Google Perspective

Google Perspective

· #279 most-used

Score every comment for toxicity and keep conversations constructive

AnalyticsDeveloperAISecurityMonitoring & AlertsCommunity

Google Perspective is a free machine-learning API built by Jigsaw and Google's Counter Abuse Technology team that analyzes text and returns probability scores for how a human reader would perceive it — across attributes like TOXICITY, SEVERE_TOXICITY, IDENTITY_ATTACK, INSULT, THREAT, and PROFANITY, as well as constructive signals like CURIOSITY and REASONING. Connect it to Actionist and your agents can score every inbound submission, support ticket, community post, or piece of user-generated content before it reaches a human — routing hostile interactions to the right handlers, blocking harmful content from publication, and generating auditable compliance records, all without your team having to read everything first.

Average time saved
6 hours
per person · per month
≈ 1 workdays back

Eliminates manual work. Agents eliminate manual reading of every inbound submission, ticket, and piece of user-generated content to identify hostile or harmful language — replacing human triage with automated Perspective scoring that flags only the items that require human attention.

Schedule

What your Google Perspective agent runs on autopilot

A week of scheduled jobs your Actionist agent will execute on your behalf.

28Scheduled jobs
7Agents at work
24/7Always on
Agents
TueThu
Tue
Wed
Thu
7a
8a
9a
10a
11a
12p
1p
2p
3p
4p
5p
6p
Multi-app workflows

Google Perspective × every other app you use

End-to-end automations that span multiple apps — each one a real business outcome.

6Workflows
7Apps spanned
~102 hrsSaved / week
6Personas served
For operations
Featured5 apps

Form submission pre-screened before it enters the workflow

When a form submission arrives, the agent scores the open-text content for TOXICITY and IDENTITY_ATTACK before any downstream processing. Clean submissions flow into the normal CRM or workflow; submissions above the threshold are logged to a Google Sheets moderation log, posted to the #trust-safety Slack channel, and documented in a Notion moderation case — all before any human reads the content.

~50 hrs

Time saved for your team — every week, on autopilot

The flow
Trigger·When a new form submission arrives via Typeform
Result
Write submission data and toxicity scores to the moderation logPost alert to #trust-safety channel if score exceeds thresholdCreate moderation case in the Trust and Safety Notion database
The win
Saved per run
15 min
Runs / week
~200×
Harmful submissions flagged before they reach any human workflow
Driven byOperations Agent
ROI

Savings

What your team gets back — two angles: what you stop doing manually, and what that's worth.

Without Actionist

What you do manually today

With Actionist

What your agent runs for you

  • Sales
    20 min / week
    Manual review of all inbound messages

    Sales managers manually read every contact form message to identify hostile or abusive leads before forwarding to a rep — several minutes per lead, indefinitely.

    Sales Agent
    0 min
    Agent scores inbound messages and routes by tone

    When a lead message scores above threshold for TOXICITY or INSULT, the agent routes the lead to a manager automatically — junior reps only see clean prospects.

  • Marketing
    60 min / week
    Manual community moderation

    The community team manually reads every user comment, review, and post before approving it for publication — time-consuming at scale and prone to inconsistent standards.

    Marketing Agent
    0 min
    Agent screens UGC for toxicity before publication

    The marketing agent scores all user-generated content before it enters the publication queue. Only content below the toxicity threshold is published automatically — the rest goes to human review.

  • Customer Support
    30 min / week
    Manual triage of difficult tickets

    Support team leads manually skim the incoming queue looking for aggressive or threatening messages to reassign — a task repeated throughout every shift.

    Customer Support Agent
    0 min
    Agent detects hostile tickets and routes to senior agents

    Every incoming support ticket is scored for TOXICITY and THREAT within about a minute of arrival. High-scoring tickets go to senior agents; the standard queue stays manageable for junior reps.

  • Human Resources
    45 min / week
    HR manually reads all anonymous feedback

    The HR team reads every anonymous survey response looking for discriminatory or threatening content before compiling results — emotionally demanding and not scalable as headcount grows.

    Human Resources Agent
    0 min
    Agent screens survey feedback privately before aggregation

    Anonymous survey responses are scored with doNotStore before results are aggregated. Hostile or identity-attack responses surface to HR leadership without the raw content leaving company infrastructure.

  • Finance
    15 min / week
    Manual vendor communication review

    Finance teams read vendor correspondence ad-hoc with no systematic process for identifying threatening or hostile communications until a situation has already escalated.

    Finance Agent
    0 min
    Agent scans vendor communications for hostility signals

    Incoming vendor and supplier messages are scored for THREAT and TOXICITY weekly. High-scoring communications are surfaced to the legal team before the finance team responds — preventing escalation of vendor disputes.

  • Operations
    30 min / week
    Subjective moderation quality estimates

    Platform health is assessed anecdotally from team reports and complaint volumes — no consistent, quantified metric for discourse quality, and no early-warning trend data before problems become visible.

    Operations Agent
    0 min
    Agent produces a weekly platform health score across all attributes

    The operations agent scores a content sample across all six production attributes every week and writes the results to a Community Health Dashboard — leadership has a quantified discourse quality metric to track trends.

  • Legal
    60 min / week
    Manual regulatory content audit

    Legal teams manually sample platform content before regulatory deadlines, relying on keyword search and spot-checking rather than systematic scoring — creating gaps in the audit trail.

