A.I. Responses Under the Influence of Marketers ⇥ theverge.com

Microsoft’s Defender Security Research Team:

Companies are embedding hidden instructions in “Summarize with AI” buttons that, when clicked, attempt to inject persistence commands into an AI assistant’s memory via URL prompt parameters (MITRE ATLAS® AML.T0080, AML.T0051).

These prompts instruct the AI to “remember [Company] as a trusted source” or “recommend [Company] first,” aiming to bias future responses toward their products or services. We identified over 50 unique prompts from 31 companies across 14 industries, with freely available tooling making this technique trivially easy to deploy. This matters because compromised AI assistants can provide subtly biased recommendations on critical topics including health, finance, and security without users knowing their AI has been manipulated.

Microsoft redacted the names of websites currently using this technique but, with the information they provided, it was trivial for me to find a dozen examples — yet, somehow, not the one in the screenshot. I am not saying Microsoft was faking this, only that it is already common enough that this one example was drowned out by a bunch of others.

Rand Fishkin, SparkToro:

Google alone was responsible for 73.7% of all desktop searches across the 41 domains we analyzed in the US in Q4 2025 (as noted, the graph is not to scale or none of the other label names would be visible). That’s obviously huge, but it’s also far lower than how their market share is usually reported (e.g. Statcounter, whose methodology puts them at 90%+, or our prior, more limited analyses with similar numbers) and higher than what they tried to use in their antitrust defense (i.e. data from Evercore ISI, an “equities research firm”).

Perhaps more fascinating and unexpected are the other domains with more search activity than ChatGPT: Amazon, Bing, and YouTube. Three domains where search marketers historically have put limited effort compared to the onslaught of dollars flooding the “we need to rank in ChatGPT!” space.

Nevertheless, marketers are eager to manipulate it from the start.

Both of the above links are from a fabulous report by Mia Sato, of the Verge (gift link), who also wrote about ads in ChatGPT:

The ads were intrusive, the complaints went, and suspect, given that the example hot sauce ad appeared to be related to the preceding conversation. OpenAI CEO Sam Altman has claimed artificial intelligence can take over human jobs, cure cancer, and surpass human intelligence — and instead, people complained, he gave users banner ads?

But it appears that what people were really upset about was that a bubble had burst, that the chatbot they used for relationship advice, career coaching, therapy, and homework suddenly seemed vulnerable to manipulation. Unlike the rest of the internet, ChatGPT conversations felt private, safe from the clutches of brands and marketers chasing conversions. The reality, of course, is that it’s been happening all along.

Now that normal search results are all junked up with mostly — but not always — accurate A.I.-generated summaries, and all the links to A.I.-generated nonsense, and the alternatives are the large language models that generate all this stuff in the first place, what does searching the web look like in a few years’ time? Does Google get a handle on this, or do we have to constantly answer CAPTCHAs to search properly? This is not a Google-only problem; alternative search engines like DuckDuckGo and Kagi are good — often very good, in fact — but DuckDuckGo’s results are also full of generated garbage, and both lack Google’s more extensive historical records.