JBD documentation

The Reputation Score is a composite 0–100 measure of brand reputation health.

It starts from a neutral baseline of 50 and moves ±50 based on how positive vs. negative your mentions are, weighted by the reach of each mention — a single viral negative post matters more than many small positive ones.

• 70–100 → strong positive reputation
• 40–69 → neutral / mixed
• 0–39 → negative-skewed, action recommended

The Presence Score measures how visible your brand is in the monitored window, regardless of sentiment.

It blends log-scaled reach (weight 70%) and log-scaled mention count (weight 30%), each normalized so the realistic range — from a small local brand up to a global brand with reach in the billions — maps onto 0–100 without saturating at the top. Log scaling means a doubling from 10K → 20K reach is the same jump as 1B → 2B, so the score stays comparable across very different brand sizes.

The Influence Score (brand level) captures whether your brand is being amplified by a few very high-reach sources (major news outlets, large accounts) rather than only small mentions.

It sums the reach of your top-5 highest-reach mentions and log-normalizes that total. It is computed over the full mention set (not a sample), so it reflects all of your coverage in the window.

*Note*: A related per-author score exists on the Influencers page — that version ranks individual profiles.

AVE is a PR industry standard that estimates how much it would have cost to reach the same audience through paid advertising, based on each platform's typical CPM (cost per thousand impressions).

We use the following per-platform CPM estimates (USD per 1,000 impressions):

AVE is a directional metric — it's not a replacement for properly modelled media value, but it lets you talk to finance and PR stakeholders in a dollar amount they recognize.

Share of Voice (SoV) = your brand's mention volume ÷ (your brand + tracked competitors).

When no competitor data is available (Talkwalker project not yet configured), SoV defaults to 100 — treat with caution. When you configure competitor tracking, this value reflects real-world category share.

Every individual brand mention we collected via Talkwalker (X, news, blogs, Reddit, Facebook, Instagram, YouTube, Bluesky, forums, …) in the current time range. Each mention is deduplicated by external_id (platform + post id) so reposts don't double-count.

Reach is the estimated number of people exposed to a mention. For social media, reach is typically the author's follower count at the time of posting. For news / blogs, we use the outlet's estimated audience.

Total reach is the sum across all mentions in the selected window. It's an upper-bound estimate — a follower may not have actually seen the post due to algorithmic distribution.

Platform-native engagement counters combined into a single number. Different platforms weight differently (Reddit points vs Twitter likes vs LinkedIn reactions) so use this as a relative trend, not a cross-platform absolute.

We count a mention as user-generated if the platform is a social network AND we successfully resolved an author handle. This excludes news, blogs, and other non-social surfaces.

Subset of total reach that came from social platforms: X, Instagram, TikTok, YouTube, Facebook, LinkedIn, Threads, Bluesky, Mastodon, Reddit.

Compare against Non-Social reach to understand whether your visibility is driven by earned media (news) or organic social activity — each has different half-life and remediation playbook.

Reach from non-social surfaces — news, blogs, forums.

Non-social reach is typically more persistent (news articles stay indexed and rank for months) but slower to move — editorial cycles, not algorithmic bursts.

Stacked bar showing how your total mention volume decomposes across platforms, with Positive / Neutral / Negative counts per source.

Read this chart to answer: *"On which platform is sentiment trending negative?"* A platform where negatives dominate may need a dedicated response playbook.

The selected date range controls how far back in time the KPIs and charts query. Longer ranges are smoother (noise-resistant) but less responsive to this-week events.

Defaults: 30 days on Overview. Changing the range re-queries sns_mentions with the new boundaries.

To keep dashboard load times fast, tenant-level KPI snapshots are cached in memory for 3 minutes. The (cached) marker just tells you "these numbers are from cache, not re-aggregated live".

Click 🔄 Refresh to force a fresh computation — useful right after you know a new mention has come in.

The Sentiment Score is the classic −1 to +1 balance, displayed here multiplied by 100 for readability (so −50 to +50).

Unlike the Reputation Score, it does not weight by reach — it's a pure count-based sentiment index useful when you want to see whether conversation volume is shifting tone, independent of how big the accounts are.

A positive mention expresses approval, praise, satisfaction, or positive intent toward your brand. Sentiment classification happens at ingest time using the upstream source's tag (Talkwalker) or the LLM when the source doesn't tag sentiment.

Aggregate-level positive counts are reliable; individual mentions may occasionally be mis-classified due to sarcasm or domain-specific language.

Neutral mentions are those that reference your brand without a clear positive or negative stance — news announcements, list inclusions, product specs, informational posts.

Neutral volume is often the largest bucket for B2C brands and is a leading indicator of general awareness.

