Guide

How we get products named by ChatGPT, in one afternoon with Claude

Team AfterLaunch9 min read
In this article

When an engine names a company in an answer, it links to that company at most about 26% of the time. A mention with no link sends nobody to your site.

Since July we have read over 18,000 answers from ChatGPT, Gemini, Perplexity and Google AI Overviews to work out why some products get recommended and others never do.

This is the afternoon we now run on every product we study. It ends with three files on your machine: questions.md, ai-log.md and targets.md. Everything after that is email.

Save it for an afternoon you can block out. It is written to be run, not read.

Two ways to run it

Yourself: bookmark this, and when you sit down, paste this to Claude or whichever agent you use.

Read https://afterlaunch.io/guides/how-to-get-named-by-chatgpt-in-one-afternoon-with-claude and treat it as the method. Start with the 2:00 block on my homepage and pricing page: [PASTE URLS]. Stop at the end of each block and show me the file before you go on.

Or with AfterLaunch: it asks the engines your questions, keeps the log and reads the cited pages, which is the hand-done half of this afternoon. Free for 14 days, no card. Either way you leave with the method, so nothing is lost by reading on.

What separated the products the engines recommend

We took 8 categories and 32 products. In each category we picked a clear winner by GitHub stars, funding or revenue before we looked at a single AI answer. Then we read the list pages the engines cited when they answered questions in that category.

Winners were named and linked on 15.9% of those pages. Everyone else was on 4.6%. It held in all 8 categories.

58% of the sites naming and linking a winner were other software companies, often direct rivals.

Things we believed after our first study that did not survive the second:

  • Leading with "the alternative to X". 0 of 8 winners did it.
  • A big launch spike on Hacker News or GitHub Trending. 1 of 8 winners had one.
  • Product Hunt rank, blog volume and llms.txt did not separate winners from the rest.
  • "ChatGPT rewrites your question into short searches, so target those." We looked at the searches it actually ran. It searched the question word for word, 25 times out of 25.

What this does not prove: 8 categories and 32 products are a pattern, not an experiment. The 18,625 answers were collected between 19 July and 2 September 2026. ChatGPT is measured through the API with web search, not the app. Most of the data is software and developer tools. Answers move, which is why the last section exists.

So the work that most likely moves the result is getting onto the lists the engines already trust. The afternoon below is how we find them.

2:00. Twenty-five questions, none of them about you

Someone who has not heard of you asks about their problem, so that is what your questions should sound like. "What's the best uptime monitor for a small team?" is a question you can win. "Is Uptime Kuma any good?" belongs to Uptime Kuma.

We write 25 per category and keep 7 as the key ones. Paste your homepage and pricing page into Claude with this:

Here are my homepage and pricing page: [PASTE]. Write 25 questions someone would type into ChatGPT before they have heard of us. Never use our name or any competitor's name. Mix the shapes: "best X for Y", "how do I Z", "cheapest way to", "X or Y for a team of 5". Write them the way a person types, not the way a marketer writes. Then mark the 7 closest to the moment someone picks a tool. Save all 25 to questions.md with the 7 at the top under "Key questions", and stop.

2:30. Fifty-six answers, in a log

Claude cannot see what ChatGPT, Gemini, Perplexity or Google answer today, so this is the one part you do by hand. Ask your 7 key questions on each of the four engines, twice, in a fresh chat each time. That is 56 answers. Copy each answer and every link it shows, and label the paste with the engine and the question.

Why all four: they do not read the same internet. Across 18,625 cited answers in our data:

  • Reddit shows up in 46% of Perplexity's answers, and 0% of ChatGPT's
  • YouTube shows up in 59% of Google AI Overviews' answers, and 4% of Perplexity's
  • LinkedIn shows up in 22% of Perplexity's answers, and 0.1% of Gemini's
How often each platform shows up in an engine's answers, across 18,625 cited answers collected between 19 July and 2 September 2026.

Why twice: one answer from one engine tells you little. The same product can be named in 53% of answers on one engine and 100% on another.

