Reading a Platform Policy Before You File

The stranger test, and why your own file fails it

A removal request that argues the content is false is answering a question no operator asked. A request naming the specific policy line the content breaks is answering the only question the person reading it is empowered to decide. Everything else here follows from that swap, and the reason for it is structural rather than unlucky. One operator has written it down:

“We generally do not consider evidence offered by someone with a vested interest in removing a review, as its reliability cannot be verified.”
— Glassdoor Help Center, “I'm an employer. What can I do about negative reviews on Glassdoor?”, updated 6 April 2026, read 12 August 2026

Two more sentences sit on the same page: no side is taken in factual or contractual disputes between employers and reviewers, and final decisions rest in the operator's sole discretion, not open to negotiation. Read all three together and a whole category of effort disappears. The dossier is not ignored out of laziness. It is set aside because the side wanting the content gone is the side that assembled it, and nobody at the other end can test that.

So apply a stranger test to every element of a request before it goes. Could someone who has never met you, has no access to your systems, and takes nothing you say on trust confirm this from the content and the platform alone? If yes, it belongs in the request. If no, it belongs in your file for other purposes and nowhere near the report button.

Whether a false statement about your company is legally actionable is a separate question and one for a defamation attorney. What follows is about the rule and the content.

Four documents, and only one of them binds

Every operator publishes several things that look like policy, and businesses argue from the wrong one constantly. Sort them first.

The rule set. Usually called community guidelines or content guidelines. This is the document a moderator actually applies. It is the only one worth quoting in a report.
The help articles. Business-facing pages explaining the process, what the operator will and will not do, and how to flag. Useful for understanding the route and the posture, and often more candid than the rule set, but written as guidance rather than as enforceable lines.
The terms of use. A contract between the operator and anyone with an account, governing venue, arbitration and what you gave up by signing up. Nothing to do with whether a post comes down; a great deal to do with what happens if you escalate.
The trust or transparency pages. Enforcement volumes, alert programs, published statistics. Occasionally the most useful pages of all, because that is where an operator describes what it does to businesses as well as for them.

One more sorting step. Operators describe the same rule differently to a reviewer and to an employer, and the employer-facing page usually holds the sentences a business needs most. Find both.

Telling an enforceable line from a statement of values

Inside the rule set, sentences fall into two kinds, and only one kind can carry a report.

A statement of values describes the operator's stance: welcoming positive and negative content alike, never editing or altering what users submit, applying one standard to client and non-client content. Those sentences are worth reading closely — they tell you what will never happen — but nothing can be filed against them.

An enforceable line describes a property of the content. It has an object you can point at: a count, a date range, a category of person, a format, a source. A cap of one review per employer in a year for each type of review. A requirement that the job or interview fall inside a five-year window. A ban on screenshots.

The test is the stranger test in a different coat. If a sentence could be checked by someone looking only at the content, it is a line you can file against. If verifying it requires knowing what happened between you and the poster, it is not. Which is why the most quotable sentences in these documents — the ones about neutrality and fairness — are the least useful operationally, and the tedious ones about counts and formats are where the work is.

Six questions that find the rule, on any platform

Rather than reading each operator's rules end to end every time, read them against a fixed set of questions. The same six categories turn up in almost every policy in this field, and they are the categories that decide themselves.

Was this person allowed to post at all? Eligibility rules are the strongest ground in the field, because they are checkable without reference to the dispute. Amazon requires a reviewer to have spent fifty dollars on Amazon.com using a credit or debit card in the past twelve months before creating reviews at all, and states that promotional discounts do not count toward it. The Yelp Content Guidelines confine a review to the writer's own visit — “not what you heard from your partner or co-worker, or what you saw in the news or on social media” — and shut out peers and competitors in the same industry.
How many times, and how recently? Frequency and recency limits catch the campaign that a falsity argument never reaches. Glassdoor enforces one review per employer per year per review type and removes content where it finds evidence that policy was abused.
Was it posted in the right place? Channel rules are heavily underused. Yelp requires the review to sit on the page for the location actually visited, and treats a rant about a business's employment practices as off topic rather than as a consumer experience. Amazon excludes content that only addresses the seller, the shipping, the packaging or an ordering problem from product reviews, because that belongs in seller feedback.
Who does it name? Identity rules are the sharpest lever most employers have. The Glassdoor Community Guidelines permit naming individuals in the highest positions in a company — the public face of the business — and then state that negative comments about identifiable individuals outside that group are not allowed.
What does it physically contain? Format rules are decided by looking. Screenshots, contact details, portraits, logos, substantial quoted material from elsewhere, proprietary information such as customer lists, budgets or detailed financial results.
How was it produced? Provenance rules have expanded fastest. Yelp prohibits using third-party AI tools and chatbots to create content, “including using such tools to draft or revise content” — which reaches a genuine complaint tidied up by a chatbot, not only a fabricated one. Amazon bars a review posted before delivery and treats compensation for creating, editing or removing a review as a violation. On Glassdoor, evidence that a poster was incentivized or coerced is itself a ground for removal.

