Graphs

Measured, not memorized

Base LLMs hallucinate brand affinities from token co-occurrence in training data.
Graphs returns measured ones.

The attention graph maps which entities compete for the same audiences across 8,400 brands, influencers, public figures, products, and interests — derived from real consumer engagement data, not training corpora.

Version-pinned. Citeable. Sub-50ms warm.

Query handles. Get graphs.

curl https://graphs.socialsignal.ai/v1/attention/similar/starbucks

{
  "handle": "starbucks",
  "domain": "attention",
  "version": "2026.01",
  "results": [
    { "handle": "upsstore",       "title": "The UPS Store",            "score": 0.93 },
    { "handle": "coffeeandtea",   "title": "Coffee & Tea",             "score": 0.93 },
    { "handle": "fedex",          "title": "FedEx",                    "score": 0.92 },
    { "handle": "traderjoes",     "title": "Trader Joe's",             "score": 0.92 },
    { "handle": "officesupplies", "title": "Office Supplies & Services","score": 0.92 }
  ],
  "latency_ms": 34
}

Two endpoint families.
/similar/{handle} returns entities with similar audience shape.
/affinity/{handle} returns entities your target audience over-indexes on.
Pairwise similarity by handle pair, text resolution, taxonomy, full handle search — all under /v1/attention/.
Handles are permanent. A handle issued today resolves to the same entity in every future version.
Scores are version-pinned. Each response carries its data version. Pin in production; upgrade deliberately.

Different question, different answer

/similar vs /affinity

/v1/attention/similar/starbucks

The UPS Store           0.93
Coffee & Tea            0.93
FedEx                   0.92
Trader Joe's            0.92
Office Supplies         0.92

/v1/attention/affinity/starbucks

Costco                  0.93
Costco Gasoline         0.92
Cold Stone Creamery     0.91
Bakeries & Desserts     0.90
Sunglass Hut            0.90

Entities your target audience over-indexes on — Starbucks customers also buy in bulk and indulge.

Same anchor. Two different questions. Two different answers — both grounded in measured behavior, neither inferable from training-data co-occurrence.

Agents skip the handle namespace

Graphs ships an MCP server at mcp.socialsignal.ai. Agents send a text query; the server resolves it to a handle and returns the graph result. No handle lookup, no caching layer, no integration glue.

{
  "name": "graphs_attention_similar",
  "description": "Returns entities with similar audience shape to the given query.",
  "input": { "q": "string" }
}

Configure once in your MCP client. Available free-tier by default for discovery, with rate limits. Paid keys unlock production throughput.

MCP Server Docs →

Graphs Pricing

Per call. Transparent.
No contracts.

Free

1,000 calls / day
Rate-limited ~100/hr
Email → key

Standard

Per 1,000 calls
No subscription
Pay as you go

Volume

Custom

1M+ calls / month
Volume pricing
Reach out

Pricing applies to all /v1/attention/ endpoints equally.
Top up or pause anytime. Cancel by not making calls.

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form.

How the attention graph was built.

The attention graph is derived from approximately 52 million public social media profiles, 117 million mobile devices, 200 million desktop devices, 281 million individual demographic records. The 80 deep consumer segments derived from this inform entity-to-entity audience overlap using cosine similarity on mean-centered persona affinity vectors.
‍
The actual lookup is sub-microsecond — two hashes and two array reads against a network-edge binary. The latency you observe is request handling, JSON serialization, and TLS. We report the round-trip because that's what you can measure. Edge processing time is consistently 0.00ms. It felt like bullshit to show that on the demo.
‍
Updated 1–2× per year. Every response is pinned to its data version.

Affinity graphs for AI applications

Measured, not memorized

Query handles. Get graphs.

Different question, different answer

Agents skip the handle namespace

How the attention graph was built.

Other services