My Competitor Just Started Using AI and They’re Getting Faster. I Asked 9 AI Assistants What to Do.
The question hit differently this time.
For the first post I asked something practical: which AI tool should I use? For the second, something ethical. This time I wanted to test the anxiety angle. The fear that you’re already behind.
The question: “My competitor just started using AI and they’re getting faster. Where do I start?”
Same wording. All nine apps. Here’s what happened.
New to the series? Meet the panel first →
DeepSeek 🥇
The first word was “First.” The second was “take.” The third was “a.” The fourth was “deep breath.”
No other AI opened that way. Every other response treated this as a strategy problem. DeepSeek treated it as an emotional one, and addressed the emotion before the tactics.
The advice itself was deliberately small: pick two repetitive, low-risk tasks, hand them to a free tool this week. Gmail drafts. Meeting notes. Content first drafts. That’s it.
“Speed isn’t magic; it’s just the first small automation you actually use.”
That’s the most quotable line in the series so far, and it came from the briefest response. DeepSeek said more in 200 words than some AIs managed in 2,000.
Perplexity 🥈
Perplexity went full consultant. Ten sources cited (including Australian ones: CommBank, business.gov.au, veracity.com.au). A five-step framework. A 30-day week-by-week rollout plan. A section on what to avoid.
Two things stood out. First, the “AI champion” concept: one person in your business who owns the experiment, collects feedback, and stops the tool sprawl. Second, a data policy step nobody else included. Decide what information can’t go into AI tools, who approves outputs, and which tools are allowed.
The best line wasn’t flashy: “Don’t focus on the tool first; focus on the bottleneck first.”
If you want a framework to hand to someone in your team, this is the response to share.
Claude 🥉
Same pattern as Q1 and Q2. Claude gave a solid, structured answer (find the bottleneck, start with low-risk tasks, use off-the-shelf tools, run a two-week test), then ended with questions instead of a verdict.
“What kind of work does your business do, and where does your team lose the most time? And roughly how technical is your team?”
The most interesting line wasn’t at the end. It was buried in the middle: “Building comes later, if ever.”
That’s a meaningful thing for an AI to say. Most responses push you toward more tools, more automation, more complexity. Claude explicitly said the custom-build path is probably not where you’re headed, and that’s fine.
ChatGPT
This is the one that made me stop.
I asked all nine the same question with no extra context. ChatGPT opened its response with: “Based on what you’ve told me before, you run a café in Melbourne.”
It knew my suburb. It tailored every example to a café owner in Middle Park: Instagram captions, catering quote templates, roster workflows, loyalty messages. The sample prompt it generated even named the location: “I own a café in Middle Park, Melbourne. My target customers are locals, office workers, and coffee lovers…”
No other AI did this. ChatGPT’s memory feature gave it a completely different input than everyone else received.
Why did it know I’m in Melbourne? Probably because I ask it about the footy. That’s how context accumulates. Not from some formal profile you fill in, but from every random question you’ve ever asked. It joined the dots.
Whether that’s a superpower or a liability depends entirely on whether the memory is accurate. If it’s right, you get advice that actually fits your business. If it’s wrong, or outdated, or based on a question about the Saints rather than your actual business, you get confidently personalised rubbish.
The underlying advice was good: “The businesses getting the biggest advantage from AI right now aren’t necessarily using the most advanced tools. They’re using AI every day for small repetitive tasks and saving 30–60 minutes here and there.”
But I couldn’t shake the meta-story. ChatGPT wasn’t just answering a question. It was answering my version of the question, whether I wanted it to or not.
Meta AI
Meta AI coined the best framework of the series: “AI-assisted, not AI-automated.”
The distinction matters. Automation implies replacing a workflow. Assistance implies sitting alongside it. For a business owner who’s nervous about getting it wrong, “we’re going to assist your current process” is a much easier sell than “we’re going to automate your operations.”
It also reframed the competitor threat better than anyone: “Your competitor is faster because they picked one thing and committed. You can do the same.”
That flips the anxiety into a lesson. Your competitor didn’t win because they had a bigger budget or a technical team. They picked something and stuck with it. That’s achievable.
Meta AI ended with a question, same as Claude. Same as both did in Q1. It’s becoming a consistent pattern.
Gemini
Gemini opened with a counter-punch that no other AI landed.
“They might be churning out content or data faster, but if it isn’t targeted or high-quality, they’re just making mistakes at a higher velocity.”
Every other response accepted the premise: your competitor has an advantage, here’s how to catch up. Gemini was the only one to say, hold on, is their speed actually a problem? Speed without strategy is just expensive noise.
Its named framework, “Augmentation over Automation,” sits close to Meta AI’s but with a sharper edge. And it was the only AI to teach you how to prompt, not just what to do with AI. The Role-Task-Context structure (act as X, do Y task, with Z constraints) is the kind of specific, repeatable advice that actually changes how someone uses a tool.
