Mihai Ionescu’s Post

As an EU citizen, I'm protected from LinkedIn using my data to train its AI. But I don't care. I know it's an unpopular opinion, but please allow me to explain. And if you spot gaps in my logic, I would be grateful if you highlighted them in the comments. First of all, what data are we talking about? Here’s what LinkedIn says: "This could include your use of the generative AI (AI models used to create content) or other AI features, your posts and articles, how frequently you use LinkedIn, your language preference, and any feedback you may have provided to our teams." The social network has always collected this kind of data as part of its user agreement and used it to analyze and modify how the platform works. The only news is that they have included AI in the process to make it more efficient. But what if LinkedIn creates a publication machine based on our precious posts? The quality of content on LinkedIn varies like grains of sand under a microscope. Moreover, it is often driven by trends that come and go, with no objective criteria set in stone. In fact, there's no way for AI to assess content quality beyond grammar. Would a post with many reactions, shares, and comments automatically be considered high quality? Hardly. People respond to posts for many different reasons. Aside from content provoking a genuine reaction, users engage because they: - seek the poster's attention - want to please the poster - seek attention from their network - are the poster's real-life friends or family - are part of an engagement pod - are bots - are responding to a call to action (e.g., "comment to get my one-of-a-kind sales process PDF"). All these motives have little to do with the content's quality. And I believe LinkedIn's team understands this very well. Could the "eyeball time" each post receives be a measure of quality? This would also be an unreliable metric because people with many subscribers will naturally attract more views than others. And, again, users follow others for reasons often unrelated to content quality. Finally, our posts are in the public domain and are even accessible through external search engines like Google. So, anyone can use our content to train their AI to post on LinkedIn or elsewhere. This is why I don't care about this platform training its AI on our content. Again, if you see issues in my logic, please highlight them in the comments. I'm always open to learning. #LinkedInAI #LinkedInEthics or whatever

https://lnkd.in/erPXd7qD Thank you, OpenAI and co., for your help in making the "dead internet theory" come true. Thanks to your oh so helpful garbage, the internet is increasingly becoming a place no actual human even wants to interact with. What remains is online marketplaces (themselves being completely riddled with AI-generated garbage), search engines (same BS) and "social" media, with Facebook and X increasingly being co-opted by the alt-right (and Facebook increasingly consisting of bots interacting with unmoderated pages posting an endless barrage of poorly AI-generated BS, creating "engagement" that nobody needs), while LinkedIn is a completely anti-human, alien space where CEOs keep congratulating themselves and each other for achieving nothing of any substance or importance, other than increasing their growth and shareholder value (and salary) by "increasing their efficiency" (or "efficiencies"; remember: this is an intangible concept that can somehow exist in the plural form), i.e. firing a bunch of humans and replacing them by "AI". EDIT: I kinda like the outlandish randomness that is Shrimp Jesus, though 😂 #AI #GenAI #GenerativeAI #spam #garbage #slop #deadinternet #searchengines #Google #Bing #socialmedia #Facebook #Twitter #X #LinkedIn

SearchGPT: What you need to know about OpenAI’s search engine With its characteristic cheerfulness, ChatGPT will tell you that the digital media landscape is “rapidly evolving.”  Google is regularly updating its algorithm, AI is getting smarter by the day and new tools are entering the market faster than fashion trends, which go in and out of style. That can mean opportunities and challenges for us. One of the most talked-about evolutions to the search landscape is OpenAI’s SearchGPT, a product that could rival Google’s dominance soon.  At my agency, we recently gained early access to SearchGPT. Below are our takeaways about this new tool and its implications for digital marketing. What is SearchGPT and how does it work? SearchGPT is an AI-powered search engine that combines the strengths of traditional search engines with the advanced conversational abilities of large language models. It delivers answers to user queries using real-time information from across the web.  Rather than returning a list of links for users to sift through like traditional search engines, SearchGPT provides direct answers, summaries and insights based on an understanding of context and the user’s intent. OpenAI doesn’t clearly state the exact details of how SearchGPT works, however, we can surmise that it uses something akin to retrieval augmented generation (RAG) which is a popular approach used by other AI search engines including, Perplexity and Google AI Overviews. RAG is designed to reduce the likelihood of hallucinations in responses by integrating information from a database into the LLM response to enhance accuracy. The model converts the search query into numerical embeddings that capture its meaning and searches a vector database containing trusted information sources. In this case, the web index is most likely provided by Bing based on OpenAI’s partnership with Microsoft. By retrieving the most relevant content, SearchGPT can generate precise responses while linking back to the original web content, ensuring transparency and reliability. The retrieved sources serve as additional context for the language model to accurately answer the user’s query. Key features of SearchGPT include: Conversational interface: Users can interact with SearchGPT in a more natural, dialogue-like manner. Direct answers: Instead of a list of links, SearchGPT provides concise, relevant answers to queries. Citations panel: A sidebar displays the sources used to generate the response, with links to the original content. Follow-up questions: Users can ask additional questions to explore topics further, creating a more interactive search experience. How does SearchGPT compare to Google AI Overviews?   My initial impression of SearchGPT is positive; it certainly outperforms Google’s AI Overviews (AIO).  While SearchGPT may result in fewer clicks for informational terms – and mostly for low-intent searches – in my opinion, this shift broadens the search ecosystem...

