I detect fake AI reviews by checking for specific evidence, not by trusting a smooth writing style. Look for repeated wording, vague personal details, missing proof, copied patterns, and claims that don’t match the product or service. One sign alone proves nothing; two or three matching signs justify a closer check.
The result changes with the review type, the platform, and the evidence you can see. A short review has less text to inspect than a detailed one, while a business with hundreds of reviews gives you a larger pattern to compare. Your goal isn’t to prove who used an AI tool. It’s to decide whether the review looks genuine, useful, and tied to a real experience.
Who needs to spot fake reviews?
A shopper checking one expensive product needs a fast risk check. You may have only 10 reviews to compare, so you’ll focus on unusual praise, missing details, and claims that don’t fit the product page. You don’t need to identify the exact writing tool. You need enough evidence to decide whether the review should influence your purchase.
A business owner has a different problem. You may need to separate a real complaint from a copied attack, or find reviews that break a platform’s rules. That job needs screenshots, dates, account details, and the exact wording. A vague feeling that a review “sounds like AI” isn’t strong evidence for a report or removal request.
A researcher, journalist, or site editor needs a wider pattern. You’ll compare reviews across sellers, platforms, languages, and dates. You may also need to check whether a person received a free product, used an incentive, or copied text from another source.
This page serves shoppers and editors who want a careful first review of public text. It assumes you can read the full review, compare several nearby reviews, and open the product or service page. It doesn’t assume access to platform records, private messages, or AI-detection software.
The page isn’t a substitute for a platform investigation. If a review affects a legal dispute, a large business decision, or a harassment report, save the evidence and contact the platform or a qualified professional. Don’t accuse a named person from writing style alone.
What changes the result?
The amount of detail matters most. A two-sentence review gives you fewer clues than a 200-word account. In a short review, check whether the words match the rating and product; in a long review, inspect the order of events, concrete details, and repeated phrases. Don’t treat missing detail as proof of AI, because real people can write brief reviews.
The review’s claims must fit the item. A review of a phone should mention features that model actually has. A restaurant review should match the menu, location, and opening dates. Check the product name, version, service date, or location against the seller’s page and other reliable records. A polished review with impossible or mismatched details is more suspicious than one with simple grammar.
The surrounding review pattern often gives stronger evidence than one sentence. Compare at least five nearby reviews when that many are visible, using the same language and rating range. Look for the same opening, the same benefit order, identical phrases, or several accounts praising the same unusual feature. A single shared word means little; a repeated structure across accounts matters more.
Timing changes what the pattern means. A group of reviews posted on one day could reflect a real event, such as a product launch or a delayed delivery batch. It could also show a coordinated campaign. Check the dates beside the ratings, then compare the first-person details. The common mistake is treating a date cluster as proof instead of checking what happened during that period.
The platform changes the evidence available. Amazon, Google, Yelp, app stores, and specialist sites display different labels, account histories, dates, and reporting tools. A verified-purchase label, where a platform uses one, can support a purchase claim, but it doesn’t prove the entire review is honest. Read what that label actually means on the platform’s help page.
AI detectors are the weakest part of the process. They can flag human writing, miss edited AI text, and give a score without showing evidence you can verify. Use a detector, if at all, as a prompt to investigate the text; never use its score as your final proof.
How to check a review
Use the steps below for one review first, then compare it with other reviews from the same product, business, or service. Keep screenshots before reporting anything, because a review, username, rating, or date can change later.
-
Copy the review into a note without changing its wording. Include the username, star rating, date, product or service name, and any platform label. Save a screenshot showing the full page, including the account name and date. The step worked when you can return to the exact version you saw.
-
Read the review once for its basic claim. Write one plain sentence describing what the reviewer says happened, such as “The charger stopped working after two weeks.” If you can’t state the claim without repeating praise words, the review may be too vague to use.
-
Mark concrete experience details. Highlight dates, locations, product versions, order problems, missing parts, staff actions, or changes after use. Don’t count generic words such as “excellent,” “amazing,” or “high quality” as evidence. The review has passed this step when its useful details can be checked against the item or service.
-
Check whether the details match the subject. Compare the model name, features, menu, location, service date, or subscription plan with the official listing or the page where you found the item. If the review describes a feature the product doesn’t have, record the mismatch rather than guessing why it happened.
-
Compare five nearby reviews, when five are available. Put the reviews in a note and mark repeated phrases, sentence order, benefits, and unusual examples. If fewer than five exist, compare every visible review and say that your sample is small. Repeated wording across several accounts is stronger evidence than one formal-sounding paragraph.
-
Check the praise-to-evidence balance. Count the specific experiences, not every sentence. A review saying “The team was wonderful and the quality was outstanding” offers two opinions but no checkable event. If the rating is high but the review gives no product, time, or service detail, reduce how much weight you give it.
