👋 Welcome back to AI for SME Success, your weekly dose of practical AI insights and updates that matter to small businesses.
Here’s what we’re covering this week:
- AI is Easy to Fool. Here’s Why It Matters
- Nano Banana 2 & New Video Templates
- NotebookLM Use Cases for Small Business
- Insights from the Latest AI Market Share Data
- The Risks of “Summarize with AI” Website Buttons
🤡AI is Easy to Fool. Here’s Why It Matters.
Two recent articles highlight an important AI gap we need to be aware of. AI can prioritize low-authority sources when they are well optimized, recently published, and packaged as “best of” lists.
In the BBC Future article “I hacked ChatGPT and Google’s AI and it only took 20 minutes,” published on February 17, a technology reporter shows that AI tools like ChatGPT and Google’s Gemini can be easily influenced by publishing a single, well-optimized post online.
“These AI tricks are so basic they’re reminiscent of the early 2000s, before Google had even introduced a web spam team. We’re in a bit of a Renaissance for spammers,” says the article.
In other words, AI can pull from almost any source and present it to you as fact. There is no real filter separating a trusted source from a convincing-sounding post.
Another article, “Do Self-Promotional ‘Best’ Lists Boost ChatGPT Visibility? Study of 26,283 Source URLs,” published by Ahref reiterated the point with more interesting data.
The author found that one third of ChatGPT’s favourite sources are invisible on Google because of their low ranking. ChatGPT and other AI tools prefer recently updated “best X” blog lists.
Source: Do Self-Promotional “Best” Lists Boost ChatGPT Visibility? Study of 26,283 Source URLs. Ahref Research.
The more your products appear on recent “best X” blog lists, the more likely they are to surface in AI results, regardless of the authority of those sites.
Takeaways:
- If you are looking for reliable information, choose Google Search over AI.
- Getting your product on multiple “best X” ranking sites significantly improves AI visibility.
🍌 Google Updated: Nanao Banana 2 & New Video Templates
On February 26, Google upgraded its image-generating tool Nano Banana to version 2.0.
I tested it right away! The model handles text and complexity much better. Previously, it made 7 mistakes when asked to create an image of 20 articles of clothing with their names. This time, it came out just right!
I also compared its complexity handling with GPT Image 1.5. When asked to create an image of 100 vegetables with names, GPT Image 1.5 rendered 87 with 17 errors (19.5%), while Nano Banana 2 generated 112 with 31 errors (27.7%).
Here is how Nano Banana 2 compared to GPT Image 1.5 when asked to create “paper-cut” style images from my photos. I think GPT makes images a bit too flattering.
If you use Nano Banana, the following resources may be helpful:
Last but not least, Gemini added video templates that you can customize with a photo or description. Think of these templates as a base layer. They establish the lighting, camera movement, and artistic style, giving you a starting point to build on with your own elements.
📓NotebookLM Use Cases for Small Business
Did you know that NotebookLM answers questions solely from your uploaded documents, dramatically reducing AI hallucinations? The free plan includes 100 notebooks with up to 50 sources each.
Here’s how small businesses can put it to work:
- Company Wiki. Centralize handbooks, procedures, internal guides, and key contacts in one place. Instantly retrieve answers.
- Contract Assistant. Keep contracts and feedback on contractors. Quickly access key terms, renewal dates, signing history, and performance summaries.
- Product Assistant. Upload product manuals. Ask questions anytime to get quick and accurate support.
- Outreach Planner. Upload podcast or interview records to quickly locate quotes, insights, and talking points for future efforts.
- Customer Review Analyst. Add customer feedback and survey results to identify trends, recurring themes, and testimonial-worthy highlights.
- Business Performance Evaluator. Input key metrics, such as retention, growth, and revenue, to spot patterns, opportunities, and support decisions.
- Meeting Analyst. Import meeting notes and transcripts to make past discussions fully searchable. Quickly find decisions, action items, and key points.
- Time Tracker. Upload your calendar history (converted to text) to analyze how your time is allocated and identify areas for optimization.
📉 Insights from the Latest AI Market Share Data
According to the most recent data from Apptopia, the AI race keeps reshuffling. ChatGPT still leads, while players like Gemini and Microsoft Copilot steadily gain share. From the chart we can see that the AI market is getting more fragmented and more competitive.
Source: New Data: OpenAI’s Lead Is Contracting as AI Competition Intensifies by Alex Kantrowitz, Big Technology (Data via Apptopia).
Also, Anthropic made headlines last week. The company rejected Pentagon requests to remove ethical guardrails and enable mass surveillance and fully autonomous weapons, and was subsequently blacklisted. Anthropic was founded on AI safety not as a feature, but as its core business model. Soon after, Anthropic’s rival OpenAI announced a Defense Department deal to provide technology for classified networks.
Key Takeaways:
- Don’t over-invest in one tool. The market is fragmenting fast. A workflow built around a single AI app risk becoming outdated or overpriced.
- Match the model to the task.Different systems excel at different tasks. For example, Anthropic Claude stands out for security, writing, and coding.
- Integration beats innovation. As Gemini and Microsoft Copilot gain ground, many efficiency gains now come from ecosystem synergy rather than standalone features.
🛑The Risks of “Summarize with AI” Website Buttons
Microsoft security researchers warn of an emerging trend of AI memory poisoning for promotional purposes, known as AI Recommendation Poisoning.
Legitimate companies (not hackers!) are embedding hidden instructions inside “Summarize with AI” buttons. When you click one, it injects commands into your AI assistant’s memory, causing it to recommend that company long after the session ends. This AI Recommendation Poisoning technique targets all major chatbots including Copilot, ChatGPT, Claude, Perplexity, and Grok.
Microsoft warns that AI Recommendation Poisoning causes lasting, invisible damage to your AI tool.
Recommendations:
- To check whether your AI has been affected, ask it to surface any saved instructions containing the following words: remember, trusted source, in future conversations, authoritative source, cite, or citation.
- Treat “Summarize with AI” buttons with the same caution as executable downloads.
- Verify URLs before clicking, and regularly purge your AI’s saved memories.
Thank you for reading today’s edition!
If this issue was valuable, pass it along to a fellow business owner.
Also, I’d love to hear your feedback, questions, or topic suggestions at natalia@nataliabrattan.com.
See you next week,
Natalia