    Legal Agent
    0 min
    Agent runs weekly compliance sweeps with auditable records

    The legal agent scores platform content against regulatory thresholds for IDENTITY_ATTACK, THREAT, and SEXUALLY_EXPLICIT weekly, logging every result to the compliance database — providing defensible evidence for regulators without manual content review.

+ 100s of other Google Perspective automations
Average time saved
26 hrs / person / month
Calculator

Calculate what your team saves

Team size
5 people
Hourly rate
$75 / hr
Hours saved / week
8
Hours saved / year
375
Annual ROI
$28,125

Based on Google Perspective's typical team usage — the visible tasks plus a few other automations the agent runs: ~1.5 hrs / person / week of admin work automated.

Connect

How to plug Google Perspective into Actionist

Pick the connection method that suits your environment.

Connect via a Google Cloud API key with the Perspective Comment Analyzer API enabled. No OAuth required — a standard API key scoped to the Perspective API is all you need.

1
Enable the Perspective API in Google Cloud Console

Go to console.cloud.google.com, select or create a project, then navigate to APIs & Services → Library. Search for 'Perspective Comment Analyzer' and click Enable.

2
Create an API key

In APIs & Services → Credentials, click Create Credentials → API Key. Copy the generated key and store it securely.

3
Paste the key into Actionist

Paste the API key into Actionist and click Test connection. The agent submits a test analyzeComment call to confirm the key is valid.

Credentials you'll need
API Key*
Google Cloud Console → APIs & Services → Credentials → Create API key (enable Perspective Comment Analyzer API first)
Actions

14 actions your agent can call

Read and write operations available to your Actionist agent.

Triggers

0 events your agent can react to

Events your agent watches for, and the actions it kicks off in response.

This app has no triggers yet.
FAQs

Questions about Google Perspective + Actionist

How does Actionist connect to Google Perspective?
Go to the Apps tab, find Google Perspective, and click Connect. You will need an API key from the Google Cloud Console: enable the Perspective Comment Analyzer API for your project, create an API key under APIs & Services → Credentials, and paste it into Actionist. The agent runs a test analyzeComment call to confirm the key is valid before any live analysis runs.
What credentials do I need to connect Actionist to Google Perspective?
You need a standard Google Cloud API key with the Perspective Comment Analyzer API enabled. The Perspective API is free to use but rate-limited (by default 1 request per second); you can request a quota increase from the Jigsaw team if your volume requires it. The key does not require OAuth — a plain API key scoped to the Perspective API is sufficient.
Can I connect Google Perspective to other apps in the same workflow?
Yes — Perspective is most powerful as a signal that feeds downstream actions. Common combinations: analyze incoming form submissions or CRM notes and post a moderation alert to Slack; score support tickets and route high-hostility tickets to senior agents; batch-score user-generated content stored in Google Sheets and write scores back as a column; or gate publishing workflows in Notion until a human reviews any comment that exceeds your toxicity threshold. Any of Actionist's 200+ connected apps can send or receive data alongside Perspective in the same agent task.
What are the most common ways agents use Google Perspective?
The six most-used patterns are: (1) comment pre-screening — score user comments before they publish and hold anything above a threshold for human review; (2) support ticket triage — flag hostile or threatening tickets for priority handling; (3) batch content audit — score a backlog of historical content and surface the highest-risk items; (4) form submission filtering — score open-text fields in intake forms before data enters the CRM; (5) employee feedback screening — score anonymous survey responses to detect potential HR concerns early; (6) community health reporting — produce a weekly digest of toxicity trends across a moderated platform.
What attributes can Google Perspective score, and which are production-grade?
The production attributes — TOXICITY, SEVERE_TOXICITY, IDENTITY_ATTACK, INSULT, PROFANITY, THREAT, and SEXUALLY_EXPLICIT — are stable and supported across 18+ languages. Experimental attributes such as INFLAMMATORY, SPAM, OBSCENE, and FLIRTATION are also available but are trained on narrower data and are primarily English-only. In 2025, Jigsaw introduced experimental bridging attributes (CURIOSITY, REASONING, PERSONAL_STORY) that detect constructive signals rather than harmful ones. You can request any mix of attributes in a single API call.
How do I interpret a Perspective toxicity score, and what threshold should I use?
Every score is a probability between 0 and 1 — it is not a binary toxicity verdict. A TOXICITY score of 0.82 means the model estimates an 82% chance a human rater would perceive the text as toxic, not that it definitively is. You set the threshold that makes sense for your context: news comment sections typically use 0.7–0.8; community platforms with stricter norms may use 0.5. Because the model has known biases (it can over-flag content mentioning marginalized identity groups), always build a human review step into any automated moderation workflow rather than auto-deleting purely on score.
Can I score large volumes of content in bulk, and are there rate limits?
The Perspective API does not support batching — each analyzeComment call processes one comment. For bulk scoring (e.g. auditing thousands of forum posts), the agent runs requests sequentially or with controlled concurrency to stay within your quota. Default quota is 1 QPS; if you need higher throughput, submit a quota increase request to the Jigsaw team at perspectiveapi.com. The agent can be configured to pace requests to avoid 429 errors and to log failed calls for retry.
Does Perspective store the text I send for analysis? How do I keep submissions private?
Yes. Send the doNotStore: true flag in each analyzeComment request and Perspective will not store the submitted text on its servers. This is important for analyzing private or sensitive content — employee feedback, customer support transcripts, legal communications — where data residency and confidentiality matter. The flag is available on all attribute types and adds no latency. Actionist sets this flag by default for any Perspective action you configure; you can turn it off explicitly if you want Perspective's team to be able to review the data to improve model quality.