-
Look for signs of copied or translated text. Search a distinctive phrase in quotation marks, first on the same platform and then across public search results. Exact matches, unusual grammar shared by different accounts, or the same benefit sequence can show reuse. A translation mistake alone doesn’t show AI, because a human may have written in another language.
-
Check the account and posting pattern. Look for a review history, repeated ratings, several reviews posted close together, or reviews covering unrelated products in the same voice. Don’t assume a new account is fake; real customers may create an account only to complain or praise one purchase. Treat the pattern as a reason to inspect, not a verdict.
-
Separate suspicion from a reportable rule break. If the review contains a false claim, undisclosed incentive, impersonation, or copied content, save the exact evidence and use the platform’s reporting category that matches it. If you only have a style concern, label it “uncertain” and avoid a public accusation.
Stop before contacting the reviewer if the dispute could involve harassment, threats, private data, or a legal claim. Report the evidence to the platform, your organization’s legal team, or a qualified attorney; don’t try to identify the person behind an account.
The key judgment is whether the review adds verifiable experience. A review can sound natural and still be false, while a stiff review can come from a real customer. Give more weight to details that fit the product, match the timeline, and differ naturally from nearby reviews.
Review cases at a glance
The right check depends on the review’s length, the platform’s labels, and how much comparison material is available. Start with the claims that could change a buying decision, rather than treating every sentence as equally important. Separate firsthand details from opinions, guesses, and statements about what the reviewer can’t have tested.
| Situation | What to do in it | What it takes | What to watch for |
|---|---|---|---|
| One short review | Check the main claim, rating, and whether the reviewer names a specific use or problem | Two minutes | Too little evidence, especially when the rating gives no reason |
| Long detailed review | Verify the described events, features, limitations, and timeline against the product or service information | Five to ten minutes | Polished filler, vague experience, or confident claims with no concrete examples |
| Several matching reviews | Compare exact wording, unusual phrases, posting patterns, and whether the reviewers mention different experiences | Ten minutes | Coordinated copy, repeated errors, or reviews that sound like one source |
| New account review | Inspect the review history, profile activity, and whether the account reviews related products or services | Three minutes | Single-purpose account, sudden activity, or no useful history |
| Negative review | Check the dates, stated facts, and whether the complaint concerns the current version or policy | Five minutes | A real issue dismissed simply because the review is harsh |
| Translated review | Read for concrete details, repeated facts, and clear descriptions of what happened | Three minutes | Translation artifacts, but don’t mistake awkward wording for dishonesty |
| Incentivized review | Check the disclosure label and give extra weight to specific drawbacks or limits | Two minutes | Biased praise, unusually broad claims, or benefits presented without conditions |
| Suspected copied text | Search a distinct phrase in quotation marks and compare the surrounding details, not just the words | Five minutes | False exact matches caused by standard descriptions or required wording |
If a review makes a serious claim but gives no date, example, or identifiable circumstance, mark it unverified rather than treating it as false. If several independent reviews describe the same concrete problem in different words, investigate that pattern even when each review is brief. Use the platform’s label as one clue, not as proof that the account or experience is genuine.
What strong reviewers notice
Check evidence before style
A strong check starts with the claim, not the grammar. Write down what the reviewer says happened, then ask whether the product, date, location, and version support that account. This protects you from a common mistake: treating formal vocabulary as proof of AI and ignoring a factual mismatch.
Natural writing can be clean, repetitive, or unusually formal. AI-assisted writing can also contain typos, slang, or personal details if a person edited it. Style is useful only when it appears alongside stronger signs, such as identical structure across several accounts or claims that don’t fit the item.
The order of details matters too. Real accounts often follow the event: purchase, use, problem, contact, and result. That order isn’t a rule, but it gives you a path to check. A review that lists broad benefits without saying what happened, when it happened, or what changed after use offers less evidence.
Compare details at the same level. “The battery lasted well” is a broad opinion. “It lasted through a six-hour flight with video playing” is a testable account, although you still can’t assume it applies to every user. The useful question isn’t whether the sentence sounds human. It’s what you can verify.
Leave the score alone
An AI detector score shouldn’t decide the case. A score may reflect short text, non-native English, repeated phrases, or a formal writing style rather than the source of the review.
Check the review’s context while it’s still visible. Save the date, rating, account history, product version, and surrounding reviews. If the review is removed after you report it, those records show what you actually saw without requiring you to reconstruct the wording from memory.
A useful comparison also includes reviews you believe are real. Take five ordinary reviews from the same item and compare their detail, length, and wording with the suspicious one. If your test flags almost every review, the method is measuring style rather than authenticity.
Leave alone any private information that isn’t needed. Don’t search for a person’s workplace, address, family, or other accounts to prove a theory. The practical limit is clear: public text can support a content report, but it usually can’t prove who wrote the review or whether they used AI.
