ajestIQ
Review Patterns7 min read

Review Text vs. Review Patterns: Two Kinds of Evidence

Read one looks at what a single review says. Read two looks at what all the reviews show together. Blend the two and you get accusations, not evidence.

August 14, 2026Updated November 16, 20267 min read
Discuss This Article
In Brief
  • Text evidence sits in the review itself and can be quoted.
  • Pattern evidence only shows when all the reviews are read together: timing, phrasing and review history.
  • One signal is not a pattern. A signal on its own is recorded as not proven and the review is left alone.
  • Every piece of evidence is public review content or public review history.
  • Keeping the two apart is what stops an observation from turning into an unsupported accusation.

We have filed review cases with Amazon since 2019. The most useful thing we learned in that time fits in one sentence: the review in front of you is only half the evidence.

Most sellers know one way to deal with a bad review. Read it, find the rule it breaks, report it. That works, and it is the larger share of what we do. It also has a blind spot. Some negative reviews break no rule in their text and still are not what they look like. You only see it when you stop reading reviews one at a time.

So we read every written negative review twice, and we keep the two reads apart. This article explains what each read looks at, where each one runs out, and why I think mixing them is the most common mistake in this work.

Read one: the review on its own

Read one asks one narrow thing: whether this review, on its own, breaks one of Amazon's rules.

The evidence is the text. Three models read each written negative review against Amazon's Community Guidelines. They look for three kinds of problem:

  • Feedback that is not about the product. Shipping, packaging, the seller, a refund, or Amazon itself. Amazon keeps this out of product reviews.
  • Restricted content. Private information, profanity, threats, hate speech and explicit content.
  • Promotion. Links to outside sites, discount codes, and reviews that send shoppers to another brand or seller.

Our working list has 29 read one conditions across those three groups. What Amazon Review Removal Actually Covers goes through them in detail.

Read one evidence is direct. You can quote it. The sentence that breaks the rule is on the page, and anyone at Amazon can check it in seconds. That is its strength.

Its weakness is that it sees one review at a time. A review that says "Hard pass. Wouldn't gift this to an enemy" breaks nothing. Read one leaves it alone, and it is right to.

Read two: all the reviews together

Read two asks something else: whether anything shows up, once every review on the listing sits side by side, that no single review shows.

It looks at three signals.

Timing

When the reviews arrived, against the listing's normal pace. A burst of negative reviews in a few days stands out on a timeline. It also has plenty of ordinary causes. A big sale, a launch, a bad production batch or a news story can each produce one. Timing on its own is never grounds.

Phrasing

Unusual wording that repeats across reviewers. Everyday wording counts for nothing here. "Would not recommend" appears on a huge number of honest reviews. The signal is the specific, odd phrase that keeps coming back from reviewers who have no visible reason to share it.

Review history

What else the reviewers have reviewed. Every Amazon reviewer has a public profile, and that profile shows their public review history. When the same reviewers keep turning up together on the same unrelated products, that overlap is a signal. Aletheia is our name for the mapping of that overlap. It is one part of read two, not the whole of it.

Amazon looks at related data from the inside. In its own description of how it protects reviews, Amazon says its models analyze "ties to other accounts," "risky behavioral patterns" and "review history," and that it uses graph models to "analyze and understand complex relationships and patterns." The difference is access. Amazon sees its private data. We see only what is public, which is why what we find goes to Amazon as evidence for it to investigate, not as a verdict.

Our working list has 7 read two signals.

A signal is not a pattern

This is the rule the rest depends on. Each signal on its own has an innocent explanation, so each one on its own is recorded as not proven, and the review is left alone.

A pattern is different. When several signals agree on the same group of reviews (a burst whose reviews share unusual phrasing, from reviewers who keep appearing together elsewhere), together they support a pattern of coordinated reviewer activity. That pattern gives grounds for a case.

We do not publish thresholds. A single number would be wrong for a lot of listings, and publishing one would tell the people behind coordinated reviews exactly where to stop.

