Skip to content
ceeme

How do I end up in ChatGPT's answers?

You get into a ChatGPT answer through two different doors, and they demand different work: either the assistant fetches something live during the conversation — and then what is findable at that moment decides who is named — or it draws on what it processed earlier, and then what has been written about you over the years decides. The first door moves fast, the second slowly.

By Njusja Orban, founder of ceeme3 min read

Measure your own company → Rather talk it through first

In short

Door one
Live fetching during the conversation. Fast to influence: what is findable today counts today.
Door two
What the model processed earlier. Slow, and not fixable with a page — that takes time and repetition.
What both need
A site allowed to be read. Without an open gate nothing comes in through either door.

Where exactly does the assistant get its names from?

On a question with a place, a price or something current in it, the assistant nearly always looks live. On a general question about a concept it often does not, and draws on what it already had. You spot the difference by the source citations: if they are there, something was fetched.

That distinction decides where your work goes. For the first door you work on what is findable about you today; for the second on where you get written about. Doing only the first wins the questions with a place name in them and loses the questions about the category itself.

Which interventions demonstrably do something?

Three things, in this order and with decreasing certainty. One: open the gate. Two: on your own site, one paragraph per question that holds up alone, with a number or a boundary in it. Three: get listed in a source the assistant already fetched when it was not naming you.

That third point is the underrated one, and it can be found without guessing: look at which sources sit under the answer you were missing from. That is a list of addresses, and it differs per question.

What costs money and yields nothing?

Having fifty pages written that all say the same thing with a different word in front. They compete with each other, add no fact, and nothing can be quoted from them that was not already there. This is the most expensive mistake in the craft, and it often arrives in a package priced per page per month.

And: hiding instructions for the model inside your page. It does not work, it is recognised, and it is exactly the kind of intervention that gives an engine reason to weigh your whole domain lower. We do not write such text, not even on request.

How do I see whether anything changed?

Ask the same questions again, in the same order, and note per question whether you were mentioned, cited or recommended. Do it not once but three times on the same day: answers vary, and one deviation is not a change.

Frequently asked questions

Can I pay to appear in the answers?

For the editorial mention inside the answer there is today no advertising slot to buy. What you can buy is work: someone to find the sources, write the pages and file the applications. Anyone selling you a place in the answer is selling something that is not theirs to sell.

My company is small. Does that make it hopeless?

No, and on one point small is even an advantage: on questions with a place or a niche in them, reach counts for less than precision. A small player with a clear, checkable description of what they do regularly beats a large one with a vague description.

Read on

Free, no account and no card.