How to Analyze Google Reviews Across Locations

Categorize what customers say, compare each location with its own performance a year earlier, and investigate categories that worsen for two consecutive months. Then compare the pattern across your network.

Request a demo Pricing

Fictional Cacao category matrix comparing complaint percentages and comment counts across locations

Start with categories to understand what is happening at each location

A star rating describes the overall experience. It does not tell corporate whether a location has a product, service, pricing, or coordination problem. Comparing ratings alone leaves those differences unexplained.

Categorization gives written feedback a common structure. AI can identify the aspects of the experience discussed in each review and the sentiment within each topic, allowing teams to compare the same categories across locations and over time.

The method has three parts: define useful categories, compare each location with its own history before comparing it with others, and inspect the comments behind sustained deterioration. Review volume and star ratings provide context throughout.

1. Organize categories by responsible area and subcategories by specific issue

Begin with categories that correspond to areas of responsibility, such as Product, Operations, and Marketing. Then use concrete subcategories to distinguish the experience described in the text.

For example, Product might contain food temperature and preparation; Operations might contain wait times, staff service, and coordination; Marketing might contain advertised offers and price communication. These are illustrative restaurant categories, not required defaults or a template for every industry.

Define what belongs in each category so the same type of comment is treated consistently across locations. Keep the structure stable while comparing periods. A renamed or redefined category may no longer describe the same set of issues.

Use categories to identify the responsible business area.
Use subcategories to separate specific issues within that area.
Apply the same definitions to every location included in a comparison.
Document changes to the structure before interpreting historical differences.

Fictional location detail showing a sample review with positive Product sentiment and negative Operations sentiment

2. Classify sentiment within each topic, not only for the whole review

A review can praise one part of the visit and criticize another. For example, “The food was excellent, but we waited 45 minutes and had to ask for the bill several times” contains positive feedback about Product and negative feedback about Operations.

The overall rating does not replace that reading. Four- and five-star reviews can contain complaints, while a low-rated review without text provides no written explanation to categorize.

AI helps organize this mixed feedback at scale. Keep the original wording available so the team can check whether the assigned topic and sentiment reflect what the customer actually said.

3. Read category percentages together with counts and review context

For a complaint category, divide the number of written comments mentioning a complaint in that category by all written comments at the location during the selected period. The denominator includes praise, complaints, and mixed comments.

For example, six Product complaints among 74 written comments is 8.1%. Each review counts once within a category but can appear in several categories, so the category percentages do not have to add up to 100%.

Read both the percentage and the count. More complaints may accompany a larger volume of feedback; a higher percentage based on very few comments may also need a closer reading before drawing a conclusion.

Keep this measure separate from the share of 1–3-star reviews, which uses all reviews in the period, including those without comments. The historical Google rating and its trend describe accumulated reputation; a date filter does not turn that rating into a period average. Cacao’s group rating gives each location’s historical rating equal weight.

Category complaint share: matching written comments divided by all written comments.
Low-rating share: 1–3-star reviews divided by all reviews in the period.
Historical rating: accumulated reputation, separate from the selected period’s feedback.

Date, location, and group filters used to select comparable periods for review analysis

4. Compare each location with the same months a year earlier

Start with an internal, year-over-year comparison: the same location, the same category definitions, and the same month in the previous year. This gives the branch its own reference before a network-wide ranking.

Comparing March with March a year earlier can provide more relevant seasonal context than comparing March with February alone. Still check for changes in opening hours, service format, review collection, and written-comment volume that could affect the comparison.

Cacao’s on-screen comparison of two manually selected periods covers total review volume and the share of 1–3-star reviews. To investigate category evolution, inspect the category results for each date range and read the supporting comments. Do not treat those as a separate automatic category-trend comparison.

If a location has no equivalent period from the previous year, state that limitation. Use the available history as provisional context rather than describing it as a year-over-year comparison.

5. Investigate categories that worsen for two consecutive months

A practical review trigger is deterioration in the same category for two consecutive months, with each month compared against the corresponding month a year earlier. This is a recommended investigation criterion, not a universal statistical threshold or an automatic Cacao alert rule.

Look for both a worsening complaint share and repeated descriptions of the same issue in the text. The percentage identifies a change in the feedback; the comments help establish what customers experienced.

Two months do not prove a cause. Check the written sample and the operating context before drawing conclusions. A serious individual complaint can still require attention immediately; the two-month criterion is for investigating sustained patterns.

Worked example: the same category worsens in two consecutive months

Fictional data for one location and its Operations category. Each month is compared with the same month a year earlier. The base is all written comments at the location.

Operations complaints: counts, written-comment sample, and year-over-year change
Month	Previous year	Current year	Change in complaint share
February	6 of 100 comments 6%	12 of 120 comments 10%	+4 percentage points
March	8 of 100 comments 8%	18 of 120 comments 15%	+7 percentage points

Both months show a higher Operations complaint share than the equivalent month a year earlier. The increase is not only a rise in the number of comments received: the proportion has increased too.

