What RatingSentinel reads, computes and keeps.
The full technical picture — what we connect to, what we calculate, what we store and for how long.
What the product does
RatingSentinel connects to the Google Business Profile locations a customer authorizes and analyzes the reviews those locations have received. The main output is aspect-level sentiment: what customers actually say, broken out by theme, compared across locations and tracked over time.
Alongside it we compute a Health Score for each location. That score describes how closely a location's review pattern matches what its own history would predict. It is an internal measure, and the section further down this page sets out what it is for.
The Health Score describes patterns, not verdicts. Whether any individual review belongs on a profile is a question for the platform that hosts it, and we do not attempt to answer it.
RatingSentinel is currently an analysis-only product. It does not post, edit or delete anything on a customer's Business Profile.
The connection to Google Business Profile
- The customer authorizes us on Google's own consent screen. We never see or store a Google password.
- We analyze at ingest. Health signals and aspect sentiment are computed as the data arrives.
- We keep the findings — scores, signal strengths, topic counts, review counts over time, and the Google-assigned identifiers needed to fetch a review again.
- What we fetch is cached briefly, then deleted — no more than 30 calendar days from fetch, automatically, on a rolling basis. The cache exists so the dashboard stays fast. Without it every page load would have to re-read from Google, which is slower and leaves the dashboard waiting on an external service while someone is looking at it. We do not build a permanent archive of it.
- The customer can disconnect at any time. We call Google's token revocation endpoint and delete their data.
The permission we request
| Scope | What it powers | Why nothing narrower |
|---|---|---|
| https://www.googleapis.com/ |
Reading the locations a customer manages and the reviews on each — the entire input to the Health Score and the sentiment analysis. | The Business Profile APIs publish no read-only scope. This is the only supported permission that exposes review content, so it is the minimum available. It carries write capability; we currently use it for read operations only. |
Limited Use
RatingSentinel's use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements. Data obtained through the Business Profile APIs is used solely to provide the analysis features visible in the customer's own dashboard, and is not transferred to others except as required for security, to comply with applicable law, or as part of a merger or acquisition with the customer's consent.
What RatingSentinel does not do
- Does not request the removal of any review, dispute a review, or appeal a rating, to Google or to any other platform
- Does not contact reviewers, and does not help a customer contact them
- Does not currently write to a Business Profile in any way — no replies, edits, posts or deletions
- Does not retain review text beyond the caching window described above
- Does not sell, rent or share customer data with third parties for their own purposes
- Does not use one customer's data to serve another, or pool it into a shared model
- Does not attempt to identify reviewers
- Does not collect reviews from anywhere outside the profiles a customer authorizes
The Health Score
A 0–100 measure per location of how closely its review pattern matches what that location's own history would predict, combining several independent signals. A lower score means the pattern is unusual, not that anything is wrong.
Why this score exists
Before a customer acts on a change in sentiment, they want to know the reviews behind it arrived the way reviews normally arrive, rather than in a single unusual burst that would distort the trend.
For example, a business running twenty locations wants to know if one of them ever starts soliciting reviews in a way that breaches Google's policies, so that it can be stopped. They would also want to be aware if many negative reviews were arriving in an unnatural pattern.
The score is a control a customer runs on their own data. It is not a judgment about any review. It is never sent to Google or to a reviewer. Whether a review belongs on a profile is Google's decision, not ours.
| Signal | What it looks for | Status |
|---|---|---|
| Temporal burst | Reviews arriving far faster than a location's own established baseline, measured against its history rather than a global threshold. | Phase 1 |
| Rating distribution | Implausible shapes — a heavy five-star concentration with no populated middle is the shape a concentrated solicitation push tends to leave. | Phase 1 |
| Vague enthusiasm | A significant number of strongly positive reviews carrying no specific, verifiable detail, measured by named-entity density and concrete-noun ratio rather than length. | Phase 1 |
| Text similarity | Near-duplicate or templated review text clustered across reviews, reviewers or locations, using sentence embeddings. | Phase 2 |
| Generated-text likelihood | Text showing patterns consistent with machine generation. Weighted lightly and never treated as a verdict. | Phase 2 |
What the score is not
It is an analytical estimate derived from patterns in publicly posted reviews. It is probabilistic: it is not an accusation against any reviewer, it is not evidence of fraud, and it is not legal advice. Every review that goes into a score is visible in the dashboard, so a customer can always check what a score was built from. The weighting behind the calculation is available on request.
A low Health Score is also not a finding against the platform the reviews were posted on. Platforms hold data we do not, such as IP addresses, and may reach different conclusions about the same pattern.
Sentiment analysis
One score for a whole review throws away the part a manager can act on. "The hygienist was wonderful but billing took three calls to sort out" is praise for the care and a complaint about billing, and those belong in separate columns.
So we read each review for the specific topics it raises and how positive or negative it is about each one. For a dental practice that might be wait times, the front desk, billing and insurance, the care itself, pain management and the premises. The results are totalled for each location and tracked over 30, 60 and 90 days, so a complaint that is becoming more common shows up as a trend rather than a one-off.
Review text sent for this analysis has reviewer names, location names and the customer's identity removed first. The provider that performs it is named in our sub-processor list. Under that provider's published terms, data sent through its API is not used to train its models and is kept only for a limited period.