How To Rank For Google AI Overviews We are giving away the farm, no catch 👇🏽 Last year Google launched AI Overviews, their response to ChatGPT for search.  Overall the user response was rough in the beginning but since fine tuning they are seeing better results. My buddy Neil is seeing in general a ~1% increase in clicks when ranking in AI overviews - not outrageous but I'd imagine this will increase over time. At my company Taco we've been testing and learning what works and have noticed some clear patters to drive impact. Here is how to do it: 1. Create a hit list of keywords for AI Overviews Rankings Firstly, not every keyword will have an AI overview. An example of a keyword that is would be “How to pick the best health test kit for low testosterone?” Ahrefs and other tools now tell you what keywords will show up in AI Overviews (AIOs) so you don’t have to guess :) 2. Nail optimizing for the top 10 (ideally 5) spots in the SERPs Research done by seoClarity mentioned that 99.5% of results in AI Overviews are pulled from the top 10 results. This aligns with our assessment as well. Kevin Indig does mention that AI Overviews do pull from outside the top 10, but we have found that if you rank in the top 10 spots on google, your chances on ranking for AI Overviews is MUCH higher. This starts with foundational keyword research, making sure your site is performant, creating a strong content architecture, and building links (internal and external) to the page. We have tried several methods for ranking content effectively and find it’s best to stick with taking a look at the top 5 SERPs, and generating content that is stronger than what currently exists. To take this a step further try to include first party data that showcases the uniqueness of your content and expertise by who is writing it. AI-assisted content does work - I would not rely on straight AI generated content for this. Keyword is using AI to generate the first draft and then using humans to take that draft and edit it. Make sure the content is punchy and to the point. Create a schedule for refreshing and updating old content, preferably sorted by high impressions and low clicks data. 3. Optimizing your existing content accordingly Once your content is competitive, links are being built, and it has a shot of actually ranking in the top 10 spots, have your team determine what content is currently being referenced in the AI overviews. Take that content, and copy a similar version into your article, preferably with the similar tone and style. The whole point of AI Overviews is to give a preview for the searcher so they can go to the site and learn more. 4. Index and wait Once you’ve followed steps 1-3, go into google search console and re-index your post. Of course google will do this but this helps speed things up. And thats it! Should I put out our SOP for ranking for AI chat? If enough people comment “AI Chat” I will do it :)

I dislike "prompting" as a required skill to use AI. But I admit: I admire Google’s team prompt guide for creatives and strategists. It’s a *chef’s kiss* For the last 18 months, I’ve probably written thousands of prompts. I wouldn’t say I’ve enjoyed the process, but as someone building an app using LLM models, it’s something I have to do. That said, I don’t believe every modern employee needs to have the same skills to get the most out of AI. Professionals, especially in the advertising industry, are already overwhelmed with work. They need simple solutions: not tools that require lots of training to deliver value. I’ve always opposed prompt "guides" and "libraries" because when you start copy-pasting prompts, you’re just a carbon-based AI agent. But this Google guide is different. Even though I mainly use OpenAI’s API, most prompt fundamentals are universal. What Google’s team has done is create a very straightforward, value-driven, and really sane approach. It shows how Gemini can be used naturally, focusing on outcomes rather than technical skills, making AI a seamless collaborator that enhances existing workflows. So what does the guide propose? They structured Gemini’s capabilities around four key actions: 1. Condense: Break down complex information into digestible insights. For example, condensing a lengthy market research report into key trends and takeaways for a marketing team. 2. Expand: Generate new ideas or dive deeper into topics. This could mean brainstorming creative campaign concepts, like suggesting new ad formats for a dog food product based on consumer insights. 3. Iterate: Refine existing content by exploring variations. For instance, after drafting a basic social media ad, Gemini can help test different angles or messaging styles to find the most engaging option. 4. Finesse: Polish content for clarity and impact. An example would be making an elevator pitch more compelling to resonate with senior stakeholders. One thing I’d add: If you want to get the most out of this guide, combine it with our Adaily. We’ve aggregated over 3,500 award-winning campaigns that you can use as context in Gemini, enriching these approaches with real-life executions. And that’s just for now. We’re planning to add more capabilities to our Enterprise Plan, and this Google guide shows a great direction for us. Since we’re part of the Google Founders Resident program, it would be a waste not to make the most of Gemini. Check out the full guide below. If you like it, share it with your fellow creatives and strategists!

