Help Scout AI sentiment analysis: what it has and what it doesn't

Riellvriany Indriawan
Written by

Riellvriany Indriawan

Katelin Teen
Reviewed by

Katelin Teen

Last edited November 21, 2026

Expert Verified
Illustration of two support teammates reviewing customer messages with a mood slider, next to the Help Scout logo
TL;DR

Help Scout does not document any AI sentiment analysis. Its AI features (AI agents, AI Drafts, AI Summarize, AI Assist, Topics) don't score how a customer feels. The built-in signals are the customer's own Happiness rating and workflow conditions on rating and keywords, enough for a rough tone alarm from Standard at $25 per user per month billed annually. To pull a person into upset or sensitive conversations automatically, eesel follows handoff rules you write and leaves an internal note in Help Scout. With the eesel CLI, an AI coding agent can upload approved knowledge, adjust those handoff instructions and ask test questions before anything goes live, though a mailbox admin still approves the Help Scout connection.

Does Help Scout have AI sentiment analysis?

No, not in anything I could find in Help Scout's own pages. I read the AI features page, the docs for AI Summarize, AI Drafts, AI Assist and Topics, the workflow conditions list, and the AI Transparency Hub on 2026-11-21. A search of the docs for "sentiment" returned one unrelated article about Beacon messages.

That is a fair design choice for a tool that says it wants AI to improve support, not deflect tickets. It also means a support lead who wants a "this customer is upset" flag has to build it. Here is what each piece does, and what it does not:

Help Scout featurePlanWhat it readsSentiment?
Happiness ratingsStandard, Plus and Pro (Standard reports 2 years, Plus and Pro lifetime)Customer clicks Great, Okay or Not GoodCustomer-reported, after the reply
Workflow Rating and Body conditionsBasic workflows on Standard, advanced on PlusRating value, words or regex in the messageKeyword matching, not AI
TopicsPlus and ProReason for contact, such as billing or refundNo, categorizes themes
AI SummarizePlus and ProWhole thread, one clickNo tone label documented
AI AssistPlus and Pro (per-user plans)Your own draftRewrites your tone, does not read the customer's
MCP serverStandard, Plus, Pro (reports need Plus or Pro)Conversations and reports, read-onlyOnly if you ask your own AI to judge it

The pricing page also lists an "AI Inbox assistant" under Standard. I could not find it documented as scoring sentiment, so I have left it out of the table. For what sentiment analysis normally means, my AI sentiment analysis overview and the sentiment analysis glossary entry set the baseline.

What does Help Scout measure instead?

Help Scout measures what the customer chooses to tell you. Each email reply can carry a rating link, and the customer picks Great, Okay or Not Good, with an optional comment. The Happiness report turns that into percentages and a Happiness Score: the share of Great ratings minus the share of Not Good ratings, calculated much like NPS.

Help Scout Happiness report showing Great 67 percent, Okay 33 percent and Not Good 0 percent with a Happiness Score of 67, as taken from Help Scout Docs
Help Scout Happiness report showing Great 67 percent, Okay 33 percent and Not Good 0 percent with a Happiness Score of 67, as taken from Help Scout Docs

The rules around it matter for anyone using it as a mood signal:

  • Customers can rate any reply within 10 days and can change the rating in that window, per the satisfaction ratings docs.
  • Each reply gets one rating, and a customer can rate every reply in a thread separately.
  • Ratings are indexed by the date of the reply, not the date of the rating. My Happiness report dates guide walks through why a March report can change in April.
  • Standard plans can report on the last 2 years, and Plus and Pro on the lifetime of the account.
Help Scout ratings table listing customer, user, date, rating and comment columns, as taken from Help Scout Docs
Help Scout ratings table listing customer, user, date, rating and comment columns, as taken from Help Scout Docs

The catch is timing. A rating arrives after the agent has already replied, and only if the customer bothers. The upset customer in message one has no score until a reply has gone out and been judged.

Timeline showing the customer writing and the agent replying with no built-in score, then the Happiness rating arriving up to 10 days later
Timeline showing the customer writing and the agent replying with no built-in score, then the Happiness rating arriving up to 10 days later

If you want the broader metric picture, I wrote up how to measure customer sentiment and what AI CSAT changes.

How much does each Help Scout plan cost?

Prices below come from the live Help Scout pricing page, checked on 2026-11-21.

