Quick answer
A SERP API can expose ratings or review snippets in a search result. A Reader can extract visible page text. Neither output, by itself, establishes a complete, paginated collection of individual reviews. Choose a dedicated review-data contract when individual records and coverage matter.
This guide owns that data boundary. It does not claim that SearchCans provides a complete Google, Amazon or marketplace review database. For general source retrieval, use the Reader API reference.
Choose the operation that matches the evidence
| Need | Suitable evidence contract | Important limit |
|---|---|---|
| Discover pages discussing a business or product | Search results with source URLs and snippets | Snippet presence is not a complete review record |
| Read selected public pages | Accepted page extraction with source identity and time | Visible text does not establish hidden, paginated or complete coverage |
| Collect individual Google Maps review records | A documented review endpoint and its pagination contract | Availability and returned fields must be checked for the particular request |
| Manage reviews for authorized Business Profile locations | Google Business Profile API with required authorization | An authorized account workflow is different from arbitrary public review collection |
SerpApi documents a Google Maps Reviews API with identifiers and pagination controls. Google documents authorized Business Profile review operations and Place Details fields. These are distinct contracts. Check their current access, field and usage requirements before selecting one; this page does not give legal permission or infer complete population coverage.
Normalize records without inventing missing values
The following input is an application-owned normalized shape, not the raw response schema of SearchCans, Google or SerpApi. Map each actual provider response explicitly. Use a provider-scoped stable review ID. A search snippet without one remains a reference, not a review.
import math
from datetime import datetime
from urllib.parse import urlsplit
def review_ledger(records):
if not isinstance(records, list):
raise ValueError("Expected normalized application records")
accepted, quarantine, seen = [], [], set()
duplicates = 0
for r in records:
if not isinstance(r, dict) or r.get("kind") != "review":
quarantine.append("not_a_review_record")
continue
provider, ident = r.get("provider"), r.get("review_id")
if any(not isinstance(v, str) or not v.strip() for v in (provider, ident)):
quarantine.append("missing_stable_identity")
continue
key = (provider, ident)
if key in seen:
duplicates += 1
continue
try:
link = r.get("source_url")
if not isinstance(link, str) or any(c.isspace() for c in link):
raise ValueError("Invalid source")
parsed = urlsplit(link)
host, port = parsed.hostname, parsed.port
if (parsed.scheme not in ("http", "https") or not host
or parsed.username is not None or parsed.password is not None):
raise ValueError("Invalid source")
rating = r.get("rating_1_to_5")
if rating is not None and (isinstance(rating, bool)
or not isinstance(rating, (int, float))
or not math.isfinite(rating) or not 1 <= rating <= 5):
raise ValueError("Invalid normalized rating")
date = r.get("published_at")
if date is not None:
if not isinstance(date, str):
raise ValueError("Invalid date")
datetime.fromisoformat(date.replace("Z", "+00:00"))
if not isinstance(r.get("text", ""), str):
raise ValueError("Invalid review text")
except (ValueError, TypeError):
quarantine.append("invalid_normalized_record")
continue
seen.add(key)
accepted.append({"provider": provider, "review_id": ident,
"source_url": link, "rating_1_to_5": rating,
"published_at": date, "text": r.get("text", "")})
return {"records": accepted, "duplicates_skipped": duplicates,
"quarantine_reasons": quarantine, "population_coverage": "unknown"}
if __name__ == "__main__":
# Application-owned normalized fixture, not a provider response schema.
sample = {"kind": "review", "provider": "example", "review_id": "r1",
"source_url": "https://example.com/review/r1", "text": "Example",
"rating_1_to_5": None, "published_at": None}
print(review_ledger([sample, sample, {"kind": "search_snippet"}]))
The synthetic fixture keeps one record, skips its duplicate and quarantines a snippet. Missing rating and publication date stay null. The normalized rating field uses an explicit 1–5 scale; convert another provider scale only with a documented method. Do not replace missing ratings with zero or infer a publication date from retrieval time.
Keep pagination and population claims separate
- Save each page request, cursor and response state. Deduplicating the records you received does not prove that every review was collected.
- Record the provider’s identifiers and actual source URL. Do not deduplicate people merely because display names match.
- Retain errors, inaccessible pages and unsupported fields. An empty result can have more than one cause.
- A mean calculated from returned records describes that sample. It does not automatically represent all reviewers or the entire product population.
Use search and reading for their narrower jobs
SearchCans Google Search API can help discover relevant pages. Reader can help inspect selected content. Those steps are useful for qualitative source research, but they do not replace a review endpoint’s record identity, pagination and coverage contract.
The schema worksheet, deduplication logic and invalid-input fixtures were checked locally on November 1, 2026. No live completeness experiment, sentiment-accuracy score or comparison of provider coverage is reported. Treat third-party review text as untrusted input and preserve the source needed to inspect it.