The most common mistake we see is treating AI invisibility as one problem. It isn't. A brand that ChatGPT has never heard of and a brand that ChatGPT knows but never cites are in different situations, and the work to fix each is different. Before you touch your content, figure out which failure mode you're in.
Failure mode 1: the model never learned you exist
Ask ChatGPT directly, with browsing off if you can: what is your brand, what does it do, who is it for. If the model hallucinates, confuses you with someone else, or says it doesn't know, you have a training-data problem. Your brand didn't appear often enough, or consistently enough, in the public web the model was trained on. No amount of on-page optimization fixes this quickly. The fix is presence: consistent descriptions of your brand across your own site, directories, press, and third-party mentions, so the next training run and the retrieval layer both find a coherent story.
Failure mode 2: retrieval can't find a citable page
When ChatGPT browses, it retrieves pages and then decides which ones to cite. The Ahrefs study of 1.4 million prompts found ChatGPT cites roughly half of the URLs it retrieves - retrieval is necessary but not sufficient. If your pages get fetched but never cited, the page itself is the problem: no direct answer near the top, no extractable claim, or content that restates the question without resolving it. The same study found search-sourced content is cited 88.5% of the time, versus 1.9% for Reddit - the retrieval layer heavily favors pages that already rank in traditional search. If you're invisible in search, you're mostly invisible to retrieval too.
Failure mode 3: your entity is ambiguous
If your brand shares a name with something else, or your own materials describe you three different ways, the model hedges. It might mention you with caveats, attribute a competitor's product to you, or skip you to avoid being wrong. The tell: ChatGPT knows some facts about you but mixes them with facts about someone else. The fix is entity consistency - same name, same one-line description, same category, everywhere it matters.
Failure mode 4: you lose the source competition
Sometimes everything works and you still don't appear, because the pages the model retrieves for your target prompts belong to competitors, comparison sites, and industry publications that don't mention you. This is the hardest mode to see without tooling, because nothing is broken - you're just outcompeted in the citation layer. The fix is earning presence in the third-party pages that actually get cited for your prompts, not just publishing more on your own domain.
A diagnosis sequence you can run this week
- Ask the model about your brand with browsing off. Wrong or empty answer: failure mode 1.
- Ask commercial prompts in your category with browsing on. Note which sources get cited. If your pages appear in retrieval but not citations: failure mode 2.
- Check whether facts about you get mixed with another entity: failure mode 3.
- If the cited sources are all third-party pages that don't mention you: failure mode 4.
One honest caveat: manual spot-checks are noisy. The same prompt asked twice can produce different answers and different citations, so a single session tells you less than it feels like it does. Run each prompt several times before you conclude anything. This variance is exactly why we built ClerAEO to sample repeatedly rather than trust one answer - but even by hand, repetition beats a single screenshot. Diagnose first, then fix, then measure whether the fix moved anything. Skipping the first step is how AEO budgets get burned.