PlanPer user, billed annuallyPer user, billed monthlyWhat matters for tone work
Free$0 (5 users, 1 Inbox, 1 Docs site)$0Entry plan, no sentiment features
Standard$25$30Basic workflows, AI Inbox assistant listed, MCP and API
Plus$45$54Advanced workflows, unlimited AI Drafts, plus AI Summarize, Topics and AI Assist per the docs
Pro$75$90Unlimited workflows, SLA policies, SSO/SAML, HIPAA compliance
Envoy (add-on)$0.75 per resolutionAI agent that answers customers, billed per resolution

A concrete example: a 10-person team pays $250 a month on Standard, $450 on Plus and $750 on Pro, all billed annually. The jump from Standard to Plus is $200 a month for that team, and it buys Topics, AI Summarize and advanced workflows, none of which score sentiment. Be clear on that before you upgrade for it. My Help Scout pricing guide and Help Scout review cover the rest of the trade-offs.

How do I build a rough tone alarm with Help Scout workflows?

You can get a useful, imperfect alarm from workflows alone. The condition list includes Rating (Great, Okay or Not Good), Rating Comments, Body (customer message or user note, with words or a regex), Tags, Waiting Since and Last Customer Reply. The action list includes Notify Help Scout User(s), Assign to User, Assign to Team, Add a Note, Add Tag(s) and Change Status.

Three recipes I would start with:

  1. Unhappy rating. Rating is Not Good, then add a tag like unhappy, assign to a team lead and notify them. Test it on one sample conversation first, since I could not verify from the docs how quickly a new rating triggers a workflow.
  2. Warning words. Body matches a short list such as "cancel", "chargeback" or "lawyer", then tag and notify. Help Scout advises whole words and several conditions or workflows, because partial matches catch words like "friend" when you meant "end".
  3. Waiting too long. Waiting Since is longer than a set number of hours on a tagged conversation, then assign to a team. Pair it with your SLA policy order so the two do not fight.

Two limits to plan around. A workflow only acts the first time a conversation matches, to avoid loops, and keyword rules only find the phrases you thought of. For routing, my guides to auto-assignment workflows and AI ticket tagging show how to keep the rules tidy.

Ladder of Help Scout signals: Happiness rating after the reply, workflow keyword rule when it arrives, Topics for the reason, AI Summarize on click, and a dashed row for a tone score not in the docs
Ladder of Help Scout signals: Happiness rating after the reply, workflow keyword rule when it arrives, Topics for the reason, AI Summarize on click, and a dashed row for a tone score not in the docs

Keep collision detection in mind when you notify a lead, so the lead and the original agent do not both reply, and check how snoozed conversations behave if an upset thread wakes up later.

Can Topics, Summarize or the MCP server stand in for sentiment?

Partly, and each one answers a different question.

  • Topics reads every new conversation and assigns a theme such as billing issue or refund request. When you turn it on, Help Scout reviews your history and proposes 8-15 topics, and the Topics docs say you can use a topic as a workflow condition during the beta, plus see trend reports. A spike in "refund request" is a mood signal in practice, but the label is the reason, not the feeling.
  • AI Summarize gives a bulleted recap on click. The docs say it can be inaccurate or misleading, the summary is always in English, and it is web-app only. It is good for a handoff, as I covered in Help Scout AI conversation summaries.
  • AI Assist has an "adjust tone" button, but it rewrites your draft to sound more professional or friendly. It does not read the customer's mood.
  • The MCP server is the most open door. Help Scout lists examples like finding the objections customers raise most often, and it supports pulling happiness reports. Connections are read-only, so Claude or ChatGPT can analyze tone across last month's conversations but cannot add a note or tag. That is a weekly review, not a live alarm.

If native scoring is a hard requirement, compare how other helpdesks do it:

That may also be a reason to read Help Scout alternatives or Help Scout vs Front.

Reviewers are honest about the reporting side. A G2 reviewer rated Help Scout 5/5 and still wrote:

G2

"The reporting functionality is great but basic, sometimes we have to export data or use other tools for more detailed analysis."

How does eesel fit into Help Scout sentiment handling?

I would not describe eesel as a sentiment dashboard. Its Help Scout docs describe triggers and actions, and a sentiment score is not among them. What it does well here is act on cues you define. eesel wakes when a customer opens a conversation or replies, reads the thread, and can leave an internal note, add tags and assign the conversation to a person or team, or leave a draft for a teammate to review. It learns from your past Help Scout conversations (closed ones from the last 30 days by default) and your Docs site.

The practical difference from a workflow keyword rule is that the instruction can be a sentence in plain language, such as "if the customer asks for a person, mentions an injury or says they are cancelling, hand off and write why", and the rule can be checked on past conversations before it touches a live customer. When it wakes it starts from customer messages and team replies, not your internal notes, so put the context that matters in the rule itself.

