Why an audit finds reviews you do not have
An audit cannot see your reviews. It reads your text, looks for phrases that usually sit beside reviews, and infers. That works until a phrase means two things — and then it tells a business with no customers to mark up its customer ratings, and the business goes looking for something that is not there.
The sentence that said the opposite
A software foundation was told "Reviews are shown but not marked up". The evidence the check had collected was every use of the word "review" as an ordinary verb — please review the licence, participants are invited to review the guidelines, allow ten business days to complete the review.
And one more, which is the one worth remembering. The site's own sentence stating that it does not provide testimonials was counted as evidence that it does.
That is the failure in its purest form: counting a signal being mentioned as the signal being present. A page that discusses reviews, disclaims reviews, links to a review policy or explains why it has no reviews contains the word as often as a page covered in them.
Three more phrases that had to go
- "Stars." Fine on a restaurant, meaningless on a technical site, where every repository badge says it. The word was in the list; every developer tools page in the sample matched.
- "Five-star." A wildlife charity teaching children to build a bee hotel wrote that it offers solitary bees five-star accommodation. The check quoted that joke back as its evidence of displayed customer ratings. The phrase is a claim about quality — a five-star hotel, five-star service — and a business describing itself that way is not showing anybody's review.
- A street address that was a statistic. This one is ours. The address pattern was case-insensitive and allowed several words between the house number and the street suffix, and "way" is both a street suffix and an ordinary English word. It matched a sentence in this project's own marketing copy about domains linking to a competitor "that way", concluded that a Mac software company was a location-bound local business, and produced six findings including two false high-severity headlines. Any page with a statistic in it was one sentence away from the same fate.
Why the guards cut both ways
The repair to the address pattern was to require the street name to be capitalised and to allow at most two words between the number and the suffix. That stops statistics matching. It also means the pattern is now strict in ways that can miss real addresses — and one was found immediately: a house number carrying a single lowercase letter, the Baker Street shape, which the numeric-only version had been failing on all along.
So tightening a pattern is not free, and loosening it is not free either. What you choose depends entirely on which error costs the reader more.
The asymmetry that decides where the line goes
Here the two errors are not equal, and saying why is the most useful thing on this page.
- A missed review section produces "no customer reviews found". That is advice to go and collect some. A business that has reviews reads it, knows better, and ignores it. The cost is a wasted sentence.
- A false match produces "mark up the reviews you display" to a business that displays none. It cannot be ignored, because it sounds like a technical instruction: the reader goes looking for the review section the tool says they have, and there is nothing to find. The cost is their afternoon and their trust in the report.
So the phrase list is deliberately short and deliberately specific — named review platforms, "out of five stars", "star rating", "what our customers say" — and it deliberately misses a page simply headed "Reviews" with no other phrasing. That is the safer error, chosen knowingly.
A short phrase is cheap to add and expensive to be wrong about. The source notes that this is the sixth time a phrase has been removed for being a signal in one context and ordinary English in another, and the third time in that one list — which tells you how strong the pull is towards adding just one more word.
How to read a phrase-matched finding on your own site
Any finding that claims your pages show something — reviews, prices, testimonials, locations, opening hours — was almost certainly inferred from text rather than observed. Three questions settle it:
- Does the report quote its evidence? If it does, read the quote before the headline. Most of the cases above were obvious the moment somebody looked at the matched sentence.
- Is the matched word a verb? "Review", "rate", "star", "book", "order" and "contact" are all nouns that audits look for and verbs that people write.
- Is the page about the thing rather than doing the thing? A policy page about reviews, a help article about booking, a blog post about pricing — each one will match a phrase list and none of them is the signal.
When this does not matter
On a genuine local business — a restaurant, a clinic, a shop with an address and opening hours — these findings are usually right, because the phrases mean what they look like they mean. The failure mode is concentrated on sites that are not location-bound and on technical or editorial sites, where the same vocabulary is ordinary prose.
If your site has no premises the public visits, a local-business finding is worth a sceptical read before it is worth any work at all.
The fixes that make it worse
- Adding review markup to satisfy the finding. Structured data describing reviews you do not display is a misrepresentation of your page, and it is the kind that search engines act on.
- Rewriting your prose to avoid the trigger words. Removing "please review the licence" from a licence page to quieten a tool is letting the tool edit your site.
- Adding a business address to a site that has no premises because a local-business check fired. It fired because of a sentence about something else.
How to make this finding appear on any site
Write a sentence about reviews. That is the whole trick, and it is why the finding's presence is weak evidence on its own and its quoted evidence is strong evidence. A phrase list can tell you a page mentions something; only a person can tell you the page is doing it.
Where this sits in an audit
The registered checks are local.reviews, which covers review signals, and
local.applicable, which decides whether a site is a local business at all — the one
the address pattern feeds, and the one that was wrong about us. Both sit in the local business
area.
For the related habit of reading what a finding actually counted before acting on it, see a finding's count is not decoration →. For the opposite failure — a tool reporting the absence of something it simply could not recognise — see a form is a call to action →.
Common questions
Why does my audit say I show reviews when I do not?
Because it matched a phrase rather than seeing a review. Words like review, rate and star are nouns a tool looks for and verbs people write, so a page that discusses reviews matches as readily as one covered in them.
Should I add review structured data to clear the finding?
No. Markup describing reviews you do not display misrepresents the page, and that is the kind of misrepresentation search engines act on. Confirm the reviews exist first.
Why was my site classed as a local business when it has no premises?
Usually a street-address pattern matching ordinary prose. Several street suffixes are common English words, so a sentence containing a number and one of them can read as an address.
Is it better for a check to miss reviews or to invent them?
To miss them. A missed section produces advice to go and get reviews, which you can ignore. A false match sends you looking for a review section you do not have, which you cannot.
How do I tell a real phrase-matched finding from a false one?
Read the quoted evidence before the headline, and ask whether the page is about the thing or doing the thing. A policy page about reviews matches every phrase list and is not the signal.