G2 Reviews Scraper & Switching-Signal Monitor

Turn G2 reviews into recurring competitor intelligence.

Monitor G2 product reviews and detect new negative reviews, pricing complaints, competitor mentions, feature gaps, reliability issues, and customer switching signals. Persistent state means scheduled runs can return only reviews discovered since the previous successful run.

What this Actor does

Raw review exports answer “what did reviewers write?” This Actor is designed to answer “what changed?”

• Monitors one or many public G2 product review URLs.
• Establishes a baseline on the first successful run.
• Remembers stable review IDs in a named Key-Value Store that persists across runs.
• Emits only unseen reviews on later runs when Only new reviews is enabled.
• Detects commercially useful signals without OpenAI, Anthropic, Gemini, or another paid AI API.
• Stops pagination after 20 consecutive known reviews, keeping scheduled runs small.
• Tracks public aggregate rating and review-count changes when G2 exposes them reliably.
• Writes detailed records to the default Dataset and a run-level summary to OUTPUT.

Common use cases

Competitive intelligence

Find public reviews that say customers switched from or are moving to another SaaS product.

Product research

Discover missing-feature and reliability complaints in recent competitor reviews.

Pricing research

Detect new complaints about price increases, billing, cost, and perceived value.

Customer intelligence

Identify public dissatisfaction with competing products without building a separate monitoring service.

Reputation monitoring

Detect newly published 1–2 star reviews during scheduled runs.

SaaS founder research

Turn repeated public pain points into evidence for product opportunities.

Input

{
  "products": [
    {
      "name": "Slack",
      "url": "https://www.g2.com/products/slack/reviews"
    },
    {
      "name": "Microsoft Teams",
      "url": "https://www.g2.com/products/microsoft-teams/reviews"
    }
  ],
  "maxReviewsPerProduct": 10,
  "onlyNewReviews": true,
  "detectSignals": true,
  "competitors": ["Microsoft Teams", "Slack", "Discord", "Zoom"],
  "minimumRating": 1,
  "maximumRating": 5,
  "includeReviewText": true,
  "dateCutoff": null,
  "emitReviewsOnBaseline": true,
  "stateStoreName": "g2-review-switching-signal-monitor-state",
  "proxyConfiguration": {
    "useApifyProxy": true,
    "apifyProxyGroups": ["UNBLOCKER"]
  }
}

products accepts 1–25 public URLs shaped like https://www.g2.com/products/<slug>/reviews. The URL, not the optional display name, is the monitoring identity. Plain URL strings, http://, missing www., localized paths (/es/products/...), review permalinks and other product tabs are normalized to the canonical review listing, and duplicates are removed. Entries that are not G2 product URLs are skipped and reported in OUTPUT.productErrors with status INVALID_INPUT; the run fails with a clear message only when no usable product remains.

maxReviewsPerProduct limits recent reviews inspected per product. The API default is 10. Increase it for deeper baselines; scheduled runs usually stop earlier after finding known reviews.

The Store prefilled example intentionally requests one Slack review and disables onlyNewReviews. Apify's automated Store QA requires a successful run with a non-empty Dataset within five minutes, while API and scheduled runs retain the stateful onlyNewReviews: true default.

dateCutoff is optional and uses YYYY-MM-DD (an ISO date-time is truncated to its date). Pagination stops after older reviews are reached.

emitReviewsOnBaseline controls first-run output. When enabled, current reviews are returned with baselineRun: true and isNew: false. When disabled, the first run establishes state without emitting review rows.

Recurring monitoring

1. Run the Actor once to establish a baseline.
2. Confirm baselineRun: true in OUTPUT.
3. Create an Apify Schedule for the Actor or a saved Task.
4. Keep the same stateStoreName so the named Key-Value Store remains available across runs. Use a different name to isolate another monitor.
5. Later runs emit only unseen reviews when onlyNewReviews is enabled.