A negative mention expresses dissatisfaction, criticism, service failure, or other adverse sentiment toward your brand. The Storm Alert system specifically watches for spikes in this count.

positive_count / total_count × 100

A healthy B2C brand typically sits in the 40–70% positive ratio range. Values above 80% may indicate filter-bubble / friend-network bias. Values below 20% warrant action.

negative_count / total_count × 100

Use 15% as a rough action threshold — above that, set aside team time to investigate root causes. Sustained values over 25% signal a reputational issue that needs crisis response planning.

The Composite Risk is the arithmetic mean of each engine's risk score, where each engine score is built from:

• AI post risk — how risky AI Overview / Featured Snippet content looks
• High-risk SERP items — count of suspicious/negative URLs in top-10
• Medium-risk items — lower-severity flagged items
• Negative-sentiment suggestions — autocomplete suggestions carrying negative intent

Tones used on the dashboard:

• 0–29 → low / green
• 30–59 → medium / amber
• 60–100 → high / red

For each tracked domain we maintain a 30-day rolling rank history. The 30-day delta is:

currentRank − rankThirtyDaysAgo

Because rank 1 is the best, a positive delta means the site got worse (e.g. moved from rank 3 to rank 7 = +4). The UI shows this with a red ▼ arrow. Negative deltas (rank improved) show as green ▲.

When users search your brand on Google or Bing, the AI Overview / Featured Snippet is often the first answer they see. We grade that content on 0–100:

• 0 → fully positive, on-narrative
• 50 → neutral, factual
• 100 → hostile or narrative-hijacked

An AI Overview Risk ≥ 60 warrants immediate attention because the answer itself can spread faster than any human-authored SERP result.

Across all monitored search engines, we count search results that rank in the top 10 (page 1) and carry negative sentiment — complaints, controversies, unfavorable coverage, look-alike / imposter sites.

This is the most actionable signal for a brand security team: every negative result in page 1 is something you need a response strategy for.

Count of unique domains that have appeared in the top 10 search results for any of your tracked keywords in the last 30 days. This is the tracking coverage — the wider this number, the more of your SERP landscape you're watching.

Autocomplete suggestions and "related searches" phrases that we pick up per tracked keyword. Each phrase is re-classified for sentiment daily — a negative phrase appearing means Google/Bing now associates your brand with that sentiment in its user-intent model.

Out of the 6 monitored engines (Google / Yahoo / Bing / YouTube / X / Instagram), how many currently have real data feeding the dashboard. Engines without real data render on mock / inferred backfill so the UI remains informative during setup.

Each of the 6 monitored engines has its own composite risk score. The chart overlays all 6 lines so you can see which engine is driving the overall risk. If Google is flat but Bing is spiking, the issue is likely a Bing AI Overview or a particular imposter site ranking there.

For each day and each monitored engine, we classify the top-10 SERP items as high / medium / low risk. This chart stacks those counts so you can spot:

• Drift: red band slowly growing = narrative shift
• Spike: sharp red jump = likely a viral negative article ranking
• Recovery: shrinking red band = brand security work is paying off

Each line tracks one tracked domain's rank position over 30 days. Colors: green = positive-sentiment domain, gray = neutral, red = negative.

Reading the chart:

• A red line moving down the Y axis = a negative site climbing higher on the SERP → bad
• A green line moving down = a positive site climbing → good
• Lines disappearing from view = site dropped out of top 10

Search autocomplete reflects what Google/Bing think users are searching for in connection with your brand. A rising negative suggestion (e.g. "<brand> scam", "<brand> problem") is an early-warning indicator that negative intent is becoming associated with your brand in search.

The chart plots per-sentiment counts over 30 days; the individual-phrase view shows which specific suggestions are ranking and moving.

The related queries section of a SERP (also called "People also ask" or "Related searches") is Google's model of what other queries are *semantically nearby*.

Tracking the sentiment of these related phrases tells you whether your brand is being contextualized positively or negatively in the search graph — a stickier signal than suggestions because Google considers the query context more carefully.

A 0–100 aggregate risk score specifically for AI-generated search results — AI Overviews, Featured Snippets, and similar. Because these are *the first answer* many users see, an AI Overview that frames your brand unfavorably can reach more people faster than any traditional SERP item.

Components:

• Presence of negative-sentiment claims
• Presence of unverified statements
• Presence of competitor comparisons where you lose
• Stability (how often it changes day-to-day)

For each day, how many of your top-10 results were positive vs negative vs neutral. A slowly shifting ratio tells you whether your brand is winning or losing narrative control on search.

engagement / reach × 100%

Typical ranges (rough industry benchmarks):

• Instagram: 1–3%
• TikTok: 4–9%
• X: 0.5–1%
• LinkedIn: 2–5%
• YouTube: 2–6% (for community tab; videos are different)

Use as a platform-relative trend, not a cross-platform absolute.