I will paste 56 AI answers one at a time: 7 questions, 4 engines (ChatGPT, Gemini, Perplexity, Google AI Overviews), 2 asks each. Each paste is labelled with the engine and the question. For each paste add one row to ai-log.md with these columns: date | engine | question | named (yes/no) | linked (yes/no) | products named | urls cited. If a paste is missing, leave its row blank and tell me which one. When I say "done", give me our named rate and our linked rate per engine, and the 20 URLs cited most often across every answer.

Count named and linked as two different things. The engines cite sources on 95 to 100% of answers, and still link the company they name only about a quarter of the time.

This is the block AfterLaunch does for you, every week, on the free 14 days as much as after. The rest of the afternoon runs the same either way. Connected, the 56 pastes become this:

AfterLaunch is connected. Pull this week's answers for the 7 key questions in questions.md into ai-log.md, one row per answer with named, linked, products named and urls cited, and give me the named and linked rates per engine and the 20 most-cited URLs.

3:30. The pages the engines read

Take the 20 most-cited URLs from the log. These are the pages that decide the answer, and most of them are lists.

Here are the 20 URLs cited most often on my questions: [PASTE]. Fetch each one. If a fetch fails, say so and do not describe the page from memory. For every page tell me: the kind of page (list, review, comparison, forum thread, docs); every product it names, in order; which of those it links; who wrote it (a publication, a software company, or a person); and whether we are on it. Then save to targets.md every page that names 2 or more of [COMPETITOR A, B, C] and not us, with columns: url | who wrote it | named us | linked us | the gap, in their words.

targets.md is the output of the afternoon. Everything before it was to get here.

4:30. One email per page

Write to whoever owns each page, and make each email about that page. Lead with your name and the fact that you built the product. Name their list and the gap you fill, in their own words. Give them a login so they can check for themselves. Ask for the link, not only the mention.

Do not skip the pages written by other software companies. They were more than half of the sites that named and linked the winners, and a company that already lists three tools like yours has decided the category deserves a list.

For each page in targets.md draft a short email to its author. First line: who I am and what I built. Second: their list, by name, and the gap we fill, in their words. Third: a login so they can check. Last: add us, with a link, if it fits. Under 90 words. No flattery. One email per page, even when two pages share an author, because each email is about its page. Save them to outreach.md. Send nothing.

5:00. Where to publish next, engine by engine

The same log tells you where each engine looks, which is where your next piece should live.

  • Google AI Overviews: YouTube is the single largest platform it cites in our data
  • Perplexity: Reddit, then LinkedIn
  • ChatGPT: publications and press release wires; its most cited single domain in our data is TechRadar

People read LinkedIn, and both Perplexity and Google cite it, so a post there reaches both. X is almost never cited by these engines: 21 citations out of 23,371 in our data.

Here is ai-log.md. For each engine, tell me which kind of source it cited most on my questions, and give me one thing I could publish there this week that would deserve a citation. Use only the sources in the log.

Then, once a week

The lists change. One that did not have you last month adds you, or drops you. Pick a day and rerun the afternoon's second and third blocks with this:

Here are this week's 56 answers for the questions in questions.md. Add them to ai-log.md, compare named and linked rates per engine with last week, and report only what moved. Reread every page in targets.md and tell me if we are on it now, and linked. Add any new cited page that names 2 or more competitors and not us. Draft follow-ups for any email in outreach.md with no reply after 7 days. Send nothing.

Connected, the weekly block is one line:

AfterLaunch is connected. Add this week's answers for questions.md to ai-log.md, report what moved per engine, recheck every page in targets.md for whether we are named and linked now, and add any new cited page that names 2 or more competitors and not us.

If you connect AfterLaunch

The hand-done part of the afternoon is the 56 answers and the weekly rerun. That is what AfterLaunch does: it asks the engines your questions on a schedule, keeps the log, and reads the cited pages, so the 2:30 block and the weekly block become one prompt each. We build it, so weigh that. Free for 14 days, no card.

claude mcp add afterlaunch --transport http https://afterlaunch.io/api/mcp --header "Authorization: Bearer <your key>"
AfterLaunch is connected. Show me how each engine answers my tracked questions this week, where we are named, where we are linked, every source cited, and which cited pages name a competitor and not us.

We ran the whole afternoon by hand first. Then we built it into AfterLaunch, with the other moves that came out of the same studies, and nothing it drafts goes live until you say yes.

afterlaunch.io →