Flow from a Help Scout conversation to eesel reading the thread, then splitting into a handoff branch with an internal note and a routine branch with a draft reply
Flow from a Help Scout conversation to eesel reading the thread, then splitting into a handoff branch with an internal note and a routine branch with a draft reply

Two eesel customers show the pattern, both on Freshdesk rather than Help Scout, so treat them as the same idea on a different helpdesk:

  • Original Bootcamp runs eesel on Freshdesk, and a filter lets through only the tickets it can answer from the website and past tickets. Anything about an injury or illness goes to a person even if the booking question could have been answered, and the filter let 31 tickets through between 8 August and 8 September 2026. That is a sensitive-topic rule written as a sentence, not a sentiment score.
  • SamCart runs one agent on Freshdesk tickets and another on its website chat. When a seller asks for a person or the question is beyond the help center, the ticket is tagged for handoff and the agent stops replying on it.

That is the shape I would copy on Help Scout: define the moments a person must take over, make the agent say why in a note, and let the team keep the judgment calls. For the process side, see AI escalation, the ticket escalation process and AI agent handoff best practices.

Which setup should a Help Scout team pick?

Pick by what you need to be true on a Monday morning.

If you needUsePlanLimit
A trend of customer-reported satisfactionHappiness reportAny planLagging, only customers who click
A flag on named words or bad ratingsWorkflowsStandard for basic, Plus for advancedOnly the phrases you list
Themes behind a volume spikeTopicsPlus or ProReason, not mood
A weekly tone review across many conversationsMCP plus Claude or ChatGPTStandard (reports need Plus)Read-only, manual
A person pulled in at the right moment, with a noteeesel handoff rulesHelp Scout plan plus eeselActs on your rules, no published score

My own default for a team under 15 agents is workflows plus the Happiness report first. They cost nothing beyond the plan you have, and they show you which phrases and ratings actually predict a bad outcome. Add the assistant when the keyword list starts growing past what anyone can maintain, or when the cost of missing one escalation is higher than the cost of the tool. For churn-sensitive accounts, see detecting churn risk in conversations, and for the broader question, AI sentiment analysis for support and AI ticket sentiment are the next reads.

If you review conversations for coaching, pair any of these with a quality assurance routine, and read intents and sentiments to separate what the customer wants from how they feel.

If the tone problem is really an answering problem, start with Help Scout AI draft replies and AI ticket deflection. The roundup of the best AI for Help Scout and my AI for Help Scout guide cover the options.

Setting up Help Scout customer properties such as plan tier helps you rank who gets a person first, and the glossary entries on escalation and NPS fill in the vocabulary.

Try eesel for Help Scout

If your Help Scout team needs the right person on the upset or sensitive conversations without waiting for a rating, eesel connects to your mailbox in minutes, follows the handoff rules you write, leaves an internal note explaining why, and can be tested on your past conversations before it replies to anyone. Start with drafts or notes only, then widen what it can do once the rules hold up.

Frequently Asked Questions

Does Help Scout have AI sentiment analysis?
Not as a documented feature. Help Scout's AI features page lists AI agents, AI Drafts, AI Summarize, AI Assist and Topics, and a search of the Help Scout docs for "sentiment" returns nothing about conversations. The closest built-in signal is the customer's own Happiness rating. For the general idea, see my AI sentiment analysis overview.
Which Help Scout plan do I need for AI features that touch tone or mood?
AI Summarize and Topics are on Plus and Pro, and AI Assist's tone rewrite is on Plus and Pro for per-user plans. On the pricing page that is $45 per user per month billed annually for Plus and $75 for Pro. None of them scores sentiment, so the plan decides what you can summarize and tag, not what you can measure. My Help Scout pricing breakdown has the full plan math.
How is the Help Scout Happiness score calculated?
Help Scout takes the percentage of Great ratings and subtracts the percentage of Not Good ratings, much like NPS, per its Happiness report docs. It only reflects customers who click a rating, so it is a lagging signal. The CSAT glossary entry explains how it compares to other satisfaction metrics.
Can a Help Scout workflow detect an angry customer?
Only by keywords. The Body condition can match words or a regex in the customer's message, and the Rating condition matches Great, Okay or Not Good. That catches the phrases you list and the ratings people give, not tone you did not anticipate. My guide to dealing with angry customers covers the human side.
Can I use Claude or ChatGPT to check sentiment on Help Scout conversations?
Yes, as analysis after the fact. Help Scout's MCP server lets Claude, ChatGPT and Cursor search conversations and pull reports, but access is read-only, so the assistant cannot add notes or tags. It suits a weekly review, not a live alert.
Is there an AI that follows handoff rules on Help Scout conversations?
eesel connects to Help Scout and can leave internal notes, add tags and assign a conversation to a person or team, following instructions you write. It does not publish a sentiment score, so I treat it as acting on cues you define. See AI agent handoff best practices for what to put in the rules.
Does Help Scout AI Summarize tell me how the customer feels?
It writes a bulleted recap of the key points, per the AI Summarize docs, in English only and only in the web app. Mood may show up in the wording, but the docs do not describe a tone label, so I would not rely on it for escalation. My Help Scout AI summaries guide goes deeper.

Share this article