Signs a review needs checking
A pattern is more useful than a single strange sentence. Use the table to choose a safe first check, then preserve evidence before reporting or discussing the review publicly.
Don’t treat a missing “verified purchase” label as proof that a review is false; labels vary by platform. Treat it as unconfirmed until the account, product version, and timing fit the transaction. A reviewer who names a defect, date, or use case gives you something to test. One who only repeats marketing language gives you little to verify.
| What you notice | What it usually means | What to do first | How to stop it happening again |
|---|---|---|---|
| Generic praise, no details | Low-value review | Check nearby reviews | Weight evidence, not stars |
| Exact phrase repeats | Possible copied text | Search the phrase | Report duplicated content |
| Wrong product feature | Factual mismatch | Check official listing | Verify model before buying |
| Several reviews share structure | Possible coordination | Compare dates and accounts | Inspect review patterns |
| Extreme rating, vague claim | Weak supporting evidence | Read the full review | Require concrete examples |
| Sudden review burst | Campaign or event | Check timeline and context | Compare earlier activity |
| Negative review lacks order details | Hard to verify | Check service records | Keep receipts and messages |
Save the review’s URL, screenshots, date, rating, and edits before it changes or disappears. If the claim concerns damage, delivery, or support, keep the order record and messages with it. Report the narrowest problem you can show, such as copied wording or a product mismatch, rather than alleging fraud from tone alone. If the warning remains unverified, seek independent evidence before relying on it for a purchase.
Rules that govern fake reviews
The rules depend on the platform, country, and relationship between the reviewer and the business. A review can be poor quality without breaking a law, while a review that looks natural can still violate a platform’s policy.
In the United States, the Federal Trade Commission’s Consumer Reviews and Testimonials Rule addresses fake or false reviews, certain review suppression, undisclosed insider reviews, and buying reviews based on a required sentiment.
The FTC rule doesn’t give you a magic test for AI-written text. It focuses on deceptive conduct and how reviews are created, obtained, displayed, or used. That means an AI-written review isn’t automatically illegal, and a human-written fake review isn’t made acceptable by looking natural.
Platform rules may be stricter than law. Google, Yelp, Amazon, app stores, and other services can set their own standards for conflicts, incentives, impersonation, spam, and relevance. Read the platform’s current review policy and reporting page for your country and account type.
Don’t make a legal accusation from an AI score. If the issue involves paid reviews, undisclosed incentives, impersonation, threats, or a business dispute, preserve the evidence and ask the platform or a qualified attorney what process applies. Stop self-investigation when it would require private data or access you don’t have.
The real cost of checking reviews
A quick check costs time, while a wrong conclusion can cost money, reputation, or a useful warning. For one short review, comparing the claim with the listing may take about two minutes. A detailed pattern check across five reviews can take five to ten minutes, depending on how easy the dates, accounts, and product details are to see.
The cost rises when you check several platforms. You may need to compare different names, product versions, translations, and review displays. A business should also budget time to save screenshots and records before reporting, because the platform may remove or change the material.
The highest cost comes from false certainty. Ignoring a real negative review can lead to a bad purchase or hide a service problem. Accusing a real customer of using AI can damage trust and may create a defamation or harassment risk, especially when you publish the accusation with the person’s name.
AI-detection software can add a subscription cost, but payment doesn’t solve the evidence problem. Before using one, check whether it explains its result, supports your language and text length, and lets you keep the original text private. If it gives only a percentage, treat that percentage as a weak signal rather than a finding.
Your own value is easiest to read from the decision you’re making. For a low-cost purchase, a fast comparison may be enough; for a medical, financial, legal, or business decision, spend more time checking independent facts and don’t rely on reviews alone.
Keep your review evidence clear
Save the original review before you report it. Record the page address, username, rating, date, product or service, platform label, and a screenshot that shows the surrounding page. Store the note in a place where you can find it again without editing the original wording. Keep the original screenshot file, rather than only a pasted image. Note whether the review was edited, removed, or still visible when you checked it.
Keep your own observation separate from the reviewer’s words. Use labels such as “exact phrase match,” “feature mismatch,” “same posting day,” or “no concrete event.” This makes the record easier to inspect and stops a suspicion from becoming a claimed fact. Quote only the relevant passage, preserving its spelling and punctuation. Put your explanation in a separate field, and identify which page or record supports it.
Review the evidence again if the product page changes. A model may be updated, a menu may change, or a service may move location. Check the version and date before calling a claim false. If the review still conflicts with the available information, report the precise mismatch instead of writing that the whole account is fake. If the screenshot doesn’t show enough context, mark the point as unverified and capture the surrounding text on a second check.
