Why does advice from other people matter?

Searching the internet is easy now. Knowing whose opinion to believe—for you—is harder.

Your own preferences are the foundation. Advice from a friend or expert you already trust can make the answer better. That's the multiplier—not a giant social network you have to feed.

People already pay for chat over a trusted source (askJancis is one example). The scarce thing isn't more search results—it's knowing where a recommendation came from.

The hard part of old recommendation apps

Recommendation networks have a supply problem.

Asking people to write reviews, tips, or lists creates work before the network provides much value. Apps like Foursquare tips, Yelp friends, Path, and Nuzzel ran into versions of that problem—among other reasons they struggled. Supply was hard to sustain.

Old ask

Write the review. Maintain the list. Feed the network.

What I'm testing

You already said you loved it. May that help someone you've said yes to?

Trust here is declared—Maya marks Leo as someone whose restaurant taste she trusts—not an algorithm claiming “similar taste.” Figuring out affinity automatically remains unsolved.

A few people, with reasons

Not a giant friends list—a small circle she chose.

Maya’s neighborhood

Shared and connected through the graph

Extracted context links people who matter—so Maya and others can decide with judgment that already exists nearby.

In Maya’s usable graph Nearby · not usable for her Locked

Maya · her context

Visiting CDMX · prefers quiet · avoids dairy

Anniversary trip · held back

Leo · childhood friend

Loves food · childhood friend · Maya marked his restaurant judgment as trusted

Quintonil + Contramar excellent · Pujol too formal · quiet-night shortlist

Sofía · CDMX food blogger

Maya found her on TikTok

Quintonil quieter tonight · dairy-free handled well · Roma Norte energy

4 more nearby · not usable for her
Private context · not shared

Who said it beats how popular it is

Why this is different from “ask ChatGPT to research restaurants.”

ChatGPT can already produce a good list. What it can't do alone is weight that list by people Maya specifically trusts.

Population model

Many records → one segment

Segment-level prediction

“People like Maya choose Pujol.”

Scale finds similarity.
Context can establish relevance.

With Si

A few trusted sources

See Maya's dinner →