Most GEO advice is noise. Here's what actually gets you cited by AI.
The “AI SEO” gold rush has produced a lot of confident, wrong advice. A grounded look at what the controlled research really found — keyword tricks fail, sourced evidence wins — and how we think about it at LoopCite.
There’s a new gold rush, and with it a new wave of snake oil. People — and increasingly their AI assistants — ask ChatGPT, Gemini or Google’s AI a question and act on the handful of businesses those engines name. Naturally, an entire cottage industry has sprung up to tell you how to “win” at it. GEO. AEO. AI SEO. Pick your acronym. Most of the advice is noise, and some of it will actively hurt you — because a lot of it is old SEO instinct wearing a new hat, and the new medium doesn’t reward the same reflexes.
The advice to ignore
“Stuff your pages with AI keywords / raise your semantic density.” This is the oldest reflex, repackaged — and we’re not guessing about it. In the first controlled academic study of what actually changes AI citations, out of Princeton and IIT Delhi, keyword stuffing was the one tested tactic that returned no benefit at all, and slightly degraded visibility on some engines.1 Writing “for the embedding” is folklore.
“Just add an llms.txt file and you’re AI-ready.” A tidy file is a nice-to-have. It is not a citation lever, and Google is blunt that it ignores llms.txt and similar AI-only markup for its features.2 We ship one anyway, because it’s cheap and some non-Google assistants read it — but if a vendor sells you a file as the silver bullet, that’s the tell.
“Rank #1 on Google and the AI will cite you.” This one sounds right, which is what makes it dangerous. Answer engines retrieve and stitch together passages, weighing things the blue-links ranking never did. Ranking helps you get considered. It does not guarantee you get quoted.
“Fabricate authority — spin up stats, seed reviews, publish ‘best of’ lists that rank you #1, blast cheap press releases.” Beyond the ethics, this is a trap. Invented statistics fail the moment an engine cross-checks them, and gaming tactics are exactly the fragile signals that get discounted. You’re building on sand.
What actually matters
Strip away the hype and the picture is calmer, and more demanding.
1. Readiness is not the same as visibility. Two different questions. Can an AI read, understand and trust your site at all? And does AI actually name and cite you when a customer asks? A site can be beautifully ready and still invisible. Most tools blur these two. They are different problems with different fixes.
2. AI lifts what it can quote and attribute. The same controlled research that debunked the keyword tricks found the real movers were genuine content signals: real statistics with a named source, short authoritative quotations, and explicit citations — worth up to a 30–40% relative lift in visibility, and more for sites that don’t already rank near the top.1 Not more words. Better-sourced ones.
3. A clear, cross-referenced identity beats clever copy. AI grounds its confidence on who you demonstrably are — a coherent entity, consistent across the places it already trusts. Thin, contradictory or JavaScript-only identity signals give an engine nothing to stand on.
4. Honesty is a strategy, not a constraint. Every shortcut that fabricates — a metric, a review, an authority — is a liability the moment an engine or a customer checks. The durable play is to be genuinely citable.
None of this is secret. It’s just harder than a checklist, which is exactly why the checklists sell better.
If an AI-visibility playbook is mostly clever tricks to fool a machine, be suspicious. The engines are getting harder to fool, not easier.
How we think about it at LoopCite
We built LoopCite because the market was full of tools that grade your meta tags and hand you a number you can’t check. We do it differently on a few stubborn principles:
We measure what AI actually says about you — across ChatGPT, Gemini, Copilot and Google’s AI Overviews — not a proxy for it. If we tell you you’re invisible for a query, it’s because we watched it happen, and we show you the answer verbatim with the sources each engine cited.
We test each engine on its own terms. They see you differently, and blind-averaging them hides the truth. We separate readiness from visibility, so you know whether the problem is “AI can’t read me” or “AI reads me and picks someone else” — because the fix is completely different. You can see that split for yourself in a head-to-head against a competitor, free.
We’re built for this market — South African businesses, and the languages South Africans actually search in, not an English-only view of the world. And we refuse the tricks. We will not sell you keyword-stuffing, fabricated stats or seeded reviews. We’d rather tell you the unglamorous truth: get genuinely good at being citable, then check the receipts.
We call the model behind it VERA — our Visibility & Engine-Readiness Assessment. What’s under the hood is ours. The philosophy isn’t a secret at all: measure reality, refuse shortcuts, and coach toward what the evidence actually rewards.
The bottom line
The businesses that win the answer over the next few years won’t be the ones who gamed it. They’ll be the ones who were genuinely worth citing — and who bothered to measure whether it was working. That’s the boring, durable truth. It’s also the one we’re building on.