The illustrative sample on our home page shows how this plays out. It is a set of invented reviews of an invented product. Read one files on a few and leaves the rest. Read two takes them together, finds a burst, finds reviews sharing the same odd phrases, finds their reviewers linked by overlapping review history, and moves those reviews into a pattern. Other reviews carry a single signal each, and they stay exactly where they were.

Observation, inference and proof

A defensible case keeps three layers separate.

  1. Observation. What the public record shows. These reviews arrived on these days. These phrases repeat. These reviewers share these products.
  2. Inference. What the observations support together. A pattern of coordinated reviewer activity.
  3. Proof. What has actually been established. From the outside, that rarely includes who wrote a review or why.

Our case files state the first two and stop. We do not tell Amazon who is behind a review or what their motive was. Amazon holds data we do not, and it is the one in a position to establish the third layer.

Why the two reads stay separate

Blending them causes two kinds of failure.

The first is a weak text claim propped up by a hunch. A review that breaks no rule gets reported anyway, with a vague line about suspicious activity attached. That is not evidence. It is an accusation with nothing underneath it.

The second is a strong pattern buried under weak text claims. A real burst of coordinated reviews gets filed as a stack of separate complaints, each one arguing a rule it does not actually break. The pattern, which was the finding, never reaches Amazon in a form it can use.

There is a legal reason too. The FTC's rule on consumer reviews treats a public false accusation, made knowingly or with reckless disregard for the truth to get a review removed, as review suppression. Calling a reviewer part of a coordinated group without the evidence to back it up is exactly that kind of accusation. Keeping observation, inference and proof apart is how we stay well clear of it.

What the evidence never includes

Every piece of evidence in both reads is public: the review text, the review date, the reviewer's public profile and public review history, and the Verified Purchase label Amazon shows on a review.

That label deserves a note. Amazon allows reviews without it, so its absence alone is grounds for nothing. It counts only alongside other signals.

We do not use private Amazon data. We never contact reviewers. Nothing in either read comes from a seller's account, and nothing we file depends on access to one.

Reading your own reviews this way

If you are sorting your own negative reviews, the same discipline works at any scale.

  1. Read each review on its own first. Mark the ones that break a rule in their text, and the ones that are fair criticism of the product. Leave the fair ones alone.
  2. Put the rest on a timeline. Look for bursts.
  3. Read the reviews inside a burst side by side. Look for wording that repeats and is not everyday wording.
  4. Open the reviewers' public profiles. Look at what else they reviewed.
  5. Write down what you saw, not what you suspect. Report only what the record supports.

By hand, this works for a few dozen reviews. Past a few hundred, the review history step stops being practical, because each reviewer's history has to be read against every other reviewer's. That is the gap read two exists to close.

Common questions

What is coordinated reviewer activity?

Reviews that, read together, show several signals agreeing: they arrive in a burst, share unusual phrasing, and come from reviewers whose public review history keeps overlapping on unrelated products. The term describes what the reviews show, not who wrote them or why.

Can one suspicious review be reported as part of a pattern?

A pattern is a finding about a group of reviews, so a case built on one rests on the group. A single review that carries one signal, such as arriving during a burst, is recorded as not proven and left alone.

What is Aletheia?

Aletheia is our name for the review history mapping inside read two. It maps public review history to show where reviewers keep appearing together on unrelated products. Read two also looks at timing and phrasing.

Do you use private Amazon data?

No. Both reads use public review content and public review history only. Nothing comes from a seller's account.

Does read two replace read one?

No. Read one is the larger share of our work. Read two finds what read one cannot see, and the two are filed on their own evidence.

Share This Article
Joe Nilsen
Joe Nilsen
Founder

Builds the evidence systems behind both reads: review-by-review checks against Amazon's rules and review history mapping across reviews.

Follow on LinkedIn

Ready to Assess Your Written Negative Reviews?

Send us an ASIN and your written negative-review count. We will confirm whether the account meets the 300-review minimum and explain the next step.

Start a Review Assessment