Stop maintaining a case when the evidence is only a writing-style impression. Keep the review as low-confidence information, or ignore it, rather than adding more personal research. Don’t treat repeated wording alone as proof of coordination; compare the dates, accounts, and concrete claims before escalating it. If the content includes threats, private data, or coordinated abuse, use the platform’s safety process and contact local authorities when there’s an immediate danger.
When AI review checks are wrong
Style-based checking is the wrong approach when the text is very short. A sentence such as “Great service and fast delivery” gives you too little language to analyze, and a generic phrase may be perfectly genuine. Compare the rating with other reviews, check the seller’s record, and use independent product information instead. If the review makes a concrete claim, verify that claim rather than treating its wording as evidence about the writer.
It’s also the wrong approach for translated or accessibility-assisted writing. A non-native writer may use formal phrases, unusual word order, or a translation tool, while speech-to-text or predictive text may introduce errors. Don’t treat those features as proof of deception. Focus on factual consistency, disclosure, copied wording, and the person’s stated experience. Ask whether the same account shows a stable pattern over time, but remember that a change in style can reflect a changed tool or circumstance.
Don’t use this method alone for a high-stakes decision. Reviews can be manipulated by people, businesses, bots, or AI tools, but authentic reviews can also be mistaken. A confident-sounding review isn’t proof that its facts are correct.
For health, legal, financial, or safety choices, check the relevant government, professional, or manufacturer information and speak with a qualified person. If the decision depends on one review, pause until you can confirm its key claim through a source with responsibility for the answer.
This page isn’t for identifying an anonymous writer. Public clues rarely prove authorship, and trying to uncover private identity can cross ethical or legal boundaries. Don’t publish guesses or contact someone’s employer based on a style score. If you need a formal finding, preserve the material and use the platform’s investigation process or get professional advice.
Frequently asked questions
Can AI-written reviews be detected reliably?
No, not reliably from text alone. AI detectors can flag human writing and miss edited AI text, especially when the review is short or the writer uses another language. Check the review’s concrete claims, product details, repeated phrases, account pattern, and disclosure instead. You can report strong evidence, but you usually can’t prove authorship or AI use from a score.
Why do fake reviews sound so polished?
Polished wording can come from AI, professional copywriting, translation, editing, or a person who writes carefully. Smooth language isn’t proof of a fake review. Look for a mismatch between confident praise and missing experience details, then compare the wording and structure with nearby reviews. A factual mismatch or repeated copy gives you stronger evidence than a formal tone.
How many reviews should I compare?
Compare five nearby reviews when five are available, using the same product, service, language, and rating range. If fewer than five exist, compare every visible review and record that the sample is small. Five reviews won’t prove a campaign, but repeated phrases, identical event order, and similar posting patterns become easier to spot than in one review.
Is it safe to use an AI review detector?
It can be safe as a low-confidence prompt, but don’t upload private, personal, or confidential review data without checking the tool’s privacy terms. Treat the result as a reason to inspect public evidence, not as proof. Don’t publish an accusation based only on a detector score, because the tool may misclassify real writing.
What happens if a review uses AI but describes a real experience?
AI use alone doesn’t show that the experience is false or that a rule was broken. The important questions are whether the claims are accurate, whether the platform permits the content, and whether the reviewer disclosed an incentive or relationship when required. Check the platform policy and report deception, copied content, or undisclosed promotion rather than AI use by itself.
Does it matter if a review is from a new account?
Yes, but only as context. A new account may belong to a genuine customer who joined to post one review, so the account age doesn’t prove deception. Check the review’s details, date, product match, exact wording, and nearby account activity. If several new accounts post similar praise in a short period, preserve the pattern and report it.
How to check fake reviews first
Start by saving the full review, including its date, rating, photos, and username as displayed, then rewrite its single checkable claim in plain language. If it makes several claims, separate them and test the one that could change a buying decision first. A statement such as “the charger failed after two days” is checkable; “I hated it” is a reaction, not proof of a defect.
Compare the claim with the product listing, instructions, order or delivery information, and any photos that show the relevant detail. Then compare its wording, specific details, date, rating, and account pattern with five nearby reviews when that sample exists. Look for copied phrases or identical unusual errors. A burst of reviews, or an account that praises or attacks unrelated products in the same style, is a warning sign, not proof of fabrication.
Before acting on the review, compare its factual claims with independent evidence and label unconfirmed claims low confidence. Don’t treat a missing photo or new account as proof that a review is fake, and don’t assume polished detail proves authenticity. If threats, private information, paid manipulation, or unverifiable disputes arise, seek platform or legal help next.
Robert Miller works as a Reputation Management Consultant. He has been helping U.S. businesses for over ten years. His work involves helping people protect and enhance their online image.Robert has worked with a wide range of companies. He has helped new businesses, local places that give a service, and big companies. He helps turn customer reviews into something that helps people trust their company and drives business growth.
