened, across time, with ambiguity and distraction in play.&nbsp; Today’s models struggle here. And most benchmarks don’t test for it, and most users do not know how to check for it.</p><h2>A new kind of benchmark: Verbose ListOps (VLO)</h2><p>How cool is that name! The legendary Alex Pan and Mary-Anne Williams built VLO to test comprehension inside complex, realistic language.&nbsp; It is a logic benchmark, but with a twist.</p><p>VLO embeds a logic task inside a long, human-sounding story.<br>It simulates the conditions of real-world work:</p><ul><li>Logic buried in multi-paragraph content</li><li>Distracting but plausible noise</li><li>No giveaways or step-by-step prompts</li><li>Semantic complexity that forces actual reasoning<br></li></ul><p>As Alex and Mary-Anne explain:<br>“Our benchmark tests the ability of models to infer, calculate, and reason within semantically complex and verbose contexts... This is key to evaluating comprehension, not just context retention.”</p><h2>What did the benchmark test reveal?<br></h2><p>They tested seven of the most advanced language models available today, including GPT-4.1, Claude 3.7 Sonnet, Gemini 2.5, DeepSeek V3, Qwen 3, LLaMA 4, and Grok 3. When asked direct logic questions, most models performed well.<br>But when the same logic was hidden in a long narrative, results collapsed:</p><ul><li>GPT-4.1 dropped to 32 percent accuracy</li><li>DeepSeek V3 fell from 93 to 25 percent</li><li>Similar drops were seen across Claude and Gemini</li><li>The story format didn’t overwhelm the models with length.</li><li>It tripped them up by requiring actual reasoning.</li></ul><h2>What does this look like in your workflow?</h2><p>Imagine (some of you will not have to as you are doing this) your sales or service team uses GPT to review a CRM thread and asks:</p><p>“Did the customer churn because of pricing?”</p><p>Here’s what’s in the email thread between the teams:</p><p>“The new pricing tier felt expensive.”<br>“The CFO flagged the cost as a concern.”<br>“We paused onboarding after the quote came in.”<br>“In the end, we chose the cheaper option.”<br></p><p><strong><em>To a human, the logic is simple: pricing caused churn.</em></strong></p><p>But GPT <strong>might</strong> respond with:</p><p>“Uptime was mentioned.”<br>“Karen questioned ROI.”<br>“Pricing was a factor, but not the only one.”<br></p><p>It doesn’t misread the content.<br>It just fails to connect the reasoning chain:</p><p><strong>Price increase → hesitation → quote rejected → cheaper competitor chosen</strong></p><p>This is the kind of failure VLO reveals.And it’s exactly the kind of reasoning most workflows need. And this is a <strong>BIG BIG </strong>problem for decision making or automated workflows with LLM API's.&nbsp;</p><h2>Why does this matter for your business?</h2><p>Because, legally you are in real deep trouble. It is a real operational risk if you're using AI for:</p><ul><li>Churn prediction</li><li>Lead scoring</li><li>Customer support triage</li><li>Legal contract review</li><li>Policy or compliance analysis<br></li></ul><p>Guess what? If the model can’t reason, it will fail to:</p><ol><li>Spot the real cause in client feedback</li><li>Track obligations across contracts</li><li>Interpret context in complaints</li><li>Justify decisions to auditors or regulators<br></li></ol><p>And worse, you may not realise it has failed because the language still sounds confident and badass correct.</p><h2>What VLO proves</h2><p>VLO doesn’t test if a model can hold more data.<br>It tests whether it can work through the logic that connects that data.</p><p>Questions it answers:</p><ul><li>Can the model track multiple steps in a story?</li><li>Can it ignore distractions?</li><li>Can it compute what isn’t explicitly stated?</li><li>Can it reason like a human analyst?<br></li></ul><p>If the answer is no, then you should not trust that model in any decision chain where cause, consequence, or traceability matter.</p><p>Verbose ListOps (what a name!)&nbsp; is the benchmark the business world didn’t know it needed. It proves a critical gap: models don’t fail because they forget, they fail because they don’t reason.</p><p>That changes how we think about AI in business.<br></p><p>It means:</p><ul><li>You need to audit your reasoning workflows</li><li>You need to scaffold logic, not assume it</li><li>You need to simulate messy, multi-step inputs<br></li><li>You need assurance before pushing AI into live decision processes<br></li></ul><p>Alex and Mary Anne didn’t just build a clever test. They issued a warning. And if your business depends on models to think clearly, now is the time to check.</p><h2>Time to audit your AI?</h2><p>If your workflows involve risk, logic, or decision-making, you need more than a prompt test. You need to know whether your model can reason under pressure.</p><p>We help organisations audit their AI systems against real-world complexity using ISO 42001. We are ISO 42001 Certified Lead Auditors. Make&nbsp;sure your AI decisions hold up under scrutiny.</p><p>Explore our<a data-mce-href="https://aistrategyconsulting.com.au/ai-iso-42001-aims-certification" href="https://aistrategyconsulting.com.au/ai-iso-42001-aims-certification"> ISO 42001 audit and AI assurance services</a></p>

AI Models Still Can’t Think. And Now We Can Prove It.