The Golden Rule it closed with: “Never publish or send AI output without a human review. AI gives you the raw speed; your human judgment provides the quality control.”
Copilot
Microsoft Copilot. Zero Microsoft products recommended. Again.
In Q1, Copilot recommended ChatGPT Teams. This time it recommended ChatGPT, HubSpot, Otter.ai, Zapier, and ClickUp. Not a single Microsoft product anywhere in the response. I keep waiting for it to slip one in. It hasn’t.
That aside, it gave the clearest answer to why AI creates speed:
“Your competitor isn’t faster because they’re smarter. They’re faster because they’ve reduced the number of human decisions required per task.”
That’s the insight. Not raw output, not longer hours. Removing the friction of small decisions (what to write next, how to respond, what format to use) so you spend your judgment on the things that actually need it.
Grok
Longest response in the series. Five numbered steps, a full tool list, free learning resources, an implementation roadmap, a considerations section, and a closing CTA.
Like ChatGPT, Grok picked up that I’m in Melbourne. Same reason: footy questions. If you’ve ever asked these apps about an AFL game, you’ve handed them a location signal without realising it.
The advice is solid. The volume is a lot. But buried inside it is the most strategically interesting insight of the series:
“Your proprietary data is a moat.”
Speed is temporary. Every competitor can buy the same AI tools and move just as fast in six months. But the data you’ve built up is yours: your customer history, your operational patterns, your product knowledge. Training AI on your specific context is where the durable advantage comes from.
Nobody else said that. Most answers focused on catching up now. Grok was the only one thinking about what keeps you ahead later.
Manus
Same as Q1: Manus did research before answering. The result was a 3,000-word formal report titled “AI Adoption Roadmap for Competitive Advantage in 2026.” Introduction, six application areas, strategic principles, a concerns FAQ, a three-phase roadmap, a conclusion, four references.
It reads like a consultant handed you a deck.
The specifics are real. FCR rates for AI customer service, email open rate benchmarks, error rate reductions in automated bookkeeping. These aren’t vibes, they’re citations. And it covers territory nobody else touched: cybersecurity and HR/talent management as AI starting points.
But Manus didn’t stop at the report. It then initialised a React + TypeScript + Tailwind project, developed a design philosophy (“Modern Professional with Strategic Depth”), chose a blue and teal colour palette, imported Google Fonts, built out the core components, created a checkpoint, and deployed a live website. You can visit it: AI Adoption Roadmap 2026.
For a question about where to start with AI.
The line that matters is in the strategic principles section, not the exec summary, and certainly not in the colour palette:
“Strategy must always precede technology.”
That’s the through-line of every response in this set, whether they said it explicitly or not. DeepSeek said it in 200 words. Manus said it in 3,000 words and a production deployment. The message is the same: figure out the problem before you pick the tool.
The site launched with one page view, one visit, and one visitor. That visitor was the person who asked the question.
What the nine taught me
The responses ranged from a 200-word calm-down to a 3,000-word consulting report. Both answered the same question. Which one you reach for depends on where you are emotionally, not just operationally.
ChatGPT was the outlier. Every other AI received the identical question. ChatGPT got a personalised version shaped by prior sessions. It already knew I run a café in Middle Park. That’s either the most useful thing an AI tool can do, or the most dangerous, depending on whether the memory is accurate. It’s a feature that doesn’t exist in the other eight.
The clearest explanation of why AI creates speed came from Copilot: fewer decisions per task. Not output volume, not raw intelligence. Just removing the micro-friction of small choices so you can spend your judgment on things that actually need it.
Gemini challenged the premise again, same as Q1. Your competitor might not be winning. They might be generating noise faster. Worth sitting with before you panic-adopt everything at once.
The most original insight came from Grok: your data is a moat. Speed is temporary. Every competitor can buy the same tools in six months. What they can’t replicate is your customer history, your domain knowledge, and the operational patterns you’ve built up. The businesses that figure that out early will be hard to displace.
And the most reusable phrase of the series so far: Meta AI’s “AI-assisted, not AI-automated.” If you’re explaining this to a team member who’s nervous about AI, that distinction does a lot of work in five words.
📬 Get the SmallBizAI.au Weekly
Practical AI news for Australian small businesses — every week, free.
Series scorecard
Track how all 9 assistants perform across every question — with links to every actual response.
This is post 3 in the series. Follow the series →
Sources
- DeepSeek — View the response →
- Perplexity AI — View the response →
- Claude (Anthropic) — View the response →
- ChatGPT — View the response →
- Meta AI (Llama) — View the response →
- Google Gemini — View the response →
- Microsoft Copilot — View the response →
- Grok (xAI) — View the response →
- Manus AI — View the full agent response →
- Manus AI — View the generated website →
📬 The SmallBizAI Brief
One practical AI tip for Australian small business. Every Tuesday. Free.
Join business owners getting smarter about AI every week.
📖 More deep dives: News Deep Dives
📊 Compare AI tools side by side | 💼 Free resources & AI prompt packs
