My Silicon Valley Girl podcast has 1.56M subscribers on YouTube. I opened ChatGPT and asked which podcasts to follow about AI and startups. We weren't in the answer.

Think about how that question used to work. We googled "best AI podcasts," got ten links, and clicked around. Now we ask ChatGPT instead, and it answers with a short list of names. No links. If your name isn't on it, you don't exist for that person.

SEO has been around for decades and it's about ranking links in Google. GEO isn't three years old yet, the rules are still forming, and it decides whether AI says your name inside the answer at all.

So how does it choose? It names the shows, brands, and products that get written about most in other people's articles: the "15 best AI podcasts" lists, the directories, the Reddit threads, the write-ups in Forbes and Business Insider. It ranks purely on how often other people mention you. Quality barely enters into it.

The first thing I needed was to see where we actually stood, so I started with a tracker built for this: Peec.

The picture was worse than I expected. Across the 1,008 sources AI pulls from to answer that question for our category, we appeared in zero. Out of 49 category questions I tracked, we showed up in 5, every one tied to a specific guest's name. The same names won every time, none of them us.

We set a plan to fix it. Results were not fast. Ten weeks in, the numbers finally started to move.

Visibility shows how often AI mentions us in its answers, and that went from 2.2% to 7.8%. But the number we were proud of was the other one, retrieved. It maps directly onto the work, because it only counts when a model opens our own site and uses what it finds. The site we built specifically for GEO had started doing its job, going from zero to 2.6%. How we built it is in week 2, below. These are still small numbers, but the direction is right.

Here's the plan, in the order that worked. Each week: what to do, why it matters, and how you'll know it worked. Anything technical sits in a PRO TIP marked for your developer. You'll see results without touching those.

You can't fix what you haven't measured, and almost everyone measures the wrong thing.

Set up a tracker. I use Peec. A tool like this asks your questions to ChatGPT, Perplexity, Gemini and the rest every day, and records whether your name came up, how high, and how you were described. Add your name and any variations people use for you.

Then write 40 to 50 questions your ideal customer would actually type, and here's the part that matters: none of them can contain your name. "Best AI podcasts for founders," not "is Silicon Valley Girl good." A question with your name in it only confirms what someone already knew to look for. The broad question is the real test.

Run them and look. You'll probably find what I found: you appear when someone types your name, and vanish on every broad question where they don't.

Then look at which websites AI keeps quoting when it answers. Your tracker shows you this. For my category, 459 of the top 1,008 were "best of" list articles. That told me exactly where I needed to be.

AI doesn't watch your videos and it doesn't listen to your audio. It reads text. The software that does the reading is called a crawler. GPTBot for ChatGPT, PerplexityBot for Perplexity, ClaudeBot for Claude. They visit a page, read whatever text is sitting there, and move on. If your work only exists as a video, they find nothing to quote, and AI won't recommend what it can't quote.

Our podcast lives on YouTube. That's the main platform and it isn't changing. So we built a text layer on top of it: a page for every single episode, over at marinamogilko.co. Same content, in a form a crawler can read. Each page has the full transcript, a description that names the guest and their real title, the timestamps with labels, and a short bio.

If you already have a website, your job this week is to make sure your best content is written out on it, not locked inside a video or a player. If you don't have one, this is the week you build something small. One page per episode, per product, per project. That's the whole requirement.

There's a quieter version of this that catches people who do have a website. If your words only appear after the page loads and runs its code, or they're tucked inside an embedded player, the crawler still arrives to a blank page and leaves. The text has to sit in the page itself. Getting our episode text to simply be there, in the page, did more for our numbers than any other single fix we made.

You can check your own in ten seconds. Right-click the page, choose "View page source," and search for a full sentence of your text. If it isn't in there, AI can't see it either.

Here's something that isn't obvious. When you ask Perplexity "best AI podcasts," it doesn't search those three words. It quietly writes its own search queries first, things like "AI podcast CEO interviews 2026," and searches those instead. Those invented queries are called fanout, and your tracker shows you the exact ones the models generate.

So read them, then put those same words in your titles and descriptions.

We already had a clue about which words those would be. Remember the five questions we did win: every one of them had a guest's name in it. Nothing broad, only names. So we pushed names everywhere, and the job title with them: "LinkedIn CEO," "Head of Gemini." A clever, abstract title gives AI nothing to match against.

One warning that comes with automating this. Ours runs from Trint to Notion to Make.com to Claude and back, and it still gets names wrong. Guests, companies, product names. A misspelled name is a name AI can't match against, so someone reads every transcript before it goes live and fixes them. Skip that step and you quietly undo the work you just did on the titles.

Then publish more pages, and fix the ones you already have. Getting cited is partly a numbers game: the more pages you have using the right words, the more chances AI has to find you. Pick your best existing content and rewrite the titles and descriptions this week, a batch at a time, rather than waiting to do it perfectly.

The fixes so far were one-time wins. This is the highest-return work you'll keep doing, and the one nobody wants to, because it isn't technical. It's asking.

Take that list of websites from week 1, the ones AI keeps quoting. For me those were podcast directories and "best of" roundups. One page alone got pulled into answers 70 times in a single analysis. Being on one of them beats anything you publish yourself, because AI trusts other people describing you more than you describing you.

So reach out, and reach out fast, within a day or two of publishing something, while it's still news. Lead with a fresh hook: one specific number or quote from your content they can't get anywhere else, and if a recognizable name is involved, put it first. "New interview with [well-known CEO], first data on X" opens doors that "please add my podcast" never will. Go for the outlets that get reprinted first, the big business ones, because one pickup there turns into ten.

Reddit, by hand. Find the threads AI quotes in your category, answer questions in them for real, get mentioned. A person writes every one of these. Automate it and you'll get the account banned.

Wikipedia, only if you qualify. Models trust it heavily, but you can't just write your own page. You need independent press already written about you, then you submit a draft through Articles for Creation and editors review it. Have the coverage? Do it. Don't? Skip it and come back later.

Then re-run your week 1 questions and compare. Report only the ones without your name in them. That's the honest number.

Four things: a tracker (Peec or similar), pages with your content written out as text, a Wikidata entry, and a way to publish regularly. My full stack is bigger. You don't need it.

It's slow. You're waiting on other people's websites and on crawlers to come back around. Our numbers moved at ten weeks, not at day 30. The work fits in a month; the payoff doesn't.

Directories are easy to get into and barely move anything, because they list everyone. What moves the needle is a real article where a human chose to include you. Those are harder, and they're the point.

So start now. This is early, and the people who show up while it's this quiet are the ones AI will be recommending a year from now.

Reply and tell me if this plan is clear, or ask me anything. I read every reply, and I'm collecting them for what I'm building next.

Marina 💜