i love looking at our search metrics, but i swear if i hear one more person brag about "AI citations" without understanding how they actually work, I'm going to lose my mind. we are completely missing the mark here.

we've all been stressing over why our blogs aren't getting cited in AI responses, trying desperately to optimize our visibility. But let's be real for a second. just because a LLM pulls your blog post as a source doesn't mean it's actually going to recommend your tool next time someone asks. Frankly, it might use your comparison article to recommend your biggest competitor instead. ouch, right?

Obsessing over getting your own URLs cited is a vanity game. It's a nice side-metric, but it doesn't move the needle. the citations that actually matter are the third-party ones-the forum threads, video platforms, and social media posts where real humans are talking about your brand. that's what signals to a LLM that your worth recommending. if other creators and users are hyping you up, the algorithm finally feels safe to do so too.

i had a colleague mention that one of their clients had outdated pricing showing up in AI search results because of an old page they deleted years ago. They published multiple new pages to fix it, but nothing changed. Why? Because the model wasn't even crawling their site for that query. it was pulling from old, cached third-party data.

over the past six months, I've been digging deep into AI visibility tracking, and some of these patterns are wild.

first off, did you know that different prompts rely on completely different sources? i tracked about a dozen prompts in the travel space across the major LLMs. In general, community forums dominated the citations. But for nearly half of those prompts, the models didn't cite forums at all. You have to analyze this at the prompt level, not just average everything out.

Also, every LLM has its own favorite neighborhoods. some models love video platforms, while others skew heavily toward social spaces. if you aren't tracking them separately, you're flying blind.

To make things more complicated, LLMs frequently don't even look at external sources. For classic "best tool for X" queries, they often just spit out answers based on their core training data. Here's a quick hack to see what you're dealing with: ask the LLM the same question twice-once with web search enabled, and once with it off. If the outdated or incorrect info only shows up in the no-search answer, it's baked into the weights. you can't fix that overnight, you have to seed the correct info across enough third-party sites so the next model training cycle catches it.

and honestly? Massive community forums, popular video hubs, and social networks are getting cited way more than prestigious media publications. if you're still chasing high-brow PR placements thinking it'll win you the AI search war, you're wasting budget.

lastly, don't trust what you see on your personal desktop. LLMs personalize search results and tailor them heavily to your geographic location. If your target market is in another country, you need to use a VPN or a specialized visibility tracker to see what your prospects are actually seeing.

is organic traffic dead? Not necessarily, but it's completely reshaped. the AI conversation has basically become the new top-of-funnel, and the actual click has just moved to step two.

So, where should we actually spend our energy? it's not about writing endless informational blog posts that get scraped for free. It's about building a footprint on third-party channels, and maybe creating high-intent, owned comparison listicles that LLMs are forced to pull from when users ask for alternatives.

Are you guys seeing similar patterns in your own tracking? or are we still pretending our old playbooks are going to save us?