State is committed once, after product processing and Dataset writes finish. If the run is interrupted, it does not advance the stored watermark and silently lose reviews. A failed product does not erase its prior state. State retains up to 50,000 recent IDs per product.

G2 applies advanced traffic protection. The production default uses Apify Unblocker, which is billed separately per successful request. The Actor sends one document request per fetched review page, does not perform a billable warm-up through Unblocker, and never reports blocked or unparseable extraction as success. Remove any country selection unless geo-targeting is required; allowing Unblocker to select the route improves reliability.

Do not purge the named state store or change stateStoreName between scheduled runs. A new store name creates a new baseline. Apify storage names may contain only letters, digits and hyphens, so other characters (for example dots, underscores or spaces) are replaced with hyphens. If the state store cannot be opened, the run still returns reviews as a baseline and records a warning in OUTPUT.warnings.

Review output

{
  "recordType": "REVIEW",
  "reviewId": "987654",
  "productName": "Slack",
  "productSlug": "slack",
  "productUrl": "https://www.g2.com/products/slack/reviews",
  "reviewUrl": "https://www.g2.com/products/slack/reviews/slack-review-987654",
  "title": "Too expensive for our team",
  "text": "We switched from Slack to Microsoft Teams because the price increased.",
  "rating": 2,
  "reviewDate": "2026-08-28",
  "reviewerName": "A. Reviewer",
  "reviewerRole": null,
  "reviewerCompany": null,
  "reviewerCompanySize": "Mid-Market (51-1000 emp.)",
  "likes": null,
  "source": "g2",
  "sourceLocale": "en",
  "scrapedAt": "2026-08-29T09:00:00.000Z",
  "isNew": true,
  "baselineRun": false,
  "signals": [
    {
      "type": "PRICING_COMPLAINT",
      "confidence": 0.86,
      "evidence": "We switched from Slack to Microsoft Teams because the price increased."
    },
    {
      "type": "SWITCHING_TO",
      "confidence": 0.96,
      "competitor": "Microsoft Teams",
      "evidence": "We switched from Slack to Microsoft Teams because the price increased."
    }
  ]
}

Unavailable public fields are null; they are never invented. When G2 does not expose a source ID, the Actor creates a deterministic SHA-256-based ID from the product slug, review URL, date, reviewer name, and title.

Signals

One review can contain multiple signals:

• NEGATIVE_REVIEW — rating of 2 stars or lower (G2 uses half stars and also shows 0-star reviews).
• PRICING_COMPLAINT — negative pricing, billing, cost, or value phrases.
• SUPPORT_COMPLAINT — explicitly negative support/service context.
• BUG_OR_RELIABILITY_COMPLAINT — bugs, crashes, downtime, errors, or poor performance.
• FEATURE_GAP — missing capabilities and feature requests.
• SWITCHING_FROM — phrases such as “switched from X” or “previously used X.”
• SWITCHING_TO — phrases such as “moving to X” or “using X instead.”
• COMPETITOR_MENTION — case-insensitive, word-safe match of a configured competitor name.

Confidence values are deterministic heuristic scores, not machine-learning probabilities. Evidence is a short excerpt around the matching phrase. Full configured competitor names are matched: configuring Microsoft Teams does not make the generic word teams a match.

Run summary

The run's default Key-Value Store record OUTPUT includes run status, baseline state, succeeded/failed product counts, fetched/emitted/new review counts, signal totals, competitor mention counts, aggregate changes, pages fetched, timing, and per-product errors. Durable review watermarks are kept separately in the named stateStoreName store.

