Chapter 08
Learning from reviews
Every review is a customer telling you, unprompted, what worked and what didn't. Read one at a time and they feel like opinions. Read together, they're the most honest operations report you'll get.
The short version
Read reviews together, not one by one
Key takeaways
- Count themes, not stars. Group what reviews describe (waits, prices, cleanliness, communication) and see which come up most, praised and criticized.
- Compare periods. Each month next to the one before shows what's getting better, what's slipping and whether a fix worked.
- Learn about the work, not individuals. Fix steps and processes; never mine staff names from reviews or rank employees by them.
The evidence
Reviews are operations data
Reviewers describe specifics. They write that the callback took two days, that the price matched the quote, that the waiting room was spotless. Those details are exactly what an operations report tries to capture, and customers hand them over for free.
Researchers have shown how much is in that text. A 2013 study trained a model on Yelp reviews of restaurants and matched it against hygiene inspection records: from the review text alone, it told severe offenders apart from restaurants with no violations with over 82% accuracy.1 Customers noticed what inspectors later found.
It matters for revenue too. A Harvard Business School working paper on restaurants found that a one-star increase in Yelp rating led to a 5 to 9 percent increase in revenue for independent restaurants, while chains saw no effect.2 It's one study of one industry, but the direction is clear: what reviews say about your operation feeds straight back into it.
Your customers already read reviews this way. Google now uses Gemini to summarize the top themes in a place's reviews,5 and in BrightLocal's 2026 survey, 82% of consumers said they read AI-generated review summaries.3 If a theme is obvious to them, it should be obvious to you first.
Themes
Sort reviews into themes
A theme is anything customers describe that you control: timeliness, price and value, cleanliness, communication, booking, the result of the work. Tag each review with the themes it mentions, and whether it praises or criticizes each one. A single review often does both: "great result, but nobody told me about the delay".
Then count. The chart below is the picture you're after: what's praised most, what's criticized most, and the themes that draw both.
Reviews that mention each theme, last 90 days
Three habits keep the counts honest. Use the same short list of themes every month, so periods compare. Count reviews, not mentions, so one long review can't outweigh five short ones. And read the reviews behind any number before acting on it: a theme is a pointer to the text, not a substitute for it.
Ratings alone hide most of this. A 4-star review that says "everything was great except the wait" carries a complaint you'd miss if you only watched the average.
Trends
Watch the trend, not the last review
One bad review on a Tuesday tells you about a Tuesday. A rating that drifts down for three months tells you something changed. The useful view is your recent reviews month by month, next to the period before.
It's also the view customers take. In BrightLocal's 2026 survey, 74% of consumers said they look for reviews written in the last three months, and 32% look for reviews from the last two weeks.3 Your all-time average is the number on your profile; your last quarter is the experience people are reading about. A business's recent reviews can sit well above or below its all-time average, so watch both.
Average rating of new reviews, by month
Read trends with care. Compare periods of the same length. Ignore a month with only a handful of reviews. And when the rating moves, look at which themes moved with it: that's the "why" behind the number.
Movers
What's getting better, and what's getting worse
Put the theme counts for this period next to the last one and two lists fall out: the themes customers praise more than before, and the ones they criticize more. The second list is your next fix. The first tells you which changes worked, so you keep them.
Getting better
- Booking praised 12, up from 5
- Cleanliness praised 38, up from 31
- Value criticized 5, down from 9
Getting worse
- Wait times criticized 9, up from 3
- Parking criticized 4, up from 1
- Customer service praised 30, down from 34
Small numbers move a lot, so "4, up from 1" is a lead to read, not a crisis. Wait times tripling over a quarter with a few dozen reviews is a pattern. Pick one item from the "worse" list, change something specific, and check the same theme next month.
Tell the team what changed and why, in the customers' words. "Nine reviews this quarter mention waiting past the appointment time" lands harder than "we need to be more punctual".
Staff
Learn about the work, not about individuals
Reviews sometimes name an employee, for praise or for blame. It's tempting to count those names and build a scoreboard. Don't. Names show up in a small, uneven slice of reviews, so a count measures who has a memorable name or a customer-facing role, not who does good work. And ranking or rating people from reviews gives staff a reason to ask only the customers they expect to be happy, which is exactly what the FTC tells businesses not to do.4
Ask what the review says about the process instead. "Nobody explained the bill" is about how bills get explained, whoever was at the desk. Fix the step, and every shift gets better.