If the underlying comments repeatedly describe long waits and service coordination problems, investigate those subcategories with the responsible team. Check the sample and operating context before deciding what explains the deterioration.

Illustrative calculation, separate from the dashboard demo data. Category changes are investigated by selecting each date range; Cacao’s native period comparison covers total review volume and low-rating share.

Fictional location comparison showing review counts, historical ratings, low-rating share, and leading complaint topics

6. Compare the pattern with the rest of the locations

After the internal year-over-year check, compare the category across the other locations using the same current period, definitions, and written-comment denominator.

If the pattern is concentrated in one branch, investigate what differs locally. If several locations show a similar deterioration, examine whether they share a process, product, offer, or other operating context. The reviews identify questions to investigate rather than proving the root cause.

Use comparable formats and service conditions where possible. A small pickup-only location and a large dine-in branch may generate different feedback patterns. Keep those differences visible instead of interpreting a network ranking as a complete diagnosis.

Fictional complaints list preserving original review text, category labels, location, date, and star rating

7. Open the comments and periodically check the AI classification

Read the reviews behind the category before involving the responsible team. Separate repeated descriptions of a specific issue from comments about unrelated experiences, and check the assigned sentiment against the original wording.

Periodically review a sample of classified comments, including mixed opinions and results that appear surprising. Correct labels that do not match the text and revisit unclear category definitions when necessary.

In Cacao, customers can edit categories and subcategories and correct classification and sentiment. Editing a category does not automatically reclassify historical reviews, so account for that when comparing old and new results.

Check whether the selected category matches the customer’s wording.
Read praise and complaints independently within mixed reviews.
Review corrections before interpreting a change as operational deterioration.

8. Use AI summaries to interpret the qualitative feedback

Cacao can generate summaries that process the qualitative information in review text with AI. Use them to organize the themes customers discuss and the experiences described within a selected analysis.

Treat the summary as a reading aid alongside the category counts, sample size, and original comments. Review the evidence behind a theme before attributing a cause or deciding what the location should change.

Keep analysis connected to timely alerts and operational follow-up

A weekly review can identify changing patterns, while negative-review alerts help the configured responsible person notice individual reviews between scheduled checks. Analytical categories can describe several topics in one review; routing an alert selects one notification category.

Use the multi-location monitoring guide for the recurring review routine. For the team’s process of discussing an issue with the location, agreeing on a correction, and confirming resolution, see how to manage reviews across multiple locations. The analysis dashboard does not track operational closure.

Apply the method with Cacao’s dashboard and review analytics

Explore the five connected views in the Google reviews dashboard: group overview, location comparison, category matrix, location detail, and complaints. Move from a category to its matching reviews with the relevant filters preserved.

For product details, explore review analytics software for multi-location businesses. Cacao keeps Google reviews and Cacao survey responses as separate analytical sources. Customers create groups within an account; each location belongs to one group.

Cacao has analyzed more than 94,000 Google reviews across 676 locations for a single anonymized multi-location organization. This demonstrates analytical scale, rather than a guaranteed improvement in operational performance.

For sharing the findings, use the multi-location reporting guide. For an initial look at one location, the free Google Review Analyzer offers a separate starting point. Book a demo to explore the workflow across your network.

Request a demo

Understand which categories are changing at each location

Explore how Cacao connects AI categorization, location comparisons, qualitative summaries, and the reviews behind each result.

Request a demo Pricing

Google review analysis: frequently asked questions

How should I analyze Google reviews across locations?

Define common categories by responsible business area and concrete subcategories by issue. Use AI to classify topic sentiment, compare each location with the same months a year earlier, and then compare it with the rest of the network. Inspect the counts and original comments behind any deterioration.

Why is a Google star rating not enough to compare performance?

The rating summarizes the overall experience without distinguishing the topics involved. A review can praise Product and criticize Operations, including when its overall rating is four or five stars. Category analysis preserves those differences.

How do you calculate a category’s complaint share?

Divide written comments containing a complaint in the category by all written comments at that location during the period. A review counts once in each applicable category and can appear in several, so percentages across categories need not add up to 100%.

Should I compare a location with other locations or with its own history first?

Start with the same location and the same months a year earlier. Then compare the pattern across other locations using the same dates and category definitions, while considering differences in format, service, and sample size.

When does a worsening category deserve investigation?

Use two consecutive months of deterioration as a practical trigger, comparing each month against the same month a year earlier and checking repeated descriptions in the comments. This is a review criterion, not a universal statistical rule or an automatic alert setting. Serious individual complaints can require attention sooner.

Does Cacao automatically compare category trends between two periods?

The native two-period comparison covers total review volume and the share of 1–3-star reviews, using manually chosen periods. Investigate category changes by inspecting category results for each date range and reading the underlying comments.

Does the AI classification need human review?

Periodically check a sample of classifications and investigate unexpected results. Customers can correct categories and sentiment in Cacao. Editing a category definition does not automatically reclassify historical reviews.

What do AI summaries add to review analysis?

They process the qualitative information in review text into summaries that help organize customer feedback. Read them together with category counts, sample size, and original comments before drawing conclusions.