{
  "status": "SUCCEEDED",
  "baselineRun": false,
  "productsProcessed": 2,
  "productsSucceeded": 2,
  "productsFailed": 0,
  "reviewsFetched": 48,
  "reviewsEmitted": 3,
  "newReviews": 3,
  "signalsDetected": 5,
  "negativeReviews": 1,
  "pricingComplaints": 1,
  "supportComplaints": 0,
  "featureGaps": 1,
  "switchingFromSignals": 1,
  "switchingToSignals": 1,
  "competitorMentions": { "Microsoft Teams": 1 },
  "pagesFetched": 2
}

If one product fails, the run status in OUTPUT is PARTIAL, successful products are retained, and the failure is listed in OUTPUT.productErrors with an actionable message. Errors are not written as Dataset rows, so you are never charged a review event for a failed product. If a later results page cannot be retrieved, the reviews from earlier pages are still saved and the product is marked complete: false with a stopReason in OUTPUT.productStats. The Actor run fails only when no product could be monitored at all.

Pricing

This Actor uses pay-per-event pricing:

• G2 review saved: $0.004 per review row written to the Dataset on the Apify Free plan, with lower prices on paid Apify plans (down to $0.0028).
• Actor start: $0.01 per GB of run memory, charged once when the run starts.

Proxy (including Apify Unblocker) and compute usage are included in these prices. Failed products, skipped products and diagnostics are not written to the Dataset and are not charged. When the run's maximum total charge is reached, the Actor stops saving reviews, skips the remaining products, and lists them in OUTPUT.productErrors.

Performance and reliability

With the default Unblocker strategy, the Actor requests public G2 review HTML with Apify's recommended got-scraping client and sends one document request per review page (no billable warm-up). Each attempt is bounded to 55 seconds. A page gets up to six attempts with exponential backoff across newest-first full, localized and review-fragment surfaces, within a 150-second retry budget per page. The surface that worked for the previous page is tried first on the next page. Authentication errors stop immediately, a missing product (repeated HTTP 404) stops after two attempts, and normal-looking pages without parseable reviews stop after three attempts; the first such page is saved to the DEBUG_UNPARSEABLE_PAGE record for diagnostics.

The remaining run time is shared fairly between pending products, and the Actor always keeps time to save the Dataset, monitoring state and OUTPUT before the run timeout. Deep crawls that reach the time budget return the reviews collected so far instead of failing; increase the run timeout for very large maxReviewsPerProduct values. Pagination also stops when a page only repeats earlier reviews. For custom or non-Unblocker proxies, a resource-restricted Chrome fallback remains available, but G2 usually blocks those routes.

Apify Unblocker is enabled in the Store input default because G2 commonly rejects ordinary datacenter and residential sessions. No private proxy credentials are hard-coded and no third-party API keys are required.

Privacy and public-data scope

This Actor collects only fields displayed on public G2 review pages and needed for review monitoring. It does not log in or access private profiles. Its default Apify Unblocker route may handle anti-bot/CAPTCHA challenges automatically; use it only when you are authorized to collect the target's public pages. Disable includeReviewText when downstream consumers only need metadata and detected signals.

Review text and reviewer display information can be personal data. Use the output only for a lawful purpose, minimize retention, and follow G2’s terms and applicable privacy rules.

Limitations

• G2 actively restricts automated traffic. Unblocker materially improves access but is not a guarantee and can add tens of seconds per page.
• Website structure and access conditions can change. A product fails with an explicit error when its first page reports reviews but none can be parsed; it is never reported as an empty success.
• Public G2 markup does not expose every field on every review; unavailable values remain null.
• Keyword/regex signals are explainable but do not understand sarcasm, every linguistic nuance, or languages other than English.
• English review pages are always tried first. When only G2's Spanish or Portuguese page is reachable, review text is G2's translation of the review and sourceLocale is es or pt; English-only signal rules then detect fewer signals.
• Switching entities that are not in the configured competitor list are best-effort phrase extractions.
• Aggregate rating/count changes are emitted only when both prior and current public values are available.
• First-run reviews are a baseline, not urgent new events, even if baseline rows are returned.

Development

npm install
npm run check
npm start

The project uses TypeScript, Crawlee/Playwright, Cheerio, Apify Key-Value Store, and Vitest. Tests cover parser behavior, missing/malformed fields, ratings/dates, signal combinations and false positives, stable IDs, state transitions, and transactional staging.