Brand reputation monitor

Brand Reputation Monitor

Monitor Hebrew and mixed Hebrew-English brand mentions from permitted exports, public-page comment files, approved APIs, review exports, news-comment dumps, and manually collected evidence. Use the skill to classify sentiment, identify urgent risk, redact personal data, and produce practical response queues for Israeli small businesses, freelancers, local service providers, and consumer-facing teams.

Use official exports, public APIs, approved vendors, or user-provided files. Do not scrape private groups, bypass platform controls, evade rate limits, collect credentials, or automate harassment.

Goals

• Find brand mentions in Hebrew, English, spelling variants, nicknames, hashtags, and local identifiers.
• Classify mentions as positive, negative, mixed, neutral, or urgent.
• Extract operational topics such as delivery, billing, campaign, pricing, safety, privacy, accessibility, staff, legal, and media.
• Score risk using severity, reach, credibility, and actionability.
• Recommend actions: log, reply, route to support, escalate, preserve evidence, or seek qualified review.
• Support Israeli localization with ₪, DD/MM/YYYY dates, Hebrew response templates, and Israeli consumer/accounting terminology.

Inputs

Recommended columns:

Column	Required	Example	Notes
date	Recommended	24/06/2026	Prefer DD/MM/YYYY for local reports.
source	Recommended	facebook, x, tiktok, news_comment, review	Normalize aliases.
brand	Optional	המאפייה של דנה	Useful for multi-brand reports.
author	Optional	@customer	Store only when lawful and necessary.
url	Optional	https://...	Public link or internal evidence reference.
text	Required	טעים אבל המשלוח איחר	Main field for analysis.
engagement	Optional	42	Likes, replies, shares, comments, or views.
topic	Optional	delivery	Use pre-filled value when reliable.
branch	Optional	חיפה	Useful for chains and regional service teams.

Mention discovery

Start with a broad keyword map, run calibration, then narrow false positives.

Category	Examples
Official Hebrew name	המאפייה של דנה, סטודיו נועה, קליניקת אביב
English name	Dana Bakery, Noa Studio, Aviv Clinic
Short name	דנה, נועה סטודיו, אביב
Misspellings	דנא, נעה, אביבב
Public professional name	דנה כהן, נועה לוי, ד״ר אביב
Product/service	עוגת שמרים, שיעור פילאטיס, טיפול שורש
Campaign phrase	מבצע חורף, השקה חדשה, קוד קופון
Location	רחוב הרצל, שוק הכרמל, באר שבע, קניון איילון
Business identifiers	Phone, domain, branded hashtag, Google review name

Hebrew normalization

• Strip niqqud for matching, but preserve original text for evidence.
• Normalize quote marks: " / ׳ / ״.
• Handle prefixes: ב, ל, כ, מ, ש, ה, ו; for example בסטודיו, להמאפייה, שהקליניקה.
• Preserve emojis because they affect tone.
• Treat common names carefully. דנה alone is not enough unless it appears with product, city, domain, branch, phone, or campaign.
• Include gendered forms: ממליץ, ממליצה, מאוכזב, מאוכזבת, מקצועי, מקצועית.

Sentiment labels

Label	Meaning	Example
positive	Clear praise or recommendation	שירות מצוין, אחזור שוב
negative	Complaint, warning, frustration, boycott, or dissatisfaction	לא להתקרב, חוויה מזעזעת
mixed	Praise and criticism in the same mention	המוצר טוב אבל השירות איטי
neutral	Factual mention or question	מישהו יודע מה שעות הפתיחה?
urgent	Safety, legal, privacy, discrimination, media, regulator, fraud, or severe claim	קיבלתי מוצר מסוכן והעסק מתעלם

Useful Hebrew cues

Positive: מעולה, מצוין, מדהים, ממליצה, ממליץ, אדיבים, טעים, נקי, מהיר, אמינים, מקצועית, הוגן, תודה, אחלה, וואו.

Negative: גרוע, זוועה, לא מומלץ, לא להתקרב, רמאים, איחור, מלוכלך, התעלמות, בושה, אכזבה, חיוב כפול, חוצפה, לא עובד.

Urgent: תביעה, עורך דין, חדשות, כתבה, חרם, מסוכן, אלרגיה, הרעלה, הטרדה, אפליה, פרטיות, דליפה, משטרה, משרד הבריאות, הרשות להגנת הצרכן.

