# The Second Audience: How Machines Read, Represent, and Act on Companies Across the Web
> How machines read, represent, and act on companies across the web. A measurement course on the Second Audience: the machine-mediation layer between companies and people. Learn to investigate what machines receive, classify roles by proximity to intent, and keep measurement honest.

Canonical URL: https://audiencetwo.com/second-audience
Markdown twin: https://audiencetwo.com/second-audience.md
Canonical thesis: https://audiencetwo.com/second-audience/thesis
Thesis markdown: https://audiencetwo.com/second-audience/thesis.md
Provider: AudienceTwo

## 01. THE COURSE: The internet has a second audience.
How machines read, represent, and act on companies across the web.
A measurement course built on a canonical thesis. Not a prompt pack. Not a ranking checklist.

Read the canonical thesis: [The Second Audience: How Machines Read, Represent, and Act on Companies Across the Web](https://audiencetwo.com/second-audience/thesis)

## 02. ROLES: Not all bots are the same.
The useful taxonomy is not primarily by AI company. It is by role, and by how close that role sits to human intent.
GPTBot and ChatGPT-User can involve the same vendor and still be different economic events. A training crawl may be months removed from any customer. A live retrieval may occur because someone is considering a purchase right now. An agent authorized to purchase is different again.

### Role taxonomy (must stay split by role, not vendor)
- **Live reader**: Person asked; page fetched to answer. Examples: ChatGPT-User, Claude-User, Perplexity-User
- **Answer indexing**: Index for future answers. Examples: OAI-SearchBot, Claude-SearchBot, PerplexityBot
- **Training**: Corpus / foundation-model crawl. Examples: GPTBot, ClaudeBot, Meta-ExternalAgent
- **Search**: Classic SEO crawl. Examples: Googlebot, Bingbot, Applebot

### Same-vendor pairs
- GPTBot is not ChatGPT-User. Training crawl vs live reader fetch. Same company. Different economic event.
- ClaudeBot is not Claude-User. Corpus building vs a person who asked seconds ago.

## 03. OUTCOME: What you will be able to do.
Take any product URL, classify the machine traffic around it, inspect what live AI readers actually receive, score which commercial facts survive, design a controlled experiment, and distinguish what you measured from what you inferred.

### For
- Ecommerce, growth, SEO, and analytics leads who suspect AI answers matter but cannot see machine traffic in GA
- Brand and agency PMs / engineers who need a shared vocabulary with marketing
- Operators who want evidence over influencer frameworks

### Not for
- People looking only for ChatGPT prompt packs
- Anyone wanting a guaranteed rank-in-AI checklist
- No-code agent builders hunting automation recipes

## 04. CURRICULUM: The map. Then the inspectable surfaces.
The thesis is larger than the course. We cannot instrument all seven roles equally. This course concentrates on machine access, representation, fact survival, referrals, and experiments.
- **Part 0 - The map of the Second Audience**: Training → Indexing → Retrieval → Answering → Decision → Agentic → Transactional. Five pillars. Then what we can measure today.
- **Part I - Who is on the machine side?**: Roles by proximity to intent. Same-vendor pairs stay split. Live readers are one region of the map.
- **Part II - What do machines receive?**: Same URL, different experience. Invisible middle. Blocked, unusable, wrong, repair, disagreement.
- **Part III - Surfaces and the economic chain**: Discoverability vs consumability. Representation through transaction. Agent Offers as one experiment.
- **Part IV - What is actually happening?**: Growing brand study. What instruments cannot see. Unanswered questions. Run the experiment yourself.

## 05. EVIDENCE: Real product pages. Not hypotheticals.
Public brand sheets score what live readers receive: price, availability, shipping, returns, size, offer.
- [Bombas](https://observe.audiencetwo.com/bombas): 0 of 6. Blocked at the edge (Vercel Security Checkpoint).
- [Brooklinen](https://observe.audiencetwo.com/brooklinen): Repair archetype. Most commercial facts do not survive the extract.
- [Unbound Merino](https://observe.audiencetwo.com/unboundmerino): Readable extract. Offer often absent. Different problem. Same method.
Browse the full public set at observe.audiencetwo.com.

## 06. DISCIPLINE: What we know. What we don't.
Every claim in the course carries a label. That is the differentiator.
- **Observed**: Seen in a specific fetch, extract, or answer under stated conditions.
- **Supported**: Repeatedly true across multiple brands or tests. Still not universal law.
- **Unknown**: Important, but unmeasured or underpowered.
- **Speculation**: Allowed when labeled. Never sold as fact.

## 07. WAITLIST: Get notified when it opens.
The course is in preparation for Udemy. Join the waitlist for launch access and updates. Read the canonical thesis anytime.
- Thesis: https://audiencetwo.com/second-audience/thesis
- Thesis PDF: https://audiencetwo.com/course/The-Second-Audience-Thesis.pdf
- Waitlist: submit name, email, and role via the HTML form on https://audiencetwo.com/second-audience#waitlist (POST /api/lead with context=course-waitlist)

## The experiment continues.
- Here is the tooling.
- Here is the methodology.
- Here is the growing dataset.
- Here is what we've observed.
- Here is what remains unanswered.

Go measure it yourself.

AudienceTwo builds the instrumentation. The Second Audience teaches the discipline.

- Workbook: https://github.com/camburley/second-audience-workbook
- Home: https://audiencetwo.com/