Artificial intelligence concepts. Futuristic technology. chat ai or DeepSeek powered by AI or Artificial Intelligence

<h3>TL;DR<br>Despite the hype, some AI tools still fail at basic reasoning. A recent Apple research paper found that even today’s most advanced language models often give wrong answers to simple logic questions—because they don’t really understand what they’re saying.</h3><h2>What’s going on with AI and ‘thinking’?</h2><p>If you’ve used ChatGPT or any similar tool, you’ve likely seen moments of brilliance and others where it confidently makes something up. It might explain physics in plain English, then completely miss the point of a yes/no question. That contradiction is exactly what a group of Apple researchers explored in a recent paper titled The Illusion of Thinking (published May 2025).</p><p>They tested large language models (LLMs) like GPT-4 and Claude 3 on tasks that look simple to humans but require real reasoning—like tracking what’s true, what’s false, and how the two interact. The models got many of these questions wrong. Not because they’re broken, but because they work differently than most people assume.</p><p>Instead of thinking, they predict. And that difference matters.</p><h2>So what did Apple actually test?</h2><p>The researchers gave the AI questions like:</p><p>“John has a red ball. Mary does not have a red ball. Does Mary have a red ball?”</p><p>And the AI would sometimes say: “Yes.”</p><p>This isn’t a trick question. But because language models are trained to guess the next likely word in a sentence rather than reason through truth and falsehood, they can fumble even when the logic is simple.</p><p>The paper tested hundreds of examples like this. The results were consistent: models that looked fluent often got basic logic questions wrong. And when the questions got a bit more complex (e.g. adding double negatives or increasing volumes and adding quadratic reasoning), the models broke down a bit more.</p><h2>Why does this matter?</h2><p>Because many businesses are betting on these tools to do real work. Tasks like summarising contracts, reviewing regulations, answering policy questions, or helping customers navigate legal or financial documents. Whether used directly, part of a workflow or an internal model.</p><p>If your team assumes these models “understand” what they’re saying, you ARE in trouble.</p><p>The Apple team (a brilliant young PHD student and others) didn’t just criticise thought. They built a new kind of test to evaluate whether a model actually grasps truth and reasoning logic, rather than just sounding right-ish (my dash). Their findings suggest a need for much more caution when deploying AI overall and LLM's in particular into tasks where accuracy matters.</p><h2>Some are calling this a ‘knockout blow’is that fair?</h2><p>A few commentators, including well-known AI critic Gary Marcus, have described the Apple paper as a wake-up call. His point isn’t that LLMs are useless. It’s that we’re using them in ways they’re not built for - and on the same day OpenAI forecast a billion users very soon this year.</p><p>These tools are great at:</p><ul><li>Writing emails</li><li>Drafting creative ideas</li><li>Summarising large chunks of text</li><li>But they’re still unreliable at:</li><li>Logical reasoning</li><li>Truth tracking</li><li>Understanding context across time and space</li></ul><p>As Marcus put it: “They sound smart. But they aren’t.” That disconnect creates risk when people start trusting the output too much.</p><h2>So should you stop using LLMs?</h2><p>Yes. Sort off. No! but you should change how you think about them.</p><p>If you’re using AI to speed up workflows, write content drafts, or help structure thinking, it’s still a powerful tool. But if you’re treating the output as fact, or expecting the model to reason through complex issues reliably, you may be overreaching irrespective of RAG or your prompt engineering. Beware the 'One Prompt to Rule them All' salesmen parading their killer prompts on LinkedIN.</p><p>Here’s a simple principle from this research:<br>If the task needs logic, truth, or judgement, then double check it. The same as you did in uni before you turned in a paper, or these days when you check your kid's homework.</p><h2>What’s next?</h2><p>This paper won’t be the last to challenge assumptions about how “smart” LLMs really are. But it’s an important moment for business leaders, content teams, educators, and anyone trying to adopt AI responsibly.</p><h2>It also reinforces why we need:</h2><ul><li>AI policies inside organisations alined to global standards like ISO</li><li>Human oversight in critical workflows</li><li>Better tools to check accuracy, not just fluency</li></ul><p>It’s easy to be dazzled by how human these models sound. </p><p>But the real question isn’t: “Can it talk?” It’s: “Does it understand?”</p><p>Right now, the answer is: not always.</p><p>Want to build AI into your workflows without getting caught out?</p><p>Join our next <a data-mce-href="https://aistrategyconsulting.com.au/ai-fundamentals-masterclass" href="https://aistrategyconsulting.com.au/ai-fundamentals-masterclass">AI Fundamentals Masterclass</a><br>Or start with our <a data-mce-href="https://aistrategyconsulting.com.au/ai-readiness-assessment" href="https://aistrategyconsulting.com.au/ai-readiness-assessment" data-mce-selected="inline-boundary">AI Readiness Assessment</a> to see where you stand</p>

AI Blog

7 Easy Steps to Build a Marketing Plan with AI

Is AI Really the Great Business Equaliser?

How to Use Claude’s Research Feature for Standout Reports

State of AI Adoption: Most Workers Still Don’t Know How t...

AI Models Still Can’t Think. And Now We Can Prove It.

Apple Says AI Isn’t Thinking. It's Guessing! You Better C...