Live signal · Starter+

For every episode you publish, who was on it and how the audience reacted to each of them.

Video sentiment. Host, guest, mentioned. Top praise and top criticism, ranked by engagement. Claim Truth verdicts on the cast. Conflation warnings when the audience appears to be talking about a different-but-same-named entity. Plus a Guest Impact roll-up of which appearances actually moved AI Visibility 7 days pre → 7 days post.

Per-episode cast read

Video sentiment plus per-cast-member (host / guest / mentioned) sentiment with sample size. Top praise and top criticism ranked by likes. Everything a §8 genuine-empty read when the stream is silent — no fabricated shares.

Claim verdicts on the cast

A sponsor claim that we've verified as contradicted, a host claim now under dispute — the Claim Truth Engine's verdict is stitched onto the cast entity so it lands next to the audience reaction, not in a separate tab.

Conflation warnings

When several comments attributed to the guest arrived at low NER confidence and the ambiguous share is material, we flag it: possible conflation — several comments appear to refer to a different-but-same-named entity. Hygiene warning, not a verdict.

Guest Impact roll-up

For each guest appearance in the window, compare mean AI Visibility 7 days pre vs. 7 days post. Top lifters and top drags surface honestly; null delta when either window is empty, so a brand-new show never invents a lift.

See it live

Guest Intelligence is enabled on every demo entity — see the band on their entity page.

PBD Podcast· Podcast (YouTube + audio)
Piers Morgan Uncensored· Interview / opinion channel
MrBeast· YouTube channel

How it works

1. Ingest. The YouTube + podcast comment pipeline writes per-entity comment_signal rows (sentiment + intent + NER attribution). The episode pipeline writes episode_entityrows (host / co-host / guest / mentioned).
2. Reduce. A pure reducer joins the two per-episode: per-cast sentiment (positive / neutral / negative counts → sample-size-carrying %), praise and criticism ranked by likeCount, and honest §8 empty reads where nothing was attributed.
3. Stitch. For each cast member the reader adds (a) Claim Truth verdicts (from the coded Claim Truth Engine — no LLM grades truth), and (b) a low-confidence NER conflation warning when material.
4. Roll up.Guest appearances in the last 90 days are paired with the show's own visibility snapshots; a per-appearance 7d pre / 7d post delta rolls into top lifters and top drags. Never rated when the window is empty.

Get the data

REST — GET /api/v1/entities/{id}/guest-intelligence (layer_1_profile).
MCP — entity_guest_intelligence tool (mirrors the REST payload).
CLI — entidex guest-intelligence <slug>.
SDK — await client.guestIntelligence('slug').

FAQ

What is Guest Intelligence?

A per-episode multi-role read for a show or creator. For each episode we surface the video-level sentiment, then break it down per cast member — host, co-host, guest, mentioned, and (heuristic) sponsor — with sample size, top praise and top criticism ranked by engagement, a Claim Truth verdict where the cast member has made a claim we can check, and a conflation warning when the NER attribution to that entity was low-confidence on this episode. On top of that we roll a window up: which guest appearances actually moved AI Visibility 7 days pre → 7 days post.

How is it different from YouTube Intelligence or the Perception module?

YouTube Intelligence tells you what the audience is asking, what intents dominate the comment stream, and where the show anomalises. Perception tells you where each of the AI engines describe and recommend the entity. Guest Intelligence sits in the middle: the unit of analysis is the episode + its cast, and the reads on each cast entity are honest — a guest with zero attributed comments comes back with a null sentiment and a sampleSize of zero (§8 genuine-empty), never a fabricated positive share.

How is Guest Impact computed, and when does it appear?

For each guest appearance we compare the show's mean AI Visibility in [pivot − 7 days, pivot − 1 day] against the mean in [pivot + 1 day, pivot + 7 days]. A delta is only recorded when both windows have snapshots — otherwise the guest carries a null delta, so the read never invents a lift or a drag. That means Guest Impact for a brand-new show lights up over weeks, not immediately.

What does a conflation warning mean, and why is it different from a fabrication verdict?

It is a first-pass conflation check: when several comments were attributed to the cast entity at low confidence and the attribution share is material (≥3 comments and ≥20% of the entity's signal on this episode), we flag it. The signal is 'the audience may be talking about a different but same-named entity' — a hygiene warning, not a fact-check verdict. Claim verdicts, when present, come from the Claim Truth Engine (a separate coded pipeline).

What does this cost to run?

Zero LLM cost, no supplier calls. Every read is a coded reduction over stores the pipeline already writes: comment_signal (per-entity, sentiment + intent, VADER-scored), episode_entity (host / guest / mentioned links), score_snapshots (AI Visibility history), and the claim-truth store. The band self-fetches an internal route so it lands with no billable request.

Go deeper with Entidex

The full entity intelligence platform — beyond Explore Entidex

Continuous multi-source entity observation
Alerts when sources diverge or drift
Cross-surface consensus + visibility over time
Evidence-anchored intelligence reports

See pricing