Scores individual public profiles that mention your brand:

log(reach + 1) + engagement × 0.1

Higher = more likely this single profile is worth engaging with or befriending (media tour, guest post invite, etc.).

These are the windows during which posts by you or your competitors in your industry historically earn the highest engagement rate. Use as a starting point — your own audience will refine these over time.

A unified calendar view of every post that's scheduled, drafted, or pending approval across all connected platforms. Each day cell shows platform-colored dots — at a glance you can spot gaps (days with no content), stacks (too many posts competing), and platform mix (e.g. "Mondays are too IG-heavy").

Drag to reschedule. Click a cell to see the full list of posts for that day.

Each connected platform shows:

• Handle (@username)
• Follower count with 7-day delta (+/-)
• Posts / week rolling count
• Engagement rate per platform (platform-native)
• Last sync timestamp — shows "Reconnect" if the OAuth token expired

Accounts needing reconnection are visibly grayed out so you can fix them before your scheduled posts fail.

Chronological list of the next 14 days of content. Each item shows:

• Platform icons (a post can publish to multiple platforms simultaneously)
• Scheduled time
• Status badge: scheduled (approved + queued) / draft (not yet submitted) / approval needed (waiting on client)
• Caption preview

Click any item to edit or reschedule.

For agencies and in-house teams with a review layer. Each card shows the post preview, who submitted it, how long it has been waiting, and Urgent flag if it is close to its scheduled time.

Clicking Approve releases the post to the publishing queue. Request edit sends a structured comment back to the author.

Each connected platform renders as a separate line so you can see which channels are compounding and which are cooling. Click a legend entry to isolate one platform.

Sudden drops often mean the algorithm changed. Sustained rises often mean a narrative is working — check Top posts for the winners.

Top 5 posts ranked by a blend of likes + comments + shares + reach. Use this to extract content themes that work and pitch them into next month's calendar.

One pane of glass for all inbound social messaging. Each item shows:

• Platform icon
• Author handle
• Type (@mention / DM / comment)
• Sentiment badge (green / gray / red)
• Urgent flag (triggered by certain keywords like "help", "broken", "angry")
• Message preview + time

Replying from the inbox creates a native platform reply.

The Auto SNS workspace includes three on-demand AI tools powered by the agent runtime LLM:

• CaptionAI — given a topic or link, draft platform-optimized captions (X tone vs LinkedIn tone).
• ImageAI — brand-voice-constrained visual generation for post headers.
• Inbox auto-reply — draft a reply for any social inbox item, ready for you to edit and send.

All outputs are suggestions, never sent automatically without your click.

The agent watches for emerging topic clusters and hashtag velocity across platforms and surfaces the top 5 of each. Use as content jumpstart — not as must-post.

A Storm Alert fires when one of these metrics deviates by more than 2 standard deviations from its 7-day rolling baseline:

• Daily mention count
• Daily total reach
• Daily negative-sentiment count
• Daily positive-sentiment count

Severity is assigned by how far past 2σ the value went:

• minor (2σ–2.5σ)
• moderate (2.5σ–3σ)
• major (3σ–4σ)
• critical (≥4σ)

The autonomous agent then writes a short explanation (≤2 sentences, using the sample mentions from the anomaly day) so you don't have to investigate manually.

The cloud-tick runs the anomaly detector (the tick fires every 30 minutes; the detector itself is guarded by a TTL so it does meaningful work a few times a day). It:

1. Aggregates today + previous 7 days of mentions / reach / sentiment counts. 2. Computes rolling mean + standard deviation. 3. Flags any day in the last 3 days that exceeds 2σ deviation. 4. Inserts a row into metric_anomalies (idempotent via unique index on tenant/metric/date). 5. For the single most-recent unexplained anomaly, calls the LLM with a prompt that includes up to 5 sample mentions from that day, and stores the explanation.

This is the Storm Alert + AI Anomaly Detector pattern.

Every 6 hours, the autonomous agent takes the latest 150 mentions and asks the LLM to cluster them into 5–8 topic clusters, each with:

• A short human-readable label (e.g. "Pricing and evaluation")
• A 1-sentence summary of what the cluster is about
• The member mentions

Per-cluster we compute:

• Mention count
• Total reach
• Total engagement
• Sentiment breakdown (positive / neutral / negative)
• Share of Voice — cluster mentions ÷ total in batch

This is the AI Topic Analysis feature — a qualitative summary of what the market is talking about, not just how much.

When the agent runtime writes a section report (rotating through all sections, each refreshed when it is older than ~2 hours), it also proposes specific next moves with:

• An explicit recommended action
• The evidence that triggered it
• Expected impact

Proposals can be pinned to stay visible in the Strategy workspace.