I run a GEO agency, so I went looking for proof that my own industry flattens people's writing. What I found: the thing everyone sacrifices their voice for mostly doesn't work, and the thing that does work looks far more human than the playbooks admit. There is a complaint I hear every week, and I think it is basically fair. It goes like this. You start writing for AI search. You add the comparison table. You break the argument into H2s that begin with the query. You strip out the sentence you actually liked because it was "too conversational." You add three statistics you had to go find, and a quotation from someone more famous than you. At the end you have a page that ranks, or gets cited, or at least ticks the boxes in some tool, and that nobody, including you, would ever read for pleasure. The complaint has a stronger version. It says this is a structural problem rather than a matter of personal taste. That optimising for machine readers systematically strips out exactly the things that make writing worth reading: the digression, the argument you're not fully sure about, the anger, the joke, the specific memory. That we are all being sanded down into the same tone by the same tools. I sell generative engine optimisation for a living, so I had an obvious incentive to find out whether the strong version holds. I read the peer-reviewed work, the preprints, the platform documentation and about a dozen vendor studies. Below is what I found, argued honestly in both directions, with every number linked so you can check me. The short version: the fear has misidentified its target. There is a real mechanical penalty on human writing, and it falls on self-containment rather than on emotion. And the specific sacrifice most people are making, turning prose into tables and bullets, has close to no measured effect on whether AI cites you at all. First, what the machine is actually reading Almost every argument about "writing for AI" gets confused because people picture a model reading their page. That is not what happens.

How to make ChatGPT recommend your APP (not website!)? AppTweak research (https://www.apptweak.com/en/aso-blog/ai-app-discovery-llm-search) shows that App Store rankings and ChatGPT recommendations are two completely different lists. Some top-ranked apps don't appear in AI answers at all. Because App Store ranks on downloads and keywords. AI recommends based on how well your app is represented across the open web. Users now ask ChatGPT: "best budgeting app for freelancers" or "offline meditation app no subscription" — get a shortlist of 3–5 apps, then open the store to convert. The store is still where installs happen. It's no longer where discovery starts. What actually moves the needle in AI recommendations: → Third-party mentions (Product Hunt, Reddit threads, niche media) — LLMs trust earned media over your own site → Reviews on G2, Capterra, Trustpilot — AI pulls heavily from these → Web content structured around user intent, not keywords — pages that answer a specific question in the first 50 words get cited more → Fresh content — a 2025 guide beats a 2022 guide every time AppStore handles conversion. GEO handles consideration — before the user ever opens the store. Ignore GEO now, and you're handing the top of the funnel to competitors who aren't. Are you already thinking about AI visibility for your app — or still focused purely on how to make your AppStore more converting? BTW, I'm Alena — analyst, ex-University of Cambridge associate, 12 years researching markets for businesses. Now I'm building hi.visibilitypro.ai — a GEO-focused product for brands that want to show up where AI answers live. I write about GEO, market shifts, and what it's like to found something in a space that's still being defined. Follow along if that's your thing.

I’ve spent the last few months doing something very exciting (no): talking to competitors, studying GEO tools, dashboards, pitches, and business models. Not as a spectator — as someone actually selling this stuff. If you simplify hard enough, there are two types of GEO players. 1) Former SEO agencies They already have clients and trust. So they upsell GEO as a new service. Clients cautiously agree to try. It works — mostly because the relationship already exists. 2) GEO-first startups They build dashboards, invent metrics, explain how AI works, give smart recommendations. Some are even raising venture rounds. But the problem is they don’t have many clients ready to pay real money for experiments. What I keep hearing from companies is very simple: “Do everything for us” — nice, but expensive “Here are dashboards and metrics, do it yourself” — cheaper, but confusing and easy to drop And this is where Visibility Pro comes in. We measure AI visibility (an actual score to make it all tangible). We give specific recommendations (real detailed fixes). And — the most important part — we can implement them ourselves. Not only GEO advice. Not only dashboards. We make sure your brand actually starts appearing in AI search: ChatGPT, Perplexity, Claude, Gemini. Two models in one! That’s the point. So here’s the deal: I’m happy to do a free AI visibility analysis of your website — to clearly outline what exactly you need to do to start showing up in AI search. 👉 Fill in the short form in the comments I’ll do the work for free. I get real use cases to test the product. You get clear instructions on how to start working with GEO. Fair exchange.

Over the past year, companies began noticing a pattern: customers check information in AI assistants first — not on Google, not on websites. This changes how brands are discovered. If an assistant cannot explain your product, your positioning collapses at the very first touchpoint. This short breakdown explains how generative engines retrieve and rank information (2 min): https://www.youtube.com/watch?v=2Jya6Wru8P4 The main point: AI systems don’t “rank” pages — they surface brands based on clarity, structured signals, and consistency across sources. Data - The broader geo-marketing / location-intelligence market is projected to hit $21.4B in 2025 and grow to $147B by 2035. - 84% of marketers already use some form of AI-enhanced targeting and geolocation. - Technical analysis shows that AI assistants increasingly rely on structured data and external references rather than website content alone. This is already happening across SaaS, DTC, and services. One useful action to start: Ask ChatGPT, Gemini, and Perplexity the same simple question: - “Which companies help with [your category]?” - If your brand doesn’t appear — the assistant isn’t “confident” in you. - That’s the first signal to review your AI-visibility footprint.

I fell deep into the AI 2027 report by Daniel Kokotajlo’s research team: https://x.com/DKokotajlo Every section hits — China, computational resource races, even their version of “Skynet” (“OpenBrain Project”). But one concept pulled me in hard: misaligned AI. Here’s the clean definition: Misaligned AI = the system’s actions diverge from human or societal goals. It optimizes what it thinks is correct — even if that’s not what we actually wanted. Rules followed, outcome wrong. Why it matters: 1️⃣ Metric ≠ real goal. Tell a model to “minimize errors,” and it may pick a strategy that’s harmful or unethical but improves the metric. 2️⃣ Autonomy amplifies mistakes. If the system touches many processes at once, a wrong objective scales instantly. 3️⃣ No intent, strong optimization. It doesn’t need bad motives. Just optimizing the wrong thing is enough. Examples: • AI moderation blocks legal content “for safety.” • Recommenders push harmful material because it boosts retention. • Ad optimizers waste budget on irrelevant audiences because CTR looks higher there. If this interests you — or scares you — go read the report. It’s worth every minute. 👉 https://ai-2027.com/