Meta AI Studio Walkthrough and Product Record

Meta AI Studio Product Walkthrough Record

Meta AI Studio is a Meta product for creating and sharing custom AI characters. The safest way to use this record is to separate the interface Dalton tested in July 2024 from what Meta documents now, then verify the authenticated product again on the day of publication.

The live product route is ai.meta.com/ai-studio. Access, fields, eligible accounts, regions, sharing surfaces, controls, policies, and data behavior can change.

flowchart TD
    A["2024 Venture Step recording"] --> C["Historical product record"]
    B["Current Meta sources"] --> C
    C --> D["Publication-day authenticated test"]
    D --> E["Supported current guidance"]
    C --> F["Methods that remain product-neutral"]

What the 2024 recording shows

In episode 28, Dalton accessed AI Studio through the web, explored public characters, and created Curio, a friendly character focused on fun facts.

The recording shows a configuration flow built around a name and identity, a description, instructions, a welcome message, introductory prompts, example dialogue, testing, and revision. Dalton described the interface as approachable for someone without programming experience.

He also reported an account boundary from his own experience: creation worked through his personal account but not through the business profile he tried. That is dated personal evidence. It should not be presented as a current eligibility rule.

The most durable part of the walkthrough is the loop after creation. Dalton tested Curio, found its fact responses too list-like, changed the instruction to encourage storytelling, and tested again.

What Meta announced

Meta's July 2024 launch record described AI Studio as a place to create, share, and discover AI characters. It said a character could be private, shared with followers and friends, or made discoverable across certain Meta surfaces.

For creator AIs, the announcement described customization based on Instagram content, topics to avoid, links, reply controls, and visibly labeled responses. Those statements establish what Meta announced at that time. They do not prove that every feature remains available to every account or region.

Meta later introduced other AI products, including a standalone Meta AI app, Creator Assistant, and Meta Business Agent. Those products should not be collapsed into AI Studio. A current article must name the exact product being tested.

A current verification pass

Before publishing instructions, a reviewer should record the date, country, account type, device, app or browser version, entry route, authentication state, and every visible creation field.

The reviewer should create a private test character, capture the available identity and behavior controls, inspect any content-source options, note distribution choices, test visible AI disclosure, find reporting and deletion routes, and document whether the character can be disabled.

No screenshot should expose private messages, account identifiers, access tokens, unpublished content, or other people's personal information. Product screenshots need capture dates, alt text, source attribution, and a replacement trigger when the interface changes.

Configuration is not model training

Names, descriptions, instructions, and example dialogue can steer a model. They do not establish that the user trained a new foundation model.

Likewise, a character described as an expert has not earned expertise merely because the description says so. Reliability depends on the underlying model, available sources, retrieval behavior, safeguards, evaluation, and the actual question.

The distinction matters for public claims. "I configured an AI character to answer cooking questions" is materially different from "I trained a cooking expert."

Privacy and source boundaries

Meta's Generative AI privacy guide explains Meta's own description of generative AI and user controls. The current Meta Privacy Policy describes broader collection, use, sharing, retention, transfer, and rights.

These are publisher statements. A creator still needs to decide what material they have the right to use, whether it contains information about other people, who may interact with the assistant, what records exist, and how correction or deletion will work.

A public character should not receive confidential files, private conversations, unreleased business information, customer data, or another person's likeness simply because an upload field is available.

What to test before sharing

A creator should test routine questions, ambiguity, missing facts, out-of-scope requests, harmful requests, identity questions, unsupported endorsements, sensitive data, hostile language, and reporting.

The test should verify that AI status is visible, the represented identity is accurate, the character does not imply authority it lacks, and a human owner can correct or disable it.

The NIST Generative AI Profile provides a system-level risk frame. [[How to Test a Public AI Assistant Before Sharing It]] turns that idea into a practical release exercise.

Record status

As of July 28, 2026, the public AI Studio route responds, and Meta's historical launch record remains available. This draft does not claim that Dalton's 2024 fields, account limitation, or distribution options are current.

For a broader comparison of AI Studio with Gemini Gems, use [[Gemini Gems and Meta AI Studio Product Record]] from episode 33. Recheck every product-specific statement before publication.

This record was developed with AI assistance from the recovered E028 YouTube captions, Meta's official product and privacy pages, and the linked research record. Dalton Anderson remains the author. Authenticated product, regional, account, interface, policy, privacy, safety, accessibility, source, and founder review are mandatory before publication. Publication is not authorized.