I was just speaking with a coworker who asked an AI for help in opening a brokerage account, and he told me how it doled out some general tips before recommending a specific investment firm, praising its easy-to-use website and even serving up a ready-to-click URL.
Wait, my colleague wondered: Was that a sponsored answer?
I’m pretty sure the answer my friend got wasn’t actually an ad; more likely, the AI had searched some how-to sites on the web and may have picked up one written by that particular brokerage.
Welcome to another edition of Prompt Mode, your weekly AI newsletter.
I’m your host, Ben Patterson. Each week on Prompt Mode, I’ll be serving up analysis of the AI trends that matter to everyday users like you and me. Stay tuned for practical AI tips, hands-on experiences with the latest AI tools, and–you guessed it–prompts to help you get the most out of your AI assistants.
Thanks for reading, and if you like what you see, just sign up right here.
But that very doubt about the AI’s reply is the point. We’re already wary (as we should be) of AI doling out hallucinations in its conversations with us. Soon, we’ll be wondering more and more whether an AI’s answer has been bought and paid for.
Just this week, Google demonstrated how it’s starting to blur the line between AI answers and ads. In a new “Conversational Discovery” ad that Google is testing in AI Mode, Gemini’s answer to a user question would include a sponsored, Gemini-written “creative” that’s “tailored to that search,” basically weaving the ad into Gemini’s answer.
In another AI Mode ad unit that Google is testing, a sponsored ad crafted by Gemini could be included as a “Highlighted Answer” to a question, appearing alongside other non-sponsored content.
In each of these ad units, Google says that Gemini will serve the role of an “AI explainer” that “evaluates and synthesizes information about a product or service” and “displays that context alongside the advertiser’s creative.” Gemini’s “coherent, independent response ensures transparency and builds trust,” Google continues.
Now, it’s important to note that Conversational Discovery and Highlighted Answer ad units (as well as new AI-powered Shopping ads in Google Search) are clearly labeled as “Sponsored.” And in the case of the “conversational” ads, the sponsored Gemini content appears in a marked-off section just below the main, non-sponsored answer.
Still, these new Gemini-written ads are going a step beyond the familiar static ad units served in ChatGPT. With Google’s new ads units, Gemini is integrating the sponsored content into the context of the user’s question. That’s a big–and probably inevitable–change.
And while Google is being scrupulous about labeling sponsored Gemini answers as ads, I can imagine the temptation for other marketers to make those “sponsored” bugs smaller and smaller, while weaving AI-written ad copy into AI answers tighter and tighter.
More in AI this week
- My colleague Mark Hachman gave Google’s new AI glasses a try and liked them. The integrated camera is still creepy, though. (PCWorld)
- Why do speakers keep praising AI during commencement addresses? Don’t they know what will happen by now? (AP)
- This pint-sized AMD box can be your own personal ChatGPT, running powerful LLMs locally rather than in the cloud. It’ll cost you, though. (PCWorld)
- Google’s new Spark AI agent will tirelessly do your bidding, running 24/7 in the cloud and accessing (if you let it) all your core Google data. But it’s only for Google AI Ultra users, meaning it’ll cost you at least $100 a month. (PCWorld)
- Alongside its Spark reveal, Google also made big changes to how it calculates Gemini usage limits, with its AI subscriptions moving to “compute-based” usage. (PCWorld)
- ChatGPT can now access your financial accounts if you’re willing. Here’s why I’m not. (PCWorld)
- A company gave four AI models $20 each to start their own radio stations. Needless to say, the experiment quickly went off the rails. (CNET)
Prompt of the week: The “anti-sycophany” prompt
AI just loved to praise us. If I had a nickel for each time ChatGPT told me I’d “hit the nail on the head” or made a “beautiful” observation, I’d go buy lots of things.
Flattery is so ingrained in AI models that it can be hard to get them to stop, even when you tell them directly. A better approach is a prompt that’s so blunt, so direct, it acts as a shock to AI’s system, the equivalent to Cher’s “snap out of it!” moment in Moonstruck.
This “anti-sycophany” prompt doesn’t literally slap an AI in the face, but it will get it to drop the praise and–finally–tell it to you straight.
That’s all for now!
Thanks for reading the latest issue of Prompt Mode. Want more next week? Don’t forget to sign up to start receiving this newsletter in your inbox. See you next time!