Gemini just made prompting idiot-proof. Very good guide. Over the last year or more, my prompts have become more detailed and very specific, resulting in better output. And then the refining starts.

Gemini's "Prompt Guide for Strategists and Creatives" 🤖 Check out that value-driven full guide below 👇

ICYMI: https://lnkd.in/ekW6C7-C OpenAI's now testing their own search engine, SearchGPT, where users can interact with it like how they interact with ChatGPT. What does this mean for SEO/marketers? - If OpenAI’s joining the search game, then I’d say that we’re doubling down towards a future where users interact with search engines more like they do with ChatGPT vs. Google today. (Fairly long-tail queries vs. short-and-sweet queries) - And without having seen SearchGPT - and seeing if they’re choosing to more-prominently display sources than Google’s AI overview - I would say that brand marketing and diversifying your company’s reach beyond just the SERP should be a growing consideration to the marketing strategy. (Because if these companies don't really display sources used for their results generation, some searchers may just not bother to look at the companies used to generate those results) - Also continue writing for humans and not for search engine. (Helpful, not overly jargon-y, has user proof, backed by expertise) But of course, this is all personal opinion. Not data backed. I’m personally interested in testing the second bullet. Working for a B2B software company with a global presence, there’s probably an equal, if not more benefit, that in-person meetups can have for demand generation than pure digital marketing. We’ll see! Interesting article though.

So, LinkedIn did a thing. Where to begin? Let's start with the fact that LI does not pay me for my content, so using it without my consent is ethically questionable, regardless of purpose. And yes, the image below depicts my option for removing consent - but I was directed to this option by other users' posts on LI. The platform never informed me it was silently harvesting my content for commercial use. Now, let's talk about purpose: I come to LinkedIn to read other people's content. I want to hear about their actual experiences, their actual views, their insights, ideas, aspirations, inspiration. I don't care what a machine thinks about impostor syndrome based on liquified and pulped insight aggregated from a global cohort of users. I am not interested in the mean, median or modal opinion on anything, as it is frequently underinformed and ill-considered. I absolutely care about some people's views more than others, which is why I carefully curate my follow and connection lists, whence I manage what appears in my feed. Simply put: this feature holds no value for me. Why silently opt me into contributing towards it? For that matter, why is AI data collection so often conducted with an ‘act first, ask questions later’ attitude? I suspect the answer is simple: if you asked people whether they would like their thoughts, insights and experience used to train a commercially run AI platform with no compensation to them, I wonder who would say yes? Not to mention the shady consumption of copyrighted content used to train gen AIs of all types. This being the case, why is there so little regulatory oversight of how these companies are using this dragnet approach to entire swathes of human cognitive and creative content? The same organisations that duck responsibility for political fact checking or hate speech claiming “we’re just a platform, please don’t blame us for the content on here, we're only hosting it" are now indiscriminately misappropriating the same content in order to repackage and sell it back to us. If a company is using AI to generate content on a social platform, it is surely impossible to claim amnesty from the content itself. And if an AI decides to generate content based on political propaganda or hate speech, who is responsible for the outcome? The owners and operators of the AI, or the creators of the data set? And why are we only asking these questions years after these tools entered common use? Why do they remain unanswered even as our data is siphoned from our back pockets? If LinkedIn actually cared about its users' views, they would do two things: 1. Advertise the setting pictured below, and ensure they had obtained *informed* consent from users. 2. Add a second setting beneath it enabling me to also suppress all content produced by generative AI, showing me only material created by actual humans. I won't hold my breath.

Really good practical guide to prompting - for strategists and creatives!