I tried the same ChatGPT prompt Sam Altman used to build a website. It helped me finally learn how to vibe code.

Reporter Uses Prompt Attributed to OpenAI CEO to Generate Trip-Planning Website

Business Insider reporter Kelsey Vlamis tested a prompt she said originated from OpenAI cofounder and CEO Sam Altman. The prompt directed an AI model to analyze chat history, generate three long-weekend trip options for a group of nine, produce a full-stack coordination website, and draft an invitation email. Vlamis copied the text into her ChatGPT account on the $20-per-month Plus plan with the GPT-5.6 Sol model set to high intelligence.

Nine minutes and 47 seconds later, the model returned three destination options, source code for a website, and an email draft. The locations were Santa Ynez Valley, South Lake Tahoe, and Todos Santos. Vlamis reported that two of the suggestions aligned with places already on her personal list.

The returned files included a README that described installation and deployment steps. Vlamis stated she lacked the technical background to interpret the instructions. She then asked the model for plain-language, sequential guidance. The model directed her through specific actions, including opening the Terminal application on a Mac with Command + Spacebar, installing Python, creating a GitHub account, and deploying the site through the Render hosting service. She completed local testing before making the site publicly accessible.

Vlamis paid Render $7.25 per month for hosting. When she later inquired about free alternatives, the model recommended Cloudflare and stated the deployment steps would have been comparable. An OpenAI spokesperson told Business Insider that a feature called Sites, available in public beta to Plus subscribers, allows users to create, preview, edit, and host sites directly inside ChatGPT without third-party services. The model acknowledged it could have suggested this option initially.

The resulting site permitted users with an access code to create profiles, enter budgets, vote on destinations, select available dates, indicate interest in listed activities, and add comments. The site also included a mechanism for the organizer to lock in a final destination and generate assignable tasks. Vlamis did not write any original code; she copied and pasted short command strings supplied by the model.

After the site was live, Vlamis transferred the project into the Sites workspace and requested additional changes, such as allowing users to edit or delete comments and repositioning the access-code field. A separate chat session processed the revisions in 14 minutes and 8 seconds.

Vlamis noted that the prompt she used was presented without modification from the version she attributed to Altman. She reported that the exercise demonstrated a workflow for directing the model to handle technical tasks outside her prior experience. The article described the process as taking under two hours while other work continued in parallel.

The published account did not include a link to the original X post or independent verification that Altman posted the exact text. All other details in the report consist of Vlamis’s direct observations of the model’s output and the resulting website functions.