Google Maps Review Sentiment & Priority Analysis

Turns Google Maps review records you already have into aggregated, business-level insights. It uses no AI/LLM (everything is deterministic and rule-based), makes no external requests, and does not scrape Google. Reviewer names, IDs, profile URLs and photos are ignored, and review text is never exported.

Built by PAL (Practical AI Lab).

What you get per business

• Rating distribution: counts and percentages for 1-5 stars, average, median, standard deviation, positive/neutral/negative share.
• Rating trend: linear slope of monthly average rating (improving, stable, declining, or insufficient_data).
• Frequency trend: reviews in the last 90 days versus the previous 90 days, measured from the newest review in the data (not from today).
• Monthly series: review count and average rating per month, with empty months filled in.
• Recurring issue topics: a keyword lexicon (service/staff, wait time, price/value, cleanliness, product quality, booking/billing, noise/atmosphere, parking/access). It reports how many reviews mention each topic and how many of the 1-3 star reviews do.
• Outlier patterns (aggregate counts only): rating/text mismatches, repeated identical texts, burst days, polarized distributions, empty-text share.
• Owner response rate.

Keyword matching is simple and English-oriented. Treat topics as a triage signal, not a verdict. Businesses with fewer than 10 reviews are flagged lowSample.

Try the ready-to-run sample

The Input form is prefilled with 24 fictional, synthetic demonstration records. Click Start to see the report without scraping Google Maps, purchasing a dataset, or providing an Apify dataset ID. Change the input to your own legally obtained dataset for real analysis. Example data is not a real customer or source.

Input

Field	Description
records	JSON array of review objects.
recordsJson	The same, as pasted text.
sourceDatasetId	Optional dataset ID from your own Apify account (used only when no records are supplied).
minTopicMentions	Minimum mentions before a topic is reported (default 2).
includeMonthlyTrend	Include monthly rows (default true).

Recognized fields: placeId/cid, title, stars/rating, text, publishedAtDate/publishedAt/date (ISO strings or epoch seconds/milliseconds), responseFromOwnerText. Up to 100,000 reviews per run.

Output

Dataset rows by recordType: summary, business_insight, monthly_trend, issue_topic, and error (never charged). The OUTPUT record in the key-value store holds the run summary.

Pricing event

Pricing is not configured yet. The planned event name is maps-review-trends-report, charged once per successful report. Failed runs are not charged.

Local tests

python3 -m unittest discover -s tests

Multi-location manager review priority (new)

The default demo contains 24 synthetic reviews across two fictional locations: one store needs attention and the other performs well. The new manager queue shows the flagged store without sharing reviewer identity or text.

The Business insights Dataset view now shows a reviewPriorityTier, a reproducible reviewPriorityScore, and suggested manager follow-up actions. The summary includes an aggregated managerPriorityQueue sorted across locations, with the count of unanswered one- or two-star reviews. This helps an operations team decide which locations to examine and recheck in a later reporting cycle.

Scoring method: 50% weight on the share of 1-2 star reviews, 25% on the percentage of 1-2 star reviews without a recorded owner response, plus 20 points for a declining monthly rating trend and 5 points for recurring issues detected in low-rated review categories. Scores are capped at 100. Fewer than 10 reviews returns `insufficient_data` priority even if a numeric score is computed. Review response metadata can be absent from exports; interpret the queue as a candidate list rather than proof of actual non-response. Review text and reviewer identities are never exported.

Monthly paid-use workflow: Import a fresh dataset of authorized location reviews, open Business insights, triage high/medium priority locations, assign actions internally, then rerun after the next update to measure changes. Only one successful report event is charged; check the official Pricing tab. This Actor does not contact or respond to reviewers.