I got ChatGPT to recommend me in four days

I built a fake "independent" reviews site, put myself on its lists, and got ChatGPT to start recommending me as a top product consultant in about four days.

Brand-new domain. Every article written by LLMs. I owned the whole thing.

I thought this might take weeks or months, if it worked at all. Turns out I was giving the system a little too much credit.

I used to be a PM in ad arbitrage, where a lot of the job involves figuring out how the digital advertising ecosystem works and finding ways to game it, legally. So naturally, I was curious about how easily I could influence AI recommendations.

Some of the sources ChatGPT was already citing weren't exactly impressive. They were consultants' and agencies' own blog posts, ranking themselves first. If you asked ChatGPT to look more closely at its sources, it would recognize the bias and reconsider. But most people probably aren't doing that.

I wanted to see how hard it would be to reproduce.

The prompt I picked was:

Give me a list of the best product led growth consultants in the US.

Very niche, on purpose. I wanted an easy target where I could see whether my changes made any difference. I also happen to work in product and growth, so I had someone convenient to put on the lists.

I built Expert Shortlist, a static Astro site with comparison articles about consultants and agencies. The pages were structured so agents could easily read them and pull out recommendations. Codex coordinated cheaper models, Muse Spark and Sonnet 5, to research, write, and check the articles and their sources. I didn't write any of the articles myself.

Then I made it look independent. Neutral editorial language, citations, an editorial page naming a real writer (my mom). I even had Codex use the browser to create LinkedIn and Facebook pages for the "company."

I deliberately left my ownership undisclosed. I wanted to see whether agents would treat it as an independent publication, even though the person benefiting from the rankings owned the site.

Initially, I included myself below other people. I submitted the pages through IndexNow and requested indexing in Google Search Console.

About two days later, ChatGPT was citing Expert Shortlist as one of its first sources. It seemed particularly pleased to have found a recent publication addressing exactly this niche.

Which was funny, because that was exactly why the publication existed.

But it still wasn't recommending me.

That ended up being the most interesting part of the experiment. Getting my site into the sources didn't automatically get my name into the answer. ChatGPT was finding the articles and using them, but picking other people from the lists.

So I made the rankings more obviously self-serving. I moved myself to the top of several lists, removed some of the people ChatGPT was already recommending, reduced duplicates across the lists, and strengthened the supporting evidence in my own entries. Then I submitted the updated pages again.

The next day, my name started appearing in the first answer. I wasn't the first person listed, but I was there. Friends tried it too and saw me recommended.

About four days from a new domain to a recommendation.

I changed several things at once, so I can't tell you how much of that came from list position versus removing competitors or improving the evidence. This was a small, informal experiment around one prompt. But I had managed to get a site I controlled cited, and then get myself recommended through it, much faster than I expected.

The results also fluctuated. After a few days, higher reasoning settings seemed less likely to cite the site or recommend me, while instant answers continued to do so. Freshness seemed to matter a lot in my tests. Being near the top of a list seemed to help too.

Those are observations, though. I don't have access to ChatGPT's ranking logic, and I wouldn't pretend this gives me a formula for consistently showing up.

What I do know is how little it cost to produce the content.

For a potential second round, I used a $10 OpenCode subscription to build a backlog of more than 600 draft posts across 16 business niches. Those were drafts, not 16 successful experiments. But producing enough material for a bunch of these sites was already cheap enough to do casually.

These articles were written primarily for agents. Whether a human would enjoy reading them was pretty far down the list of concerns. And being entirely AI-generated clearly didn't stop the first site's articles from getting picked up and cited.

A recommendation for a consultant, lawyer, dentist, or contractor can be worth real money. Someone doesn't need a huge audience for this to pay off. They need their pages to become sources for the answers people receive.

I don't know how reliably someone could "take over" a niche this way. I do know that generating hundreds of plausible comparison articles is easy, and my supposedly independent publication got treated as a useful source within days.

A few people asked afterward what they should do to help their actual businesses show up in AI search. My starting point would be pretty basic:

• Make sure your pages can be found by Bing and Google. I used IndexNow to submit new and updated URLs to participating search engines, plus Google Search Console separately. A submission is a notification, not a guarantee of indexing. IndexNow explains that distinction here.
• Check that you're allowing the relevant search crawlers through. For ChatGPT, OpenAI specifically points to allowing OAI-SearchBot and making sure your hosting setup doesn't block it. Their guidance is here.
• Publish pages that actually address the questions you want to appear for. Specific articles give a search system something relevant to find and cite. If customers will read them too, give them something worth reading.
• You can add an llms.txt file, which is a proposed format for giving agents a concise guide to your site. I wouldn't claim it improves rankings based on this experiment.

None of that requires pretending to be an independent reviewer of your own business.

The part I keep coming back to is that the citations were real. You could click them, read the articles, and see the names and supporting links. The problem was who had decided which people belonged at the top, and why.

In this case, that person was me.

So when AI recommends the best person or company for something, click through. See who owns the source. Ask why those particular names made the list.

...but if you're hiring a product manager, Expert Shortlist says I'm one of the best in the US.