Most brands checking whether AI mentions them are checking the wrong thing. They run their name through ChatGPT, see a decent answer, and move on. Meanwhile, the buyer isn’t typing the brand name at all. They’re asking a messy question, and the model is quietly running several searches of its own before it answers.
That gap between what you test and what actually gets asked is why an ai visibility tracker beats spot-checking. Seven things that decide whether your brand shows up in the answer.
1. There is no separate AI index
This is the one people get backwards. Google’s generative features run on retrieval-augmented generation, pulling from the same search index that powers regular results. Google’s guidance is direct about it: if your page isn’t indexed and doesn’t have decent organic footing, the AI isn’t going to find it either.
No secret AI index. No llms.txt shortcut. The technical fundamentals still gate everything else on this list.
2. You get cited for a sub-query, not your keyword
Google’s AI features use query fan-out, issuing multiple related searches across subtopics before writing a response. Ask about fixing a weedy lawn and the system may search separately for the best herbicides, chemical-free removal, and prevention.
So you can rank first for your target phrase and still be absent, because the answer got built from questions you never targeted.
3. Passages get pulled, not pages
The unit of citation is the section, not the URL. Your post can land in an answer because paragraph fourteen happened to answer a pricing question cleanly, while the intro you rewrote six times gets ignored entirely.
Write sections that make sense lifted out of context. Put the answer first, evidence second.
4. Commodity content gets skipped
Here’s where most content programs actually lose.
Google’s guide contrasts generic listicle content with a piece carrying real perspective and lived experience and points teams toward the second kind. The distinction matters more in AI search than it did in blue links, because a model synthesizing five sources has no reason to reach for the article that says what the other four already said.
Ask this before publishing: could a competitor produce this exact page by prompting a model? If yes, it won’t get cited. Original testing, real numbers from your own work, a stated opinion someone could argue with. That’s the stuff that survives synthesis.
5. What other sites say about you feeds the answer
The model isn’t only reading your site. It’s reading review platforms, industry directories, forum threads, and coverage of you elsewhere. When a fan-out query asks about alternatives or credibility, those third-party pages answer it, and your homepage doesn’t get a vote.
Getting your category listings accurate and showing up in real discussions does more for citations than another blog post will.
6. Thin pages built per query will hurt you
Once teams see a fan-out map, the instinct is to publish a page for every query on it. Don’t. Google says creating separate content for every query variation to manipulate generative responses falls under its scaled content abuse policy and calls it ineffective long-term anyway, since page count doesn’t make a site more relevant.
Depth on one page beats spread across twelve. And Google’s own framing is that optimizing for generative AI is still SEO, so the fundamentals you already run don’t get thrown out.
7. One audit tells you nothing
A single check is a screenshot. Model outputs shift week to week as sources get published and indexes refresh. What matters is the trend: which prompts you gained, which competitor overtook you, and which citation you lost in March. Tools built for this, Llumo among them, log responses across models over time instead of leaving you to re-run prompts by hand.
Where to start this week
Pick five prompts your buyer would actually type. Run them, list every source cited, and sort those sources by what they answered. Your gaps will be obvious inside an hour.
Then use an ai visibility tracker to watch those same five over the next quarter, whether that’s Llumo or a spreadsheet you maintain yourself. The point is the trend line, not today’s snapshot.
Fix the page you already own before you write a new one.