Don't
- Pull staff names out of review text
- Rank employees by reviews
- Give a person a star rating
- Reward staff for 5-star reviews
Handle with care
- A complaint about one person: deal with it privately, as you would any complaint
- Praise that names someone: pass it on, without a tally
Do
- Tag reviews by step, service, shift or location
- Count how many review requests each desk or device showed
- Share themes with the whole team
Competitors
Compare yourself with competitors
Your reviews tell you how you're doing. Your competitors' reviews tell you what customers in your field care about, and where the bar is. Read the reviews of the three or four businesses customers most often choose instead of you: what they're praised for that you aren't, and what they're criticized for that you could get right.
Then compare on the numbers people see when they compare you: overall rating, total reviews, and how many reviews came in recently. A competitor with a lower rating but three times as many recent reviews can look busier and more current to someone deciding today.
| Measure | You | Competitor A | Competitor B |
|---|---|---|---|
| Average rating | 4.6 | 4.8 | 4.3 |
| Total reviews | 412 | 380 | 205 |
| Reviews, last 12 months | 140 | 96 | 61 |
| Praised most for | Wood-fired oven | Fast delivery | Price |
| Criticized most for | Wait for a table | Cold food on long trips | Small portions |
Your real competitors aren't always the businesses next door. They're whoever customers are offered instead of you, and more and more, that's whoever an AI assistant suggests. Chapter 09 covers how assistants pick.
One line not to cross: never post, buy or arrange reviews about a competitor.
Routine
A monthly review routine
One hour, once a month
- 1Read every new 1 and 2-star reviewBefore any chart. The text tells you what the numbers can't.15 minutes
- 2Update the theme countsSame theme list as last month, praised and criticized.10 minutes
- 3Compare with the period beforeRating, review count and each theme. Note what moved.10 minutes
- 4Pick one fixThe theme that got worse and that you control. Write down the change.5 minutes
- 5Check last month's fixDid that theme improve? Keep it, or try something else.5 minutes
- 6Look at competitorsWho's new, who moved ahead, what they're praised for.10 minutes
- 7Share it with the teamThemes and quotes, no names, and the one change you're making.5 minutes
Grades.com
Know what to fix next
Grades.com collects your reviews from every site you connect and does the counting for you.
- Review Analytics: what customers praise and criticize by topic, your rating month by month next to the period before, topic scores next to your category's median, and every location side by side.
- Competitor Map: the businesses AI assistants suggest instead of you and the ones nearby, compared with you on grade, rating, total reviews and reviews in the last 12 months.
- Answers in plain words: ask Grades.com AI, or Claude and ChatGPT through the Reviews MCP, what your timeliness reviews have in common.
Topics come from what reviews describe; for healthcare practices, topics about conditions or care are left out.
14-day free trial of Pro, no card needed.
Frequently asked questions
How many reviews do I need before the themes mean anything?
Enough that one review can't swing the picture. Treat a theme that appears in two or three reviews as a lead to check, not a finding, and compare periods of the same length. With few reviews, read every one; counting starts to help once you have dozens a quarter.
Should I track which employees are mentioned in reviews?
No. Names in review text are rare and uneven, so they make a poor measure, and ranking or rating people from reviews pushes staff to ask only happy customers, which the FTC warns against. Learn about the work instead: the step, the shift or the process the review describes.
How often should I look at my review trends?
Read new negative reviews as they arrive, and look at themes and ratings once a month, each month next to the month or quarter before. A monthly rhythm is long enough to see a change and short enough to act on it.
Is it fair to read competitors' reviews?
Yes. Reviews are public so that people can compare businesses, and reading them tells you what customers in your field value and where others fall short. Never post, buy or arrange reviews about a competitor.
Do higher ratings really bring in more business?
There's evidence they do. A Harvard Business School working paper on restaurants found that a one-star increase in Yelp rating led to a 5 to 9 percent increase in revenue for independent restaurants, with no effect for chains. It's one study of one industry, so treat it as a direction, not a forecast.
Notes
Sources
- Kang, J. S., Kuznetsova, P., Luca, M. and Choi, Y. "Where Not to Eat? Improving Public Policy by Predicting Hygiene Inspections Using Online Reviews." Proceedings of EMNLP 2013, Association for Computational Linguistics, October 2013. Paper
- Luca, M. "Reviews, Reputation, and Revenue: The Case of Yelp.com." Harvard Business School Working Paper 12-016, September 2011, revised March 2016. A working paper, not peer-reviewed. Record
- BrightLocal consumer review survey, 2026 (published February 11, 2026): 1,002 US adults on a representative SurveyMonkey panel. Survey
- FTC, "Soliciting and Paying for Online Reviews: A Guide for Marketers", January 2022: "Don't ask for reviews only from customers you think will leave positive ones." FTC
- Google Maps Platform blog, "Discover more, faster: AI-powered summaries for places, areas, and reviews are now generally available", September 23, 2025. Google