Your Brand Through ChatGPT's Eyes

¶ CONTEXTTHE TEST

01Open a fresh chat and ask about your own company

Do it right now. New conversation, no history, and type: what is [your brand] and should I trust it. Then read the answer the way a stranger would, because a stranger is exactly who just read it.

That paragraph is doing more selling than your homepage. It reaches people before your site does, it arrives with the authority of a machine that supposedly checked, and you did not write a word of it. The answer an AI gives about your company is your real homepage now, and most founders have never once read it.

I have run this test on all 21 of my products. Some answers made me proud. One of them taught me something I am still a little embarrassed about.

¶ WAR STORYTHE TOOL THAT
FORGOT ITS
OWN NAME

02The tool that could not get itself cited

I built a product whose entire job is to check whether ChatGPT, Perplexity and Gemini mention a brand. An AI citation checker. So one afternoon I pointed it at itself.

The models could not name it.

I dug into why, and the answer was sitting in my own HTML. The homepage title said "AI Visibility Tool." The logo in the corner said "Citation Monitor." Nowhere on the page I most wanted quoted did the actual product name appear as a claim a machine could lift. I had built a tool to get brands cited, and forgot to name my own brand on the one page that mattered.

Think about what a model has to do to recommend you. It has to be able to say a sentence like "X is a tool that does Y." If your page never puts your name next to what you do, the model has no sentence to say, so it says nothing. Absence is not neutral. Absence is a competitor getting mentioned instead.

The fix took ten minutes. The title became the product name followed by exactly what it does and which engines it checks. The wordmark got its full name back. Naming yourself plainly is not branding vanity. It is handing the machine the sentence you want it to repeat.

EXTRACTED — THE SENTENCE THIS ENTRY EXISTS FOR

If your page never puts your name next to what you do, the model has no sentence to say. So it says nothing.

¶ DATAWHERE THE
ANSWER
COMES FROM

03The answer is assembled, and you can find the parts

The reply about your brand is not a single opinion the model holds. It is stitched together from sources, and you can go find those sources.

On one of my legal sites I measured it directly. Bing and the pages I pushed through IndexNow accounted for 87 percent of the ChatGPT citations that site earned. The much-hyped llms.txt file, on that same site, produced no measured citation effect at all. On my own traffic, Bing drove 87 percent of the ChatGPT citations, and the trendy llms.txt file drove a number I could not distinguish from zero.

That is not an argument against llms.txt. I still keep mine accurate on every domain, because an accurate one costs nothing and a lying one hurts. It is an argument for measuring instead of believing. The channels that actually feed the answer about your brand are often the boring ones you already have: a Bing-indexed site, clean pages, and a presence on the third-party sources models lean on, because listicles and community posts make up a large share of what ChatGPT cites.

¶ CLAIMTHEY ARE
ALREADY
READING

04The machines are reading you tonight, whether you look or not

People assume AI answers come from some finished snapshot of the internet. They do not. The crawlers come back constantly, and they read the current version of you.

One of my sites gets fetched by ChatGPT roughly 13 times a day. It has almost no human traffic. On that site, ChatGPT is a more loyal reader than any human subscriber I have ever had. Every one of those visits is the machine refreshing what it believes about the brand. Which means the answer about you is not fixed. It is a live reflection of whatever your site said the last time a crawler stopped by.

That should feel like relief, not pressure. It means the embarrassing answer is fixable. Change the page, let the crawler return, and the reflection changes with it. The homepage that lies about you today can tell the truth by next week if you edit the source it is reading.

¶ PLAYBOOKCHANGE THE
REFLECTION

05How to fix the answer, step by step

This is the exact loop I now run on every product before I care about anything else.

[1]

Read the answer first. Ask three engines about your brand. Copy what they say. That transcript is your baseline and your to-do list in one.

[2]

Name yourself on the page that matters. Your title tag and your first heading must say your brand name next to what you do. If a model cannot form the sentence, it will not say it.

[3]

Make one claim per sentence, with numbers. "Checks ChatGPT, Perplexity and Gemini for your brand" is liftable. "Next-generation visibility solutions" is not. Models lift sentences, not slogans.

[4]

Fix your entity everywhere. Same name, same one-line description, same numbers on your site, your schema and your profiles. A contradiction reads as noise, and models skip noisy sources.

[5]

Feed the channel that actually works. Get indexed in Bing, ping IndexNow, and earn a footprint on the third-party sources models trust. Measure which one moves your citations, then do more of that one.

[6]

Re-read the answer in a week. The crawler will have come back. If the reflection changed, you have proof the loop works. If it did not, you have a sharper question.

You would never let a stranger write your homepage and publish it unread. That is exactly what is happening every time someone asks a model about you. Audit the answer the way you audit your books, because right now it is the most-read page you do not control.

Entry 003 of the ledger. Daily.