Decision tree

flowchart TD
    A[New mention] --> B{Permitted source?}
    B -- No --> B1[Do not store text; document source issue only]
    B -- Yes --> C{Personal or sensitive data?}
    C -- Yes --> C1[Redact and minimize]
    C -- No --> D[Normalize Hebrew and classify source]
    C1 --> D
    D --> E{Urgency cue or high engagement?}
    E -- Yes --> F[Mark urgent and escalate]
    E -- No --> G{Sentiment?}
    G -- Positive --> H[Log praise or thank]
    G -- Mixed --> I[Route to support with balanced response]
    G -- Neutral --> J[Answer factual question or log]
    G -- Negative --> K{Specific and credible?}
    K -- Yes --> L[Assign owner and due date]
    K -- No --> M[Monitor for repetition]
    F --> N{Legal, safety, privacy, media, regulator?}
    N -- Yes --> O[Preserve evidence and seek qualified review]
    N -- No --> P[Prepare short public acknowledgement]

Risk scoring

Dimension	Low	Medium	High
Severity	Mild annoyance	Concrete service failure	Safety, legal, discrimination, fraud, privacy
Reach	No engagement	Active thread	Viral, influencer, journalist, regulator
Credibility	Vague claim	Specific order/date/branch	Evidence, photos, repeated reports
Actionability	No owner	Support can resolve	Management/legal/PR required

Score	Action
0-24	Log only; monitor trend
25-49	Reply if useful; route to support
50-74	Escalate to manager; respond within business day
75-100	Immediate escalation; preserve evidence; prepare formal response

Response playbook

Principles:

• Acknowledge the experience.
• Stay calm, factual, and concise.
• Do not admit unverified legal liability.
• Move private details to private support.
• Do not reveal customer details, employee discipline, payment data, address, health data, or internal logs.
• Do not threaten critics without qualified review.
• Do not copy-paste identical replies at scale.

Templates:

Late delivery

מצטערים על ההמתנה. נשמח לבדוק את פרטי ההזמנה ולטפל בזה מולך ישירות. אפשר לשלוח לנו מספר הזמנה וטלפון בפרטי?

Billing issue

תודה שהעלית את זה. חיוב או מסמך חשבונאי לא ברור צריכים להיבדק מיד. נא לשלוח לנו בפרטי מספר הזמנה או חשבונית, ונחזור אליך עם תשובה מסודרת.

Praise

תודה רבה על הפרגון. שמחנו לתת שירות ונשמח לראות אותך שוב.

Safety claim

תודה שעדכנת. אנחנו מתייחסים לנושא ברצינות ומבקשים לקבל פרטים מלאים בפרטי כדי לבדוק את האירוע באופן מיידי.

Edge cases

• Sarcasm: איזה שירות "מדהים", חיכיתי שעה is negative.
• Slang: אמאלה איזה טעים is positive.
• Mixed language: service אחלה אבל delivery disaster is mixed or negative.
• Competitor comparison: המתחרים יותר זולים is a pricing signal, not necessarily a complaint.
• Same-name false positive: a first name alone must not count as brand match.
• Screenshots/OCR: mark low confidence and verify manually.
• Minors, health, billing, IDs, and medical details: redact and restrict access.
• News comments: lower confidence unless repeated, specific, or highly engaged.

Anti-patterns

Avoid:

• Scraping closed groups or private profiles.
• Storing unnecessary personal profiles.
• Reporting only average sentiment while ignoring one severe incident.
• Auto-replying to safety, legal, privacy, discrimination, or media-risk claims.
• Treating every negative word as a complaint about the brand.
• Translating Hebrew sentiment mechanically without context.
• Publishing screenshots with personal data visible.
• Creating fake positive reviews or coordinated manipulation.
• Using collected mentions for unsolicited marketing.

Troubleshooting map

flowchart LR
    A[Unexpected output] --> B{Too many mentions?}
    B -- Yes --> C[Add co-occurrence rules and exclusions]
    B -- No --> D{Missed mentions?}
    D -- Yes --> E[Add spelling variants, prefixes, hashtags]
    D -- No --> F{Wrong sentiment?}
    F -- Yes --> G[Check negation, sarcasm, slang, mixed language]
    F -- No --> H{Wrong urgency?}
    H -- Yes --> I[Tune urgent lexicon and engagement thresholds]
    H -- No --> J[Inspect encoding, source, and date parsing]

Production checklist

2026 validation note

Use the current Israeli VAT rate of 18% for complaint triage involving ₪, מע״מ, receipts, refunds, and displayed prices. Treat platform data access as permission-bound: Facebook/Page comments, X/Twitter search, TikTok comments, Google reviews, and news comments each require the relevant official API scope, approved export, vendor agreement, or manual evidence workflow.

• Confirm lawful source and platform terms.
• Define keyword variants and false-positive rules.
• Define retention period and deletion process.
• Redact unnecessary personal data.
• Set escalation owners and response SLAs.
• Prepare Hebrew and English response templates.
• Test at least 20 scenarios.
• Review privacy, consumer, billing, and sector-specific obligations where needed.
• Verify DD/MM/YYYY and ₪ formatting.
• Create human-review fields for high-risk items.
• Document limitations and manual override process.

Example report

Period: 01/06/2026 to 07/06/2026
Mentions reviewed: 184
Positive: 61
Mixed: 24
Neutral: 72
Negative: 22
Urgent: 5

Main themes:
1. Delivery delays in Gush Dan after 18:00.
2. Praise for staff at the Haifa branch.
3. Coupon confusion on TikTok.

Immediate actions:
- Reply to 7 unresolved delivery complaints.
- Update coupon landing page.
- Escalate 2 safety-related claims for management review.