Educational: Mass Review Campaigns And Automation Detection Concerns

Ethical Review Coordination Toolkit

This page outlines the design and strategy behind a modular, privacy-first system that assists users in writing and posting ethical, truthful reviews about companies connected to harmful or unlawful behaviour.

Purpose

To empower individuals to:

• Coordinate civic action against unethical businesses
• Reduce friction in review writing
• Avoid violating Google Terms of Service or legal statutes
• Maintain privacy, decentralization, and ethical standards

Legal & Terms of Service Overview

Legal and Platform Compliance

Action	Legal?	Google TOS-Compliant?	Notes
Truthful human-written reviews	Yes	Yes	Fully legal and encouraged
LLM-assisted draft generation	Yes	Yes	Drafts only; human must finalize
Full automation of submission	Risky	No	Violates TOS; detectable by Google
Fake or duplicate reviews	No	No	Can lead to bans or legal issues
Headless/scripted browser posting	No	No	Highly detectable
Mass-coordinated spam	Risky	No	May be considered brigading

Detection Methods to Avoid

Google uses the following to detect abuse:

• Duplicate or near-identical review text
• Sudden bursts of review activity from same IP/device/account
• Sudden mass reviews from new accounts
• Use of headless browsers or scripted tools (e.g. Selenium, Puppeteer)
• Lack of mouse/keyboard activity (no natural mouse movement or keystrokes)
• Unrealistic form fill times (e.g. typing too quickly)
• Reviews that follow a repeated structure or template pattern
• Account age and review density (new accounts with many reviews are suspicious)
• IP address consistency and geolocation mismatch (reviewing businesses in distant or inconsistent regions)
• And Much More: See More: How Websites Can Detect Bots and Automation

Safer Design Principles

• **Human-in-the-loop**: Reviews must be finalized and submitted manually
• **Clipboard & GUI automation only**: Use tools like `pyautogui`, `AutoHotKey`, or `AppleScript` to automate the keyboard and mouse, but do not auto-submit let the human click it. Or avoid it while the website is open and only have human-like mouse inputs.
• Could use automation to just help open the review page and have the response "copied" to the clipboard already so the human just needs to press paste.
• **Max 3-5 reviews per user/day**: Stay below suspicious thresholds for people who want to keep their account, but not built with any enforced restrictions in mind, for people who are willing to lose their account
• **Randomized timing**: Introduce delays and simulate natural typing if needed, things being perfectly 1 second apart 5 times in a row, is very suspicious.
• **Varied review content**: Use LLMs to generate unique, personalized drafts
• Have a place to report getting banned so we can see if it is a problem in us helping them

System Architecture

1. Assignment Server

• Maintains a database of businesses to review (e.g. australian zionist owned business list)
• Offers an endpoint like `/get-random-5`
• No user tracking or account system, unless user chooses to give a location or something (to make results more believable)
• Simply tracks how many times a business has been served, to distribute them all over time
• Modular: Can support unrelated civic tools (reporting, verification)

2. Local Review Assistant (Client Tool)

• Requests 5 businesses from the server
• Generates 1–2 varied review drafts using an LLM
• Presents drafts to the user for manual editing and posting
• Uses `pyautogui` or similar to:
  • Open the review page
  • Allow human to Paste draft text
  • Wait for human to review and click submit
  • Move to next review or allow human to close and reopen later for the next review, so all they have to do is click on the app and then paste then submit a few times a day. (Anything even more convenient I think would be easily detectable by google)
• Includes local-only SQLite DB to track reviewed businesses, or use google review itself to make sure we don't double review anywhere

3. LLM Review Generator

• May be hosted separately (self-hosted, API, or bot interface)
• Accepts:
  • Business name & category
  • Complaint or issue type
  • User writing sample for style adaptation
• Returns unique draft(s) per input
• Users are warned to never post exact duplicates

4. Privacy & Anonymity Practices

• No account system
• Local review tracking only
• VPN encouraged for network anonymity
• Modules operate independently to reduce risk of takedown

Strategic Philosophy

This system, if made properly,:

• Encourages truthful speech
• Does not automate dishonest or mass behaviour
• Cannot be detected as automated due to real human interaction
• Is modular so banning one part won’t affect the whole ecosystem
• Preserves the image of prioritising ethical action, not spam or coercion

> "To ban us, they would have to ban regular users."

Next Steps

• Build public CLI/GUI tools for the Local Review Assistant
• Publish open API spec for the Assignment Server
• Share prompt templates for LLM generation, choose cloud methods
• Draft educational material to help users stay within safe, legal bounds
• Create tools like this get around some of the auto filter methods that prevent what should be protected forms of digital protest using the excuse of stopping illegitimate spammers.

Or if that isn't good enough...

Educational: Learning_Bot_and_Fake_Account_Detection - learn how to do or don't