All engines
Claude
ChatGPT
Gemini
Perplexity
Grok
DeepSeek
Raw training
▶ Play the Recall
◎ Explore
🎲 Same query, different answer
⏳ Why memory fades
🔀 How fan-out works
❔ New here? What is this & how to use it
AI Memory · narrator
Hover or tap a node
This is a simplified model of how an AI answer comes together. Each node is a stage or engine; the glowing links are how your data flows and combines. Click one to fire a memory pulse, or ask the graph below.
Curious where YOUR brand actually sits in this machine? We map it for real.
Get a free AI Memory Audit →
What you're seeing Each sphere is one memory an AI engine holds about a brand; lines are the associations it follows between them. Hover any sphere for its label; click to focus. Move around 🖱 Drag rotate · Scroll zoom · Right-drag pan 👆 Drag rotate · Pinch zoom · Two fingers pan
engine
Why you don't get the same answer twice
Same query · different answer
A fact may be available, but the source cited can change from run to run. Press the button and watch which source wins.
🎲 Run the same query again
Sampling temp
0.35
Top candidates this run (near-tied)
Press “Run the same query again” to start the distribution.
Why memory fades + gets replaced
Memory decay over time
Drag the timeline (or press Play). Your source can lose visibility as competitors publish fresher, more relevant, or more authoritative pages.
Time
Day 0
Day 0 30 90 365
▶ Play time forward
Illustrative source freshness 100%
↺ Re-establish presence
Before answering, one question often becomes many
How fan-out works
Behind the scenes, the engine often rewrites your one question into several hidden searches. Each engine does it differently. Type a buyer question, then flip engine chips up top to SEE the shape change.
best CRM for a small agency
top project management tools
best running shoes for flat feet
shape: parallel
Representative sub-questions (reconstruction)
Representative reconstruction, not the engine's literal internal queries. The concept is real; the per-engine examples are illustrative approximations.
↻ Replay fan-out
Use this: how to get AI to ingest YOUR data
Every stage in this machine is a lever you control:
• Fan-out → cover the sub-questions, not just the head term (each engine searches several).
• First result set → answer the question directly at the top of the page; chained engines build on what they read first.
• Pack → write self-contained chunks a model can lift whole (one idea per paragraph, plain claims).
• Attribute → clean structured data + consistent entity naming so the citation points at you.
• Prior-only engines → be well-documented across the open web BEFORE their next training snapshot.
This discipline is GEO →
NYFTY Labs · AI Memory Sphere
how leading AI engines handle and combine your data
Pause Replay ✕
1 / 6
Deciding whether to search…
×
NYFTY Labs · AI Memory Sphere
What is this, in plain language?
Explain like I'm 5
Engineering view
×
NYFTY Labs · AI Memory Audit
See the real map for your brand.
You just watched the machine. We'll run YOUR brand through it, which engines name you, who's cited instead, and where you're fading. Free, one business day.
💡 USE THIS: GET AI TO INGEST YOUR DATA