Review Intelligence Agent — G2, Trustpilot, Glassdoor & Capterra in One Report

Turn scattered customer and employee reviews into a one-page intelligence report. No NLP service required, no per-API key, no glue code.

Most teams need to know one thing fast: what are real customers and employees saying about a company right now — and how is that changing? Browsing G2, Trustpilot, Glassdoor and Capterra by hand is hours of work for a single company; scaling it across a target list is a week.

This actor does it in one call. Feed it any company plus the public URLs of its review pages, and it returns:

• An aggregated star rating across every connected source
• Star-NPS proxy score (4–5 share minus 1–2 share, the way investors actually look at SaaS reviews)
• Star trend — last 30, 90, 180 days vs all-time
• Sentiment breakdown — positive / mixed / negative / neutral
• Top complaint themes — pricing, support, performance, onboarding, UI/UX, feature gaps, integrations, mobile, docs, data accuracy
• Top praise themes — ease of use, value, speed, reliability, feature breadth, integrations, onboarding, support
• Notable positive and negative reviews with star, source, and date
• Recurring phrase cloud mined from the actual review text
• An AI-written executive narrative that ties it all together in plain English

Pay-per-report pricing. No subscription. No surprises.

---

Who this is for

• Investors & due-diligence teams — vet a SaaS or DTC target in 30 seconds before deeper diligence
• Competitive intelligence teams — track 50 competitors across 4 review sources without hiring an analyst
• Sales reps — walk into a pitch already knowing what your prospect's customers love and hate
• Product teams — spot complaint clusters around your competitor's last release
• CX leaders — benchmark your own brand against the field
• Agencies — sell a "review intelligence" deliverable to clients without building it yourself

---

How it works

1. You provide an array of companies. Each company is just { name } plus any subset of g2Url, trustpilotUrl, glassdoorUrl, capterraUrl.
2. The actor pulls reviews from each provided source (capped per source for speed).
3. It dedupes across sources, scores sentiment, mines complaint and praise themes, computes star trends, and writes a narrative.
4. Each company produces one report. You are charged per report (PPE).

Provide more sources = stronger signal. The minimum useful setup is one URL.

---

Input

{
    "companies": [
        {
            "name": "Acme Corp",
            "g2Url": "https://www.g2.com/products/acme/reviews",
            "trustpilotUrl": "https://www.trustpilot.com/review/acme.com",
            "glassdoorUrl": "https://www.glassdoor.com/Reviews/Acme-Reviews-E12345.htm",
            "capterraUrl": "https://www.capterra.com/p/123456/Acme/"
        }
    ],
    "maxReviewsPerSource": 75,
    "reportStyle": "full"
}

Field	Required	Default	Description
companies[].name	Yes	—	Company name (used in the report headline)
companies[].g2Url	No	—	Public G2 reviews URL
companies[].trustpilotUrl	No	—	Public Trustpilot review page
companies[].glassdoorUrl	No	—	Public Glassdoor reviews page
companies[].capterraUrl	No	—	Public Capterra product page
maxReviewsPerSource	No	75	Per-source cap. 10–300
reportStyle	No	full	full (detailed markdown) or brief (one-page exec summary)

---

Output — one record per company

Each company in companies produces one row in the dataset. Key fields:

{
    "company": "Acme Corp",
    "scrapedAt": "2026-05-16T12:00:00.000Z",
    "totalReviews": 187,
    "averageStars": 4.31,
    "starNpsProxy": 58,
    "starDistribution": { "1": 6, "2": 9, "3": 18, "4": 64, "5": 90 },
    "sentimentBreakdown": { "positive": 121, "mixed": 30, "negative": 32, "neutral": 4 },
    "topComplaints": [ ["Pricing & billing", 19], ["Customer support", 12], ... ],
    "topPraise": [ ["Ease of use", 33], ["Customer support", 21], ... ],
    "recentTrend": { "last30d": 4.40, "last90d": 4.36, "last180d": 4.33, "last30dCount": 11, ... },
    "sourceSummary": [ { "source": "g2", "count": 75 }, ... ],
    "notablePositiveReviews": [ ... ],
    "notableNegativeReviews": [ ... ],
    "aiNarrative": "Acme Corp averages 4.31/5 across 187 aggregated reviews. Star-NPS proxy is 58 (strong). Sentiment: 65% positive, 17% negative. Recurring complaints: Pricing & billing (19), Customer support (12). Praise themes: Ease of use (33), Customer support (21).",
    "intelligenceReport": "# Review Intelligence Report — Acme Corp\n…"
}

intelligenceReport is a fully formatted markdown report you can drop straight into a CRM, Slack, Notion, or PDF generator.

---

Pricing

$1.25 per company report (pay-per-event). One company in companies = one report = one charge. You only pay for completed reports.

Why this is a steal: collecting and analyzing reviews for a single company across 4 sources by hand is roughly 60–90 minutes of analyst time. Even at $40/hour, that's $40–60 of labor per company. This actor delivers the same artifact for $1.25, in under a minute, with a consistent format you can compare across hundreds of companies.

---

Common workflows

• Build a competitive matrix. Run 20 competitors in one input. Export the dataset to CSV. Sort by starNpsProxy. You now have a ranked SaaS leaderboard.
• Pre-call sales intel. Run a prospect 10 minutes before the demo. Open the topComplaints field — you now know what to lead with.
• M&A diligence shortlist. Run your target list. Filter by recentTrend.last90d falling vs averageStars. That's your "review velocity is bad" alert.
• Brand alarm system. Schedule weekly. Diff the topComplaints field. New theme appearing? That's your investigation.

---

Tips for high-quality reports

• Provide all four URLs when possible. Cross-source dedupe is what lets the report reach "executive" quality.
• Use maxReviewsPerSource: 150+ for deep dives on a small list. Use the default 75 for fast triage runs across many companies.
• For B2B SaaS, G2 and Capterra usually carry the highest-signal reviews. For DTC and consumer brands, Trustpilot. For employer-brand work, Glassdoor.
• Run it on yourself first. The report often surfaces complaint themes your own team has been brushing off.

---

What this actor does not do

• It does not scrape behind login walls. Provide public review URLs only.
• It does not translate reviews — non-English reviews are analyzed at lower precision.
• It does not call an external LLM. The narrative is built from on-device heuristics, which is what makes the per-report cost so low.

If you need true LLM-summarized review reports at scale, contact us — we run a parallel actor for that workflow.

---

Limits & honesty

Public review pages change layouts. We use multiple selector fallbacks per source and degrade gracefully — if one source returns 0 reviews, the report continues and clearly labels that source as empty.

Built and maintained by Johnson AI Consulting — the same shop that ships the Hiring Signal Intelligence Agent and the Competitor Intelligence Monitor on Apify Store.