Run a piece of content through those six and you will usually know within minutes whether there is anything to file. That is a better use of an afternoon than assembling a dossier nobody will weigh.

The rule that catches your own campaign

Two of those six point back at the business, and are worth checking your own conduct against first.

The incentive rules run in both directions. Glassdoor states that it will remove positive reviews where there is evidence employees were compensated or coerced into leaving them, and that employers may not ask workers to show confirmation of posting. Yelp goes further than most and tells businesses not to ask customers for reviews at all, which makes an ordinary campaign from another platform a rule breach on that one.

The provenance rules do the same. If a marketing team drafted or polished text for customers to post, that content sits inside the prohibition being used against the attacker.

None of this is a reason not to file. It is a reason to know your own history first.

When the rule you need does not exist

Some of the time the honest outcome of this exercise is that nothing fits, and the most useful example is the largest platform of all.

Meta's Community Standards run to roughly twenty-five policies covering harm, fraud, harassment, privacy, impersonation, intellectual property, account integrity and inauthentic behavior. None is about false statements. The routes that reach a business are harassment, privacy, impersonation, intellectual property and account integrity — every one asking a question other than whether the content is accurate.

The harassment route in particular has an explicit hole in it. In the tier protecting private adults, Meta permits “negative character or ability claims, except in the context of criminal allegations and business reviews against adults.” Business reviews are written out of that protection by name, so a post saying a company cheats its customers is not a bullying violation and reporting it as one will fail.

Reading that before filing saves a wasted report and redirects the effort. Where no content rule fits, the question worth asking is whether the accounts break a rule even though the posts do not — a different inventory, built from timing and behavior rather than words.

Read the date at the top of the page

These documents are versioned, they move quietly, and a business acting on a policy quotation from an article written two years ago is arguing from a rule that may no longer exist.

The revision dates on a single operator's pages make the point. Reading Glassdoor's policy set on 12 August 2026, the community guidelines carried an update date of 15 April 2025, the employer guidance 6 April 2026, the paid-services answer 8 December 2025, the legal-service page 1 July 2026, and the appeals article 7 August 2026 — five days before I read it. That is one operator's rules living at five different dates.

Two habits follow. Quote the operator's page as it reads on the day you file, not a summary. And keep a dated copy of the policy page you relied on alongside a dated copy of the content, because if the rule is edited later there is no way to show what it said at the time.

Which applies here as much as anywhere. What is reproduced on this page is the wording that stood on 12 August 2026, dated with the operator's own revision. Go to the source before relying on it.

Three lines and a link

Once the rule is identified, the report is short. Longer reports are not stronger, for a reason beyond anyone's patience: every sentence that requires trust dilutes the sentences that do not.

The shape that works is the rule as the operator worded it, the part of the content that breaks it quoted exactly, and the address. Then stop. No history of the relationship, no account of the damage, no theory about the poster's motives, nothing about revenue or lawyers.

Two judgment calls sit inside that. The first is how many grounds to raise: where several rules fit, lead with the one requiring the least inference, because one clear breach and three arguable ones reads as an argument rather than a citation. The second is the identifiable-person question — where a review describes someone by role, initials or a distinguishing detail rather than by name, say so, because a moderator outside the company cannot infer who is meant.

And preserve before filing. A successful report deletes the evidence; an unsuccessful one can prompt the author to edit or delete before anyone holds a dated copy. Capture, then read the rule, then file. That order is the only one that survives either outcome.

What an appeal is, and what it is not

An appeal is a request that the same operator look again under the same rules. It is not independent review, and it does not open a negotiation.

“We're open to reviewing appealed decisions, but content moderation is ultimately at Glassdoor's sole discretion, and outcomes aren't negotiable. Glassdoor reserves the right to remove community content at any time, for any reason. Repeated or abusive appeal requests may result in permanent restriction.”
— Glassdoor Help Center, “How to Appeal a Removed Community Post”, updated 7 August 2026, read 12 August 2026

The same page describes enforcement running on a three-strike system. Google publishes a one-time appeal per review with a final decision by email — one attempt, spent whether or not it was prepared. On Yelp, no named appeal process appeared on any policy page I read on 12 August 2026.

That is the argument for doing the reading first. No process here rewards persistence, and in at least one case persistence is penalized. If a report is refused, the productive move is to go back to the six questions and find a different line, not to resend the same one with more emphasis.

When nothing in the rule set fits

Say it plainly, because most of the market will not. A review from someone entitled to post, on the right page, about a real experience, in permitted language, naming nobody protected, produced without incentive or a chatbot, is content the platform is built to keep. No reading of the policy removes it, and no amount of money changes that.

What is left is not nothing. A public response is read by the next customer rather than the poster. A dated record of what exists, and when, is worth holding whether or not it is used. And what a search for the company name returns can change even when the page does not.

That is a smaller claim than removal. On most days it is the true one, and a business that hears it early stops paying for the version that is not.

Frequently Asked Questions

What is the best reason to give when reporting a review?

The rule the content actually breaks, quoted from the operator's own page, with the offending line quoted back and the address included. The strongest grounds are the ones a stranger can check without trusting you: the poster was not eligible to post, the review sits on the wrong page, it names someone the rules protect, it contains a screenshot or contact details, it is the second one this year, or it was produced with an incentive or an AI tool. Falsity is not a category on any platform in this cluster.

Should I send proof that a review is false?

It will usually be set aside, and one operator explains why in its own words. Glassdoor's employer guidance says evidence from a party with a vested interest in removal is generally not considered, on the basis that its reliability cannot be verified. Payroll extracts, personnel files, shift records and manager statements all arrive from the side that wants the content gone, and a moderator has no means of testing any of them. Send what is verifiable from the content and the platform itself, and keep the rest of the file for other purposes.

Where do I find the rules a platform actually enforces?

In the community or content guidelines, not in the marketing pages, the trust reports, or the terms of use. The guidelines are what a moderator applies. Help articles written for businesses are worth reading alongside them, because that is where operators state most plainly what they will not do. Terms of use govern your account and any escalation rather than whether a post comes down. Note the revision date on each page, because different pages from the same operator often carry very different dates.

Can I report the same review under several rules at once?

You can, and sometimes you should, but lead with the ground that requires the least inference. A report resting on one plain breach plus three arguable ones reads as an argument rather than a citation, and the arguable grounds dilute the plain one. If a review clearly names a protected individual and also might be from a competitor, the naming rule is the report and the rest is noise. Where a second ground is equally plain, state it in a separate line rather than blending the two.

How do I know whether a policy has changed since I read it?

Check the operator's own page and the revision date printed on it every time before filing. These documents move quietly and they move often. Reading one operator's policy set on 12 August 2026, five related pages carried five different update dates spread across sixteen months, the most recent revised five days earlier. Keep a dated copy of the page you relied on when you filed, because if the wording is edited afterward there is otherwise no way to show what the rule said at the time.

The review names a manager but does not use her name. Does that count?

On the operator whose rule turns on identifiable individuals, describing someone by role, initials, or a distinguishing detail can put the content inside the same prohibition as naming them. The important part is that a moderator outside your company cannot work that out unaided. If the review refers to the afternoon shift supervisor and there is one, say so in the report and explain briefly how the description identifies a specific person. Do not assume the inference will be made for you.

There is no rule that fits my situation. What now?

Stop filing. A review from an eligible poster, on the correct page, about a real experience, in permitted language, naming nobody protected, is content the platform is designed to keep, and repeat reports on a ground that does not fit achieve nothing and can attract restriction. The work that remains is different in kind: a response written for the next reader rather than for the poster, a dated record of what exists, and changing what a search for the company name returns.

Published 2026-08-12