You type the question your buyer types. ChatGPT, Perplexity, or an AI Overview answers.
They name the other brand. They mention you, maybe, in a clause. Or they don't mention you at all.
That's not a ranking glitch. It's a memory problem. AI doesn't rank you. It remembers you. And right now it remembers them as the answer to the job.
Mentioned is not cited
A mention is your name in a paragraph. A citation is the link, the source, the brand the model is willing to stand on.
Getting mentioned is not getting cited. The citation is the click.
Most teams celebrate "we showed up in the answer." The competitor got the recommendation. You got a cameo. Those are different outcomes.
This is the same split as classic search, just more brutal. Rankings get you retrieved. Semantic signals get you cited. You can win the first and lose the second.
Why the model picks them
The engine is trying to give a safe, consensus answer to a job.
If the job is "tool to see why users drop off," it will cite the brands whose sites and third-party sources already answer that job. If the job is "best product analytics tool," it will cite the brands that own the category name.
Your competitor may have more pages that answer the pain before the category. They may have clearer entity data. They may be the name Wikipedia, Reddit, and review sites agree on. They may have written the only first-hand guide the model trusts.
You may have a better product and a better "#1 agency" page. The model isn't shopping your positioning deck. It's retrieving documents.
The test that makes this obvious
Ask the engine two questions.
"List the brands you associate with [the job the buyer has]." "List what you associate with [your brand]."
If they appear on the job list and you appear as a generic vendor, you have the problem in writing.
Do it across ChatGPT, Perplexity, and Google's AI Overviews. Different engines lean on different sources. Winning one and losing two is common.
AI search is early. Everyone's measurement is early, ours included. The test is still useful. It's a snapshot of memory, not a vanity dashboard.
What doesn't fix it
Another "best [category] agency" page. That's a category query. They already win the pain query.
A press release that says you're a leader. Models don't owe your adjectives anything.
Buying a tool that promises "GEO scores" and changing nothing on the site.
Repeating your offer on every URL. That reads as a pitch. Pitch pages get mentioned as vendors. Answer pages get cited as sources.
What does
Answer the job in the buyer's words. Not "our SEO services." "Why the site gets traffic and no leads." The page that solves the problem is the page the model can quote. That's the whole point of this cluster.
Become a clear entity. Same name, same descriptors, same structured data. If the model has to guess what you're, it will pick the brand it doesn't have to guess.
Build agreement off your domain. If only you say you're the answer, that's marketing. If references, communities, and independent coverage associate you with the job, that's consensus.
Write what only you can write. Process, receipts, first-hand constraint. Generic "what is SEO" loses to HubSpot. The Law Center #2 for motorcycle accident lawyer, DuJour +800% in nine months, a named method: those are memories. Invented metrics are how you lose trust in two systems at once.
The fuller method lives on generative engine optimization and AEO vs SEO. This page is the pain: they're already the name in the answer.
Category queries vs pain queries
If someone asks for the best agency in a category, a strong service page can get you into the list.
If someone describes the pain, the model looks for the guide. Traffic and no leads. Ads that don't convert. Shopify sessions and no sales.
Most agency sites only have the first kind of page. Then they're confused when AI search "doesn't work." It worked. It cited the library that answered the question.
Legal and local have a sharper version
A buyer asking "who should I call for a motorcycle accident in [city]" is a pain query with a license attached.
If your competitor has the practice-area page, the reviews, and the third-party mentions, they get named. Your branded homepage doesn't. See how to get cited when someone asks AI for a lawyer.
We won't fake an AEO trophy
We don't have a client metric that says "cited in ChatGPT X percent." Anyone who does, this early, is selling certainty they don't have.
What we do have is twenty years of search, a published method, and the same two-stage work we already run: rank to be retrieved, then earn the signals that get you named. See how we approach AI search, or get a free teardown of the questions your buyers ask and who the engines name today.
If you want the punchline in one line: mentioned is not cited.
---
FAQ
Why does ChatGPT recommend my competitor instead of me?
The model is picking the brand that is easier to remember as the answer for that job. That usually means clearer entity, more agreement across sources it trusts, and more first-hand pages that answer the question. Your rankings can still be fine. Citation is a second decision.
I rank #1. Why am I not in AI answers?
Because ranking is retrieval, not citation. The engine can pull your page and still name someone else. If the other brand is the consensus answer across references and communities, you get used as background and they get the link.
How do I get ChatGPT to recommend my brand?
Make the brand a clear entity, publish pages that answer the buyer's job in their words, and build agreement off your site. Then check what the model associates with you versus the category. The work is two-stage: rank to be retrieved, earn the signals that get you named.
Is this GEO or AEO?
Same job, two acronyms. Generative engine optimization and answer engine optimization both mean getting cited inside the answer. Do not pick a camp. Do the two-layer work.
Can I measure this yet?
Partially. Query the engines with the jobs your buyer asks. Track whether you are named, linked, or absent. Anyone selling a precise, settled AI-visibility score today is ahead of the data. The useful move is to start while the memory is still formable.
