Build 05 / Respectful Audience Discovery
Audience Finder AI
A respectful audience discovery system with a simulated opportunity queue, fit scoring, local browser memory, and human-gated decisions. The model surfaces candidates. The operator decides what is real.
Greg's take
Audience selection in B2B marketing is usually a confident-sounding exercise over partial data. A segment gets defined, a list gets pulled, a campaign gets launched, and the results get explained afterwards.
The AI layer makes this more efficient. It does not automatically make it more reliable. The same judgement calls are still happening - they are just faster and harder to audit.
This build puts the evidence and the operator decision back into the audience selection workflow.
The problem
AI can find audiences faster, but speed does not make the audience trustworthy.
Segments hide assumptions
Audience decisions are built on partial signals, ageing data, market assumptions, and manual judgement.
Scoring needs explanation
A fit score without evidence is just a black box with a number attached.
Review cannot disappear
The operator needs to see the suggested site, assess fit, and decide whether it deserves action.
Memory prevents repeat waste
Accepted and rejected domains should be remembered so the system learns from review decisions.
Demo scope
A public demo of the review loop, not an outreach automation engine.
Static/simulated queue
Browser demo stores review decisions locally.
Human gated
Every opportunity is checked before any action.
LocalStorage memory
Reviewed domains are remembered in the browser demo.
Demo only
No outreach automation, scraping, remote writes, or auto-submission.
Agent architecture
A respectful audience intelligence loop.
This public demo shows the review-loop model: simulated discovery, fit scoring, local memory, and human review before action. The production path connects approved discovery inputs, AI qualification and scoring, memory, human review, and client-ready outputs.
Finder Agent
Discovers candidate websites, communities, directories, newsletters, podcasts, and events.
Qualification Agent
Scores fit, authority, backlink value, respect value, effort, and spam risk.
Pitch Agent
Drafts a respectful submission angle or introduction note for manual review.
Memory and review
Stores reviewed opportunities and keeps Greg in the approval loop before action.
Public demo
Audience opportunity review queue.
The queue below uses sample and simulated opportunities only. Decisions are stored in local browser memory for the demo.
Active opportunities
0
Accepted
0
Rejected
0
Known domains
0
Discovery controls
Run controlled discovery
Adds simulated opportunities from the local seed pool. No web search, scraping, remote writes, or auto-submission.
V1 limit: ~5 candidates per session
Last simulated run: never
Export
Review handoff
Browser-only exports for reviewed records and accepted opportunities. Nothing is sent anywhere.
Discovery - V1 controlled
Discovery is human-controlled in this public demo. Around 5 candidates are surfaced per session from local sample data, and nothing is written outside the browser.
Opportunities
0 visibleAll current opportunities have been reviewed.
Run discovery simulation to add another batch of suggested sites.
Greg's take
Audience selection is usually treated like a clean targeting exercise. It is not. It is a pile of assumptions, partial signals, ageing data, and judgement calls dressed up as segmentation.
The AI can help. But it should not quietly decide who matters and who does not. That decision needs to be visible, evidence-backed, and operator-controlled - not embedded in a scoring model that nobody can explain when the campaign underperforms.
This build turns audience selection into a queue with evidence, scoring, accept/reject decisions, and an operator who stays in the loop at every step. The model surfaces candidates. The human decides what is real.
An AI that recommends audiences without a review step is not a targeting system. It is a black box with an export button. The value is not in the recommendation - it is in the evidence trail and the operator decision that follows it.
Want to discuss Audience Finder AI?
For demos, speaking leads, respectful community introductions, or